2 Methods

The ridehail simulations have three components, each of which is described below:

• A geographical area. This is called a City, as most discussions about ridehail systems have focused on cities as the unit of comparison, but it need not be an actual “city”.

• Vehicles. These drive around the city, respond to trip requests, and carry passengers from the trip origin to the trip destination.

• Trips. These start as trip requests, and have an origin and destination within the city.

2.1 Cities

A city is a square grid of identical streets, creating uniformly-sized “blocks”.

The square is, “wrapped”, so that a vehicle going off the top of the city appears at the bottom, and a vehicle going off the right of the city appears on the left (topologically, it is a torus). This is not as unrealistic as it may first appear: in the real world, drivers are not bound by city boundaries: some trips begin inside the city and finish outside, while others start outside and terminate inside the boundaries. The “wrapped” layout can be thought of as an area with a constant population of vehicles, sustained through a flux of vehicles leaving and entering through the edges of the area under consideration.

In this simplest model of a city, all locations are identical.

The unit of travel is a single block. A block may be thought of as a distance, a time, or a combination of the two. It is often taken as a minute of travel time, but this is not a requirement.

2.4 Simulations

Each simulation is a sequence of moves. In each move, vehicles travel one block, from one intersection to the next. A typical move involves the following events:

• Any new trip requests are generated.

• Available vehicles are assigned to trips. The trip is assigned to the nearest available vehicle in , and if there are multiple vehicles at the same distance from the trip origin, one is selected at random. Vehicles always accept trip assignments.

• Vehicles move from one intersection to the next. Each vehicle chooses a direction and moving a block to the next intersection. For vehicles in , the direction is random; for vehicles in and , the direction is towards the trip origin or destination, as applicable.

• Any trip that reaches its destination terminates.

Figure 2 illustrates a simulation in progress. At the end of each move, the simulation records and optionally plots the state of each vehicle and statistics about the overall system (fraction of vehicles in each phase, for example).

This framework is sufficient to simulate a city with a fixed supply of vehicles, and fixed demand (rate of requests). For some of the results below, this is all that is needed. For some other topics, the simulation has to be extended.

2.6 Regions

The spirit of the simulations is to be as simple as possible, so that they can be calibrated using the very limited high-level public data available and so that they can be used comparatively. However, two extensions are necessary for some applications.

Many cities have a central “downtown” zone where the demand for ridehail traffic is high, surrounded by suburban areas where demand is lower. For example, 60% of trips in Metro Toronto originate within “Old Toronto” and East York, even though these two make up less than 20% of the total Metro area. This inhomogeneity is built into the model as two zones. A central zone has sides (so, one quarter of the city’s area) and has a higher rate of trip requests than the surrounding area, as shown in Figure 3 below. The trip destinations remain randomly distributed around the whole city. With an inhomogeneity of zero, the central zone is the same as the rest, with an inhomogeneity of one all trip requests take place in the central zone. In this way, and in the spirit of keeping the model as simple and parsimonious as possible, a single inhomogeneity parameter between zero and one captures the two zones.

The second extension, discussed in the section on Engaged Time standards below, is to allow vehicles to enter and leave the system in response to price and demand.

The use of random locations for trip destinations implies an average trip length of (regardless of inhomogeneity). To model an area where the average trip length may differ from this, a maximum trip length can be set, with trips distributed randomly over intersections within that distance.

3 Ridehail in Toronto

The ridehail market in Metropolitan Toronto was chosen as a test arena. The city has released important data in a 2019 study by the City of Toronto (Policy & Innovation Transportation Services 2019) and an update in 2021 (Policy & Innovation Transportation Services 2021), here called Toronto 2019 and Toronto 2021, which provide a valuable set of high-level data that can help validate the simulation approach.

Metro Toronto has a population of 3 million people. It consists of the City of Toronto and surrounding urban areas (Scarborough, Etobicoke, York, East York, and North York and is intermediate in size between the old City of Toronto and the Greater Toronto Area (population 6 million). Metro Toronto (henceforth just “Toronto”) has an area of 630 square kilometres. For the purposes of these simulations, Metro Toronto is considered as a square, with sides of 25 km and hence an area of 625 km^2.

3.1 Reference values

In Toronto 2021, the authors chose Thursday Feb 6, 2020 as a representative day to assess the overall state of ridehail in Toronto, pre-pandemic (p25). The first question we ask is: can the model reproduce important aspects of this day?

Although the date is now three years in the past, any reference to “Toronto” here implies “Toronto on February 6, 2020”. Relevant statistics from this study are shown in Table 1.

Quantity	Value
Trips completed (D)	193,902 (135/min)
Average trip length (L)	8.13km
Average vehicle speed (v)	31 km/h
Percent of time: Period 1 (available for trip)	40.5%
Percent of time: Period 2 (en-route to pick-up)	11.2%
Percent of time: Period 3 (with passenger)	48.3%
Active vehicles per hour	6200

There is a regular churn of active vehicles, so the reported number of active vehicles, which is the total number recorded in a given hour, overestimates the number on the streets at any one time. Using the distribution of vehicle hours from the Toronto 2021 report suggests that 40% of all vehicles will leave during any given hour. To maintain a steady volume of traffic a corresponding number will enter. The number of active vehicles at any one time is therefore more like 6200 * 100 / 140 = 4400.

We demand that the numbers fit Equation 1. After these adjustments we take the values in Table 2 as reference values:

Quantity	Symbol	Value
Average number of active vehicles		4500
Demand (trip requests per minute)		135 per minute
Average trip duration		16 minutes
Percent of time: Period 1 (available for trip)		40.5%
Percent of Percent of time: Period 2 (en-route to pick-up)		11.2%
Percent of time: Period 3 (with passenger)		48.3%
Average wait time		3.7 minutes
Average wait fraction		0.23

3.2 Calibration I: supply, demand, and utilization rate

At 31 km/hour, it takes 48 minutes to cross the city. In the simulation Toronto is represented as a 48 by 48 grid, so that each block represents a minute of travel. Trips start and finish at random locations but with a maximum duration of 32 minutes, which yields an average duration of 16 minutes to match the observed duration. The demand for trips was fixed at the observed average value of 135 trips per minute.

Figure 4 shows the results of a sequence of simulations, in which the number of vehicles is varied from 2000 to 7,000. For each value of , a simulation of 500 blocks duration was carried out, with the results averaged over the final 100 blocks. The time for each simulation depends on the number of vehicles: for 5000 vehicles the simulation takes about 15 minutes on a Dell XPS 13 9380 laptop. The first result is that this simplified model can simulate an environment the scale of Toronto on a single laptop computer.

At low vehicle counts, there are too few vehicles to sustain a steady state. Vehicles are busy all the time (), trip requests go unanswered for longer and longer (the wait fraction is literally off the chart. Things fall apart: the drivers cannot keep up with the requests.

At vehicle counts above 3000, the drivers can keep up with requests and the system achieves a steady state. As is increased further, the utilization rate falls and the excess capacity rises, as more vehicles compete for the available trips. The wait time decreases as the number of vehicles increases, but is a small fraction of the trip duration even at 3000 vehicles, and the dispatch time is correspondingly low.

This model correctly reproduces the value of 0.48 with a supply of vehicles. In some ways this is just a check that eliminates some bugs: the identity is satisfied by any steady-state behavior, regardless of geometry. The chart confirms that the utilization rate depends solely on vehicle supply , trip demand , and trip duration , but not on finer features of the city geography.

3.3 Calibration II: wait times and dispatch times

The homogeneous model of Toronto, with trip origins and destinations distributed randomly over the whole city, delivers the correct ratio of to . However, the homogeneous model underestimates dispatch time and wait time compared to the observed values. Identity II fixes the ratio, but not the absolute value of either of those quantities. To get this value correct we must allow inhomogeneity of trip distributions.

Figure 5 shows the vehicle phases as a function of inhomogeneity, using the fixed demand of 135 trips per minute and a fixed supply of 4500 vehicles derived from Figure 4. The Figure shows that the utilization rate is independent of inhomogeneity, as expected, while the other values change. An inhomogeneity of 0.4 reproduces the observed values of and , and hence , completing the essential descriptive statistics of Toronto’s ridehail market.

An inhomogeneity of 0.4 means that 60% of trip origins are distributed randomly over the entire city, while 40% are constrained to within the central square, of size . Hence, 55% of all trips start in the central “downtown area”. This corresponds roughly to the old City of Toronto (reference needed for travel times).

3.4 Application I: wait times and excess capacity

Figure 6 shows simulations for the calibrated Metro Toronto model, with varying numbers of vehicles from 2000 to 7000. The actual number on Feb 6, 2020 was , as discussed above.

The model makes explicit and quantitative the trade-off between short wait times and additional vehicles on the streets. Figure 6 shows that a decrease in wait times of one minute (from 4.2 to 3.2 minutes, or = 0.26 to 0.20, bracketing the observed average of 3.7 minutes), at a fixed demand of 135 trips per minute, adds approximately 1500 vehicles to the streets (from 3800 to 5300) with attendant congestion and pollution.

3.5 Application II: demand and utilization rates

In 2017, the New York Times published a story that highlighted the tradeoff between wait times and utilization rates, including an animation that inspired the model presented here (Scheiber 2017). Uber responded with a blog post claiming that the story missed the effect of demand: that with additional demand, both wait times and utilization rates could increase in a virtuous cycle. While Uber provided charts to back up the claim, they noticeably lacked any numbers on the Y axes. Since that time, there has been little discussion of the claim.

City	Year	Time or distance?	Utilization rate
Chicago	2015	Distance	59%
Chicago	2019	Distance	55%
Boston	2016	Time	47%
Boston	2019	Distance	55%
Los Angeles	2016	Time	42%
Los Angeles	2019	Distance	60%
New York City	2016	Time	51%
New York City	2017	Distance	59%
San Francisco	2016	Time	55%
San Francisco	2017	Distance	67%
San Francisco	2019	Distance	64%
Seattle	2016	Time	44%
Seattle	2019	Distance	53%
Seattle	2020a	Time	60%
Seattle	2020b	Time	51%
Washington DC	2019	Distance	55%

Figure 7 shows the utilization rates and vehicle counts required to sustain a 3.7 minute wait time in Toronto, and sheds some light on Uber’s counter-claim. As Uber claimed, an increase in demand does increase vehicle utilization even at a constant wait time. However, that utilization increase fades away at higher demand rates. As a result, even if the trip volume in Toronto were to increase to 200,000 per day (a 50% increase over the pre-pandemic numbers), the utilization rate would still increase to only 54% if wait time were kept at the same value.

As Table 3 shows, other cities have utilization rates (by time) of 50 to 60% , with New York City and San Francisco as outliers. Toronto has a level of demand that is lower than Chicago, which may provide a reasonable upper bound to what can be expected of around 60%.

There are two ways that utilization rates may exceed these estimates. The first is for the number of drivers to be limited, so that wait times are longer. The second is for trips to concentrate even more heavily into the high traffic central zone. But this central core is well served by mass transit, and any gain in ridehail trips would be achieved by taking riders away from mass transit and putting them in cars (Young et al. 2020), adding to problems of congestion (Erhardt et al. 2019).

Without these changes, the ridehail system has at least 40% excess capacity (vehicles not engaged in driving passengers or en route to pick passengers). There is no efficient solution to emissions through ride hailing.

4 Equilibration

A distinguishing feature of the gig economy model is that, once drivers are signed up to the platform, they control their hours of work: they are generally free to start or stop work simply by logging in or out from the application. The platforms’ ability to expand or shrink the supply of drivers in response to demand, whether in the short term (“surge pricing” and its descendants) or in the longer term through changes in commission and incentives, has been central to their success. The ridehail simulation has been extended to accommodate entry and exit.

Using the model to make quantitative predictions of ridehail system behaviour would require more data than is publicly available. At this point, the equilibration serves mainly as a pedagogical tool to illustrate the interdependencies between the different facets of ridehail markets.

5 Discussion

6 Conclusions

The simple simulation of a ridehail system may provide

1 Introduction

In as few words as possible, I hope to answer some practical questions I had when I started reading up on the topic. Like these:

• How are quantum computers built?

• Are we going to have quantum laptops any time soon?

• How can I get my hands dirty with quantum computing today? What do I need to understand to write my first “Hello world” quantum computing program?

• What advantages, if any, does quantum computing offer? This topic is key to the whole thing, but I’m going to skip over it very lightly here. Maybe a Part II will discuss it in more depth.

• What is the status and what are the prospects of quantum computing today? Well, an answer is too much to hope for, but I’ll offer some reflections.

But let’s start with three building blocks of quantum computers: qubits, gates, and circuits.

2 Building blocks

2.1 Qubits

Regular computers are built around bits, quantum computers are built around qubits. So what does that mean?

The archetypal example of a qubit is an atomic nucleus with a spin of ½, like the ⁸⁹Y isotope of Yttrium (although many quantum computers don’t actually use spin-½ nuclei, but let’s put that aside for now). The spin means that the nucleus has angular momentum, as if it were an electrical charge spinning in one direction or the other. This spinning electrical charge creates a little magnet that can be pointing either “up” or “down”.

(Aside: classically, the axis around which the rotation happens is the “direction” of angular momentum, which is a bit odd when you first come across it. Angular momentum in quantum mechanics follows the same rule. Similarly, a classical electric current moving in a circle creates a magnetic field. So quantum spin, a form of angular momentum, is following similar rules as classical angular momentum. A big difference, though, is that classical angular momentum can take any value–a bicycle wheel can spin at any speed–but quantum spin can take only specific discrete values (it is “quantized”): in the case of a spin-½ particle that is +½ or -½, which we usually call “spin up” or “spin down”. These two values can represent and when used as a qubit in a quantum computer, just as a classical “bit” might be a transistor that can take on a state of “on” or “off”. It’s often said that an electron or nucleus with angular momentum is like a little ball spinning around an axis, except that it’s not a ball and it’s not spinning.)

Qubits, as has often been said, exist in a state that is a superposition of 0 and 1, but let’s leave that to one side for now as well.

Qubits may be atomic nuclear spins, using nuclear magnetic resonance, but in principle any other system that displays this two-level behaviour can be used as a qubit. Some research groups have built quantum computers with atomic-scale qubits, like arrays of charged ions trapped in space by electromagnetic fields (e.g., Quantinuum or IonQ) or arrays of neutral atoms (e.g. QuEra). Others (e.g. Google and IBM) are exploiting macroscopic quantum phenomena – for example, qubits based on superconductivity in circuits of maybe a micrometre in size: the current in a superconductor is macroscopic (although still small by our standards) but is quantized. And there are other approaches too : photonics (e.g. Photonic and Xanadu), topological “anyons” (Microsoft) and more. These different approaches are often called “modalities” in industry jargon.

A feature of quantum computing is its division of intellectual and physical labour. Some people build quantum computers using a particular modality, and learn how to work with the components that they need. Others spend their time developing quantum algorithms that could, in principle, run on a quantum computer built with any of these modalities.

The large number of qubit “modalities” being pursued is often seen as a strength of the field, a sign of its vitality, and this makes sense from a “don’t put all your eggs in one basket” perspective. The number of modalities increases the chance that one of them will surely a scalable quantum computer. But I wonder… is it also a sign of the uncertainty that remains? After all, if there were a clear path from a particular modality to a general purpose quantum computer, then the industry would have coalesced around that modality. The number of modalities being actively pursued suggests that each has obstacles that it has not yet surmounted. It would be interesting to chart the progress in the industry, not by what each company and research group says about its own modality and its own breakthroughs, but by what they each say about the limitations of everyone else who is competing for the same funds. But OK, back to quantum computing proper.

2.2 Gates

Whatever approach you choose, you want to build a computer out of these qubits. So the next thing you need for a computer is logic gates. In a classical computer the simplest gate is this NOT gate.

The NOT gate has one input and one output. If the input is 0, then the output is 1, and if the input is 1 then the output is 0.

In classical computers, the lines are wires and the shield-symbol is etched into a chip and the bits are charges moving through the gates. In the world of quantum computers, the qubits stay where they are: it’s called “in-place computing”. If you shine a laser at one particular qubit and cause it to flip from one of its states to the other, then that’s like a NOT gate: it turns a 0 into 1 or a 1 into 0. In quantum computing circuit diagrams, the lines are not wires, they are before and after states of the qubit.

(In the figure above, the state of each qubit is represented in what’s called the “Dirac ket notation”, like or, equivalently, . This notation is really useful and standard in all areas of quantum mechanics, but here it’s just a symbol that shows the value of a qubit.)

Most classical gates, like XOR and NAND, have two inputs, and many quantum gates can also have two inputs. Here is a “Controlled NOT gate” or CNOT gate, which is an important component of quantum computers. If the top qubit is spin down (0), the bottom qubit is unchanged. If the top qubit is spin up (1), the bottom qubit is flipped.

Again, if this were a classical computing gate, the “bits” would be charges travelling along wires, going in to the gate from the left and out on the right. But in quantum computing the gate is not a feature of a circuit etched into a chip. It’s an operation—a combination of laser or microwave pulses that changes the state of the qubits from a before state (on the left) to an after state (on the right)—and the qubits stay in place.¹

Before moving on, I should mention one gate you will see all the time in quantum computing, but not in classical computing. It’s called the Hadamard gate. If you prepare an input qubit in state then a Hadamard gate puts it into a superposition of states. Many circuits start with Hadamard gates way to take advantage of this superposition. I just mention this because it is useful and ubiquitous, and also because it is needed in the next section.

2.3 Circuits

Once you have qubits and gates, the next thing to do is implement algorithms. In classical computers you have a fixed circuit and programming instructions send bits through these fixed gates to implement an algorithm. In a quantum computer, you start with a set of qubits in a prepared state (usually all ), and apply a sequence of gate operations and then measure the output. So a circuit is how you implement an algorithm.

Here is a circuit diagram for a quantum fourier transform (QFT), which turns out to be important in the Shor algorithm for factoring integers. You have qubits, you prepare them in some initial state , you subject them to a series of gates, and you end up with some outputs which you measure.

To be useful, a quantum computer would need many qubits and many gates, and so far there are strict limits on how many qubits can be assembled, and how many gates can operate on them before they lose their well-defined state. How many and why? We’ll come to that later on.

3 Will I have a quantum laptop? (No)

When I started looking at quantum circuit diagrams I thought it contradicted the idea of quantum computers as general purpose computers, because you need different circuits for different algorithms and I assumed that any computer would have a fixed set of circuits (like CPU circuits are etched into silicon) and hence be limited in what algorithms it can execute.

But once I realised that a circuit is not a permanent thing but is instead a sequence of operations (gates) that implement an algorithm, then I realised it’s not a contradiction at all.

Perhaps I’m belabouring this point, but for some reason, many introductions to quantum computing seem to skip over this difference between regular electrical circuits, and “quantum circuits”. I read several before I realized what was going on.

So are quantum computers general purpose? Well yes and no. Formally, they are. That is, there is a set of gates (different to the gates for classical computing) which satisfy the requirements for Turing completeness. So that’s nice.

But what about “general purpose” in the more everyday, practical sense? Let’s go back to how you build quantum computers. No matter what approach you take, the set of qubits must exist in what is called a single “quantum state”, and quantum states fall apart or “decohere” if they interact with the wider world, so they have to be isolated. Also, thermal energy (which is just one form of interaction with the wider world, of course) will jiggle these qubits among the states and you can’t have that, so quantum computers have to be operated at very low temperatures, meaning about 1K.²

This fragility of quantum states has consequences:

• You won’t, in the foreseeable future, have a quantum computing laptop. Quantum computers need cooling and will be built in fixed locations.

• It’s going to be expensive to build them. And so your company probably won’t be buying many quantum computers any time soon. It makes sense to have specialised providers who will provide time on them through a cloud service.

• For any algorithm that can be executed either classically or in a quantum computer, classical is going to be roughly a gazillion times cheaper for the foreseeable future. You won’t be running most of your code on a quantum computer.

BUT there are certain tasks that may be done more quickly on a quantum computer than on a classical one, if a powerful enough quantum computer can be built. So what will most likely happen is that specific function calls within a larger program may be executed on a quantum computer, and the rest will be done on a classical one.

The emerging paradigm, at least as at October 2024, is of a cloud-hosted data centre where an application can use classical computers for many parts of an application, but have access to quantum computers to run specific functions. Perhaps it’s not surprising that Microsoft, IBM, and Google are at the forefront and Amazon (AWS) is now investing heavily: they want to host them, and rent them out to their cloud users.

4 How do I program a quantum computer?

The companies providing quantum computers want people to use them, so they provide libraries or modules that you can use. For example, Google and IBM provide python libraries (cirq and Qiskit, respectively) and Microsoft provides its own Quantum Development Kit. These libraries provide an interface to the host quantum computers and, more practically for most of us, provide a way to simulate quantum computing on a local classical computer, for small cases.

So here is some code that uses Google’s cirq library for python. It defines two qubits and a list of gates that operate on those qubits. Then it creates a circuit from the gates, and runs the circuit on a simulator on your local computer. As the algorithm is very simple, the quantum computation can easily be simulated on a classical computer. If you’re familiar with the python programming language, you can easily install cirq by typing pip install cirq at a command prompt. Here is a “hello world” example to give you an idea of what the code looks like, which each step described by a comment.

Providing quantum computing at the level of a library makes it possible to implement hybrid applications (which, as the last section showed, are the most likely form) with the quantum components implemented as functions of a largely-classical application.

import cirq

# Define two qubits
q_control = cirq.NamedQubit('control')
q_target = cirq.NamedQubit('target')

# Define a list of gates: a "Hadamard" gate on each qubit, 
# followed by a CNOT gate, and then a measurement of the 
# control qubit q_control.
gates = [
        cirq.H(q_control), 
        cirq.H(q_target), 
        cirq.CNOT(q_control, q_target), 
        cirq.measure(q_control)
        ]

# Create a circuit from this list of gates
circuit = cirq.Circuit(gates)
print(circuit)

# Simulate the execution of the circuit
simulator = cirq.Simulator()
result = simulator.run(circuit)
print(result)

The “print” instructions display an ascii-art rendering of the circuit and the results. I found this exercise helpful, to see how an end user might construct a circuit in software and run it (albeit on a simulator). It does prompt the question: if all the quantum algorithms are going to be expressed in this unfamiliar circuit form, how expressive is this language? There is a big gap between “it’s a formally complete set of gates so you can do anything this way” and any more practical statement about real implementations.

5 Why might quantum computing matter?

Building a new kind of computer, defining new ways to express algorithms, providing new coding libraries, integrating computers with existing cloud services… this is all a lot of work. Why would you bother? Only because quantum computing has been shown to have an intrinsic advantage over classical computing, in principle, for certain kinds of problem. If I write up Part II of this introduction (link to come) then it will describe this advantage in detail, but here we shall skip lightly over the essential details. They are available in many other places anyway. The goal here is to get down the basic minimum that I needed to understand some of the descriptions you see elsewhere.

5.2 The Deutsch algorithm

The Deutsch algorithm was the first (1992) to suggest we could use these quantum phenomena to solve a problem more efficiently that one could do with a classical computer. The idea was to pose a question, no matter how artificial and useless, which quantum computers might be able to answer quicker than classical computers. And then maybe that will give us insights we can use to get to some useful questions later on.

Here is the artificial and useless problem Deutsch chose (that is not a criticism). There are four possible functions that can act on a bit data type and produce a bit output.

The four functions that take a bit as input and produce a bit as output

Input
0	0	1	0	1
1	0	1	1	0
	Constant	Constant	Balanced	Balanced

One question would be: if you have a black box that implements a function and just look at the inputs and outputs, can you tell which function is in the black box? But Deutsch picked out an even more obscure question: and are constant and and are “balanced”, which is that they have one of each. If you have a black box and just look at the inputs and outputs, can you tell if , the function inside, is one of the even functions or one of the balanced functions?

Classically you need to pose two queries: what’s and what’s ? And in a quantum circuit the naive way would also take two queries: as the measurement will only give either or with probability or , it seems you still need to make at least two measurements to find out whether is constant or balanced (an illustration that superposition, by itself, does not enable parallel computation on all its states). But Deutsch came up with an idea that you could use an extra qubit and combine the effects of superposition and interference to get the answer with just one query.

And here is the algorithm to solve it, represented by a circuit.

I won’t go through this now (many other places do, and maybe I’ll have a go in Part II, but no promises), but with this circuit just one measurement of the top qubit will tell you whether the black box, labelled here as an “Oracle”, represents a constant or a balanced function. So you can use a combination of superposition and interference to do things you can’t do on a classical computer.

6 Quantum computing in 2024

So where are we when it comes to building real quantum computers that can handle problems of interest? There are several metrics of interest: here are three (I’m sure there are others).

• Number of qubits. But implicit in the discussion above is that a qubit behaves well: that it is in one well-defined quantum state and stays there. In practice, a distinction is needed between “physical qubits” which may be less-than-perfect, and “logical” or “error-corrected” qubits, which behave properly.

• Number of gates. To be able to operate on an array of qubits with a succession of gates is demanding. Not all systems can stay “coherent” for the length of time it takes to apply a number of gates. So it is not at all clear that once you have a large number of qubits, even if they are properly error-corrected, you have a successful quantum computer.

• Connectivity. Not all quantum computers allow you to implement the same set of gates, and even if you can, for some a “gate” may be a complex, multi-step operation in practice. And in some architectures / modalities, only qubits “near” each other can be used in a particular gate. And some algorithms assume three-qubit gates, which introduces even more complexity. So you may see reference to “transpiling” algorithms, which means taking a quantum circuit and re-expressing it in terms of the gates that you have at hand on a particular quantum computer.

There is a tension relevant for building quantum computers. You need the whole circuit to be a single state, which means no interference from the environment. Yet at the same time you are poking it and prodding it with these gates, which are very explicit interactions with the environment.

Also: no quantum state is really just two states that you can pick between. There are higher energy states that start to get populated as the temperature increases, and which mess things up.

So error correction is a big part of any quantum computing effort. And typically the answer is the same as any other error correction system: redundancy. So there is a distinction between physical qubits (which are the building blocks we have talked about) and logical qubits, which are collections of physical qubits that you can trust to behave in the way the theory needs. So how many logical qubits to a single physical qubit? Depends on the approach, but the leading work says at least 50.

6.2 The big question

So far quantum computing has delivered a set of proofs of concept, together with road maps for scaling up that give an optimistic sense of predictability, when there are still fundamental unknowns to resolve.

We know there are problems for which there is quantum advantage. We know we can build some of these, like the Google sampling problem, but like the Deutsch algorithm these are mainly artificial. And we know there are problems we care about that have the potential for quantum advantage.

What we don’t know is whether we are on the left or the right side of this diagram. Or, to put it differently, how long we will be on the left side. It’s not a resolved question, and it’s not likely to be resolved soon. But don’t take this from me! Here is a recent quotation from Scott Aaronson, who did a postdoc near me at the University of Waterloo before going on to become one of the leaders in the field of quantum algorithms.

“Billions of dollars [are] being invested in quantum computing… in the hope that a quantum computer would accelerate machine learning, optimization, financial problems, AI problems…. As here, as a quantum algorithms person, honesty compels me to report to you that the situation is much, much iffier.”

Links

• Government of Quebec page on Advance Request for Medical Aid in Dying
• Montreal Gazette Explainer (October 30, 2024)
• Globe and Mail article
• Dying with Dignity
• Advance requests for MAiD in dementia: Policy recommendations emerging from a mixed-methods study of the views of the Canadian public and MAiD practitioners.

1 Executive summary

The “Uber Files” revelations of summer 2022 showed that the company “broke laws, duped police and secretly lobbied governments” of all levels to achieve its aggressive expansion goals.

Unfortunately, Uber continues to take governments for a ride. This report addresses two topics:

• Emissions. Uber makes big but vague promises about its impact on the environment, with its tagline “Your city. Our promise. Uber will be a zero-emission platform by 2040” (Uber, Inc. 2022b)). The City of Toronto takes this claim seriously by pursuing collaborative “emissions reductions targets” (City of Toronto 2021).

  Our computer simulations show that Uber vehicles in Toronto, which had a passenger in the car less than half the time in 2020, are unlikely to ever have passengers in the car for more than 60% of the time. At least 40% of the time, Uber vehicles in Toronto will be driving empty.

  Further, Uber vehicles are most highly occupied in busy, congested areas such as central Toronto, which are well suited to and well served by public transit. The expansion of Uber services in these areas can only add to congestion and draw passengers away from more sustainable and efficient public transit.

  The pursuit of short wait times, essential to Uber’s popularity with riders, is incompatible with emissions goals. A difference of one minute in wait times adds another 1600 vehicles to Toronto roads, with their attendant congestion and emissions.

• Driver Pay. Uber has persuaded governments to accept calculations based on low driver cost estimates and a treatment of “idle” drivers as not-really-working. This report argues that, at least for full-time drivers, a realistic estimate for the average pay rate is under $8 per hour.

  But the problem of driver pay is broader than this low number. The Ontario government is adopting an approach of mandating wages based on “engaged time” – the time a driver is either driving a passenger or heading to pick one up. Computer simulations show that, as fares increase, the gig-work model means that more drivers enter the system and eat away those gains by reducing the time each driver spends “engaged” with passengers. No matter what number is dictated, “engaged time” pay rates do nothing to improve driver income.

  While these results address the specific case of Uber drivers in Toronto, many of the factors at work are common to other forms of gig work and to other jurisdictions. There is a danger that an exploitative, unsustainable model of work may be cemented in place through poorly-designed regulations based on misleading lobbying from well-funded platform companies, and it must be addressed now.

2 Introduction

The unique nature of “gig work” as practiced by Uber and other platform companies is now well known. It has two main distinguishing features:

• Free entry and exit. Platform companies design their systems to allow easy entry and exit of workers. Uber, for example, attempts to make sign-up as a driver as easy as possible. And once a driver is signed up, starting work is simply a matter of opening an app on a phone.

• Piece rate payment. Platform companies have made every attempt to avoid classifying their workers as employees with an hourly pay rate. Instead, the gig-work model is to pay workers only while engaged in a task.

  Like many other jurisdictions, the Province of Ontario and the City of Toronto are introducing policies to accommodate “gig work” into the broader framework of labour standards, and to align the operations of Private Transportation Companies (PTCs) with public priorities concerning transit, including efficiency, congestion, emissions, accessibility and more. Through computer simulations that capture poorly-understood dynamics of gig work, this report addresses these two current policy directions:

• Congestion and emissions. The City of Toronto takes seriously Uber’s claim that it can reduce emissions through fleet electrification (City of Toronto 2021), and is “seeking to update by-laws to achieve vehicle electrification and emissions reductions targets for the vehicle-for-hire industry”. Can Uber and Lyft drive sustainable transit in Toronto?

• The growth of precarious work. In its “Working for Workers Act, 2022”, also called the “Digital Platform Workers’ Rights Act”, the Government of Ontario introduced a “minimum wage” standard. It states that “minimum wage shall be paid for each work assignment performed by a worker”, which has been interpreted as covering only “engaged time”. What are the actual levels of pay for Uber drivers, and how will “engaged time” regulations affect driver income?

The role of Uber drivers waiting, unpaid, for the platform to assign them a trip, and the role of full-time Uber drivers both emerge as important topics through the report.

3 Computer simulation

Discussion and debate concerning Uber and other PTCs is vigorous but often hampered by lack of reliable data. The problem is made worse by the compromised nature of research based on negotiated or commissioned access to PTC data. The obvious conflicts of interest for researchers who benefit from access to corporate data (and sometimes direct payment) while claiming to be objective, were described some years ago by the International Labour Organization and others (Berg and Johnston 2019) but the recent “Uber Files” revelations provided essential confirmation and specifics (Lawrence 2022); (Mojtehedzadeh 2022); (The International Consortium of Investigative Journalists 2022)).

Some cities have negotiated or required access to ridehail data from the PTCs, but that data is not always accessible to the broader public. In those few cases where cities do make data public, most prominently Chicago (City of Chicago 2022) and New York City (New York City Open Data 2022), key elements are missing, such as the time that drivers spend on the platform but without a trip. The City of Toronto is now collecting detailed data from ridehail platforms, including time spent without a trip, but this data is not currently available for outside analysis beyond the City and its partners.

This report attempts to address the lack of reliable data. It presents a simple computer simulation of a ridehail system in a city, which can be calibrated against broad aggregate data, which are more widely available than detailed statistics. The simulation is intended to be as simple as possible while still capturing the essential elements of a ridehail system.

One goal of the simulation is to check intuition-based claims. Some policy discussions have led to talk at cross purposes with different unstated assumptions underlying rival claims, such as the relationship between the number of idle drivers and wait times or the effect of price changes on driver incomes. Both of these are discussed below.

A simple model that is independent of a particular city geography and road layout is easier for comparisons of the ridehail experience among cities. Also, a computer simulation allows “what if?” questions that may provide insights into the trade-offs faced by platform companies, municipal governments, drivers and passengers.

The ridehail simulation has three components, each of which is described below:

• A geographical area. This is called a City, as most discussions about ridehail systems have focused on cities as the unit of comparison, but it need not be an actual “city”.

• Vehicles. These drive around the city, respond to trip requests, and carry passengers from the trip origin to the trip destination.

• Trips. These start as trip requests, and have an origin and destination within the city.

3.4 The simulation

Each simulation is a sequence of moves. In each move, vehicles travel one block, from one intersection to the next. A typical move involves the following events:

• Any new trip requests are generated.

• Available vehicles are assigned to trips. The trip is assigned to the nearest available vehicle in P1, and if there are multiple vehicles at the same distance from the trip origin, one is selected at random. Vehicles always accept trip assignments.

• Vehicles move from one intersection to the next. For each vehicle, this involves choosing a direction and moving a block to the next intersection. For vehicles in P1, the direction is random; for vehicles in P2 and P3, the direction is towards the trip origin or destination, as applicable.

• Any trip that reaches its destination terminates.

At the end of each move, the simulation records and optionally plots the state of each vehicle and statistics about the overall system (fraction of vehicles in each phase, for example). Figure 2 illustrates a simulation in progress.

This framework is sufficient to simulate a city with a fixed supply of vehicles, and fixed demand (rate of requests). For some of the results below, this is all that is needed. For some other topics, the simulation has to be extended.

3.5 Extensions

The spirit of the simulation is to be as simple as possible, so that it can be calibrated using the very limited high-level public data available and so that it can be used comparatively. However, two extensions are necessary for some applications.

Many cities have a central “downtown” zone where the demand for ridehail traffic is high, surrounded by suburban areas where demand is lower. For example, 60% of trips in Metro Toronto originate within “Old Toronto” and East York, even though these two make up less than 20% of the total Metro area. This inhomogeneity is built into the model as two zones. A central zone has sides C/2 (so, one quarter of the city’s area) and has a higher rate of trip requests than the surrounding area, as shown in Figure 3 below. The trip destinations remain randomly distributed around the whole city. With an inhomogeneity of zero, the central zone is the same as the rest, with an inhomogeneity of one all trip requests take place in the central zone. In this way, and in the spirit of keeping the model as simple and parsimonious as possible, a single inhomogeneity parameter between zero and one captures the two zones.

The second extension, discussed in the section on Engaged Time standards below, is to allow vehicles to enter and leave the system in response to price and demand.

The use of random locations for trip destinations implies an average trip length of C/2 (regardless of inhomogeneity). To model an area where the average trip length may differ from this, a maximum trip length can be set, with trips distributed randomly over intersections within that distance.

6 “Engaged time” standards do not work for gig workers

The approach of the Digital Platform Workers’ Rights Act (Legislative Assembly of Ontario 2022) to a “right to minimum wage” for gig-workers fails, not just because the numbers are too low, but because the principle of addressing only “engaged time” is mistaken.

The question of “engaged time” wage extends far beyond Ontario. Seattle recently introduced a minimum wage for Uber drivers based on engaged time (Herbold and Lewis 2022). In its “Frequently Asked Questions – Worker Protections”, Uber UK explains its approach to the National Living Wage (Uber, Inc. 2022a):

You earn the National Living Wage from the time you accept a trip on the app to the point at which the trip is completed: this is called the engaged time. 

Time waiting for a trip to be offered to you is not included in your National Living Wage calculation. This is because there is no requirement to accept any trips during this period. 

This allows you the flexibility to manage your time as you wish.

6.2 Pay is not affected by fare changes

As Uber and other gig-economy platforms do not limit driver supply, a rise in “engaged time” pay increases the number of drivers on the platform, but does not translate into increased earnings for individual drivers.

Figure 9 shows the results of the simulation. The bar on the left of each chart is a reference simulation that describes an environment similar to (but scaled down from) Toronto’s, with 1578 drivers spending, on average, 52% of their time with-passenger (P3), and earning $8.80/hour after expenses. Passengers are waiting an average of 4.9 minutes for their trip.

After a 25% fare raise, the number of vehicles increases by 25%, to 1956. Drivers now spend only 41% of their time with passengers, and their income is essentially unchanged to within the precision of the simulation ($8.40 instead of $8.80 per hour). These additional unpaid drivers do enable Uber to deliver shorter wait times, down to 4.1 minutes.

If Uber raises prices while keeping their commission the same, more drivers find it worth their while to drive, attracted by the prospect of higher pay for each trip. But more drivers mean more competition for trips, and more unpaid idle time. At equilibrium, the competition eats away all the gains of the increased fare, and the pay of the average driver is unchanged by the price hike. There are more drivers, but they earn the same as before.

It may seem contradictory to claim that Uber has been subsidising drivers by taking losses, but that pay remains low. The distinct nature of gig work is that low prices (trip subsidies) do attract more drivers to the platform, but do not end up increasing the pay rate for individual drivers.

7 Conclusions

Currently, regulators erroneously assume the fraction of “engaged” time is fixed; as our model shows, it is a dynamic variable, such that as pay for engaged time increases, the proportion of time spent engaged decreases.

Free entry and exit, coupled with a large proportion of unpaid time, makes gig work unique. Uber drivers are just the first wave of gig workers whose standard of living needs to be addressed. Fitting Uber, and other gig-work platforms, into provincial labour standards and City of Toronto sustainability goals demands that the “deadheading” phase that is essential to the gig-work business model be addressed.

Uber has an incentive to provide large numbers of idle drivers: idle drivers provide low wait times and drive demand. The company has levers, from its price and commission choices to other tactics such as “gamification” and opaque and erratic pay practices, to keep drivers on the platform even at low pay. The fact that Uber continues to lose money shows that its popularity is a result of subsidy, not efficiency.

For the City of Toronto to address its broad goals of efficient and low-emission mass transit, it must abandon the idea that Uber can be a large component of this system. Taxi-type services (platform-mediated or not) certainly have a role to play in the urban transportation mix. However, as an inherently inefficient and high-emission competitor to public transit, the City must accept that it has a limited role in a healthy urban environment. The City must resume responsibility for establishing those limits, rather than delegating this oversight to the platforms themselves.

If minimum pay regulations to implement the Ontario Government’s Act are to improve the lot of Uber drivers and other gig workers, they must take account of all the time a driver spends on the platform, not just “engaged time”.

8 The author

Tom Slee has a PhD in theoretical chemistry, and has worked in the software industry for 30 years. His interest in what was then called the Sharing Economy led to the publication of “What’s Yours is Mine: Against the Sharing Economy” (OR Books, 2016), which was translated into seven languages. A data-driven investigation of Airbnb was reported in the New York Times . He lives and works in Waterloo, Ontario.

1 Ants fight wars

Ants seem to be simple creatures, far removed from us sophisticated humans, yet somehow they work together to grow crops, to farm herds of animals, and to live in cities with populations in the millions, complete with road networks, waste management systems, temperature control, and public health systems.

And, like us, ants fight wars.

1.1 Weaver ant wars

Here’s an example: a two-minute video in which David Attenborough describes a colony of weaver ants in the Australian tropics battling to “protect its fortress” from another colony.

This conflict not only looks like a war, it also fits the definitions of war commonly used by social scientists when they talk of human wars. The key elements of most definitions are scale, lethality (through the use of weapons), organization, and the fact that the conflict takes place between “political units”: the violence is mutual and it serves a purpose for each of the communities involved.

Let’s walk through the weaver ant video again, and see how each of these qualities is satisfied. Attenborough’s narration is in the captions, and our commentary follows each scene.

To set the scene: weaver ants live in colonies with populations that may reach half a million. The colonies are in the tree canopies of tropical jungles, and are dense enough that their well-marked and patrolled borders butt up against each other, filling the available space just as nation state territories fill out the continents of the world. This territorialilty is one aspect of being a political unit.

Figure 1 shows ants from one colony intentionally crossing the border into the territory of another in a coordinated fashion. (We will come back to that word “intention”.) Here is the first indication of organized violence.

Figure 2 shows how the weaver ants of the home colony are organized to play specialized roles. Specialization is even built into their bodies. The queen has a special role as the layer of eggs, but even though they are all half-sisters there are two distinct sizes of worker ant. The “minors” mainly look after the brood, while the “majors”, which are twice as big and sometimes called “soldiers”, carry out other tasks. Older major ants act as border guards. Again, these are all aspects of organization, one of the characteristics of war.

Figure 3 shows communication at work, another aspect of organization. The guard’s pheromone signal alerts other ants, “recruiting” them to defend the colony in a form of efficient and effective purposeful communication between different groups of ants within the colony.

In her 2021 book War: How Conflict Shaped Us, historian Margaret MacMillan says that wars are distinguished from bar fights by their scale. In narrating Figure 4, Attenborough tells us that the colony might be wiped out. That’s the potential for, and sometimes the reality of, hundreds of thousands of deaths – surely enough to qualify as “mass” violence.

Figure 5 brings us to weapons. Ant weapons differ from human weapons in that they are part of the body rather than being a prosthetic device. Still, this targeted adaptation equips the ant with the ability to inflict damage on multiple enemies, at a distance. The level of lethality at work is enough that we would agree with Attenborough when he calls these adaptations “weapons”.

One not-so-obvious component of definitions of war is that the violence (or at least the potential for it) is mutual. One-sided violence in the ant world is usually called predation: war takes between political units. In Figure 6, the home guard goes on the offensive. The organized violence in this conflict is indeed mutual.

For decades, the late E. O. Wilson and Bert Holldobler were the foremost authorities in the study of ants. They wrote that “The colony is the unit of meaning in the lives of ants. The workers’ loyalty to it is nearly total”, and in Figure 7 we see worker ants that have died for the greater good and for the future of the colony. Surely these colonies count as political units.

In short, this weaver ant conflict has all the makings of what we would call a war.

1.2 Other ant wars

There are about 20,000 species of ants in the world, so there’s no one form of ant war. But weaver ants are not the only ones to engage in conflicts that fit this definition of a war. Here are some other ant wars

• In the temperate climes of Europe, wood ants form colonies which, like weaver ants in the tropics, expand until they border on each other. During winter the wood ants hibernate and boundaries between the colonies become blurred. In the spring they fight until boundaries are re-established, after which they commonly live side by side for the rest of the summer. The spring battles may be mere skirmishes, but in some cases they lead to the destruction of one colony, with the victorious colony using the bodies of vanquished ants as food to build up their nutritional stores after the winter.

• In the 19th century, Argentine ants were accidentally exported around the world on ships, and they have settled in so well in their new homes that they have established what are called “supercolonies”, with thousands of nests, thousands of queens, and up to a trillion workers. There is a massive supercolony in California, which stretches from the Bay Area down to San Diego. But near San Diego it borders another colony where a war between the two is in progress, which has killed millions of ants each month for the last hundred years. By number of casualties, it is more deadly than any other known war.

• In the tropics of the Americas, Leafcutter ants construct the most sophisticated of all ant societies. Their colonies have a single queen and are maintained by several different castes of worker ant of massively different sizes, which practise an elaborate form of fungus farming (see Figure 8). They fight nearby colonies, but they also have to fight off invading colonies of army ants, which don’t maintain a static nest but maraud through the jungle as a column of hundreds of thousands of ants on the move. Leafcutters adopt different tactics depending on the enemy: when fighting other leafcutter ant colonies, minor workers do most of the fighting, but when army ants invade it is the majors who play the most important roles.

One difference between ant wars and human wars is that all human wars are between members of the same species, but some ant wars, like those between the leafcutters and army ants, are between different species. Is this still a war? If you watched the Lord of the Rings and didn’t complain about the phrase “the War of the Ring”, you’ve already accepted the principle. Perhaps it makes sense to think of inter-species ant wars as “asymmetric warfare” between armies using different technologies and resource bases, like that carried out in Afghanistan over the last 20 years.

2 Asking questions about wars

As epigraph to Chapter 1 of her book “War”, Margaret MacMillan chose a quotation from Frederic Manning: “War is waged by men; not by beasts, or by gods. It is a peculiarly human activity.” She goes on to say that (p xi): “War raises fundamental questions about what it is to be human and about the essence of human society”. The idea that war is uniquely human, and tied to our humanity, is widespread, but ants show that war is not a “peculiarly human activity”.

Still, it is fair to ask “so what?” Isn’t ant war just a curiosity? A surface similarity? After all, when we think about human wars, we do not seek simply to classify events as “war” or “not-war”. We want to answer urgent questions, like: how do wars come about? what effects do wars have? how can wars be avoided or limited? how are wars won and lost? Addressing these questions leads us inevitably to the worlds of international relations, social institutions, human history, ethics, human nature, the nation state, technological innovation, law, diplomacy, and more. Aren’t these, as MacMillan suggests, uniquely human topics?

So here is the next point we wish to make: all those questions we ask about human wars can be asked of ant wars too, and the answers are not so different. We will go through a list of topics, but as preface we’d like to highlight two themes that run through all of them.

The first is that any study of ants encourages a “system viewpoint” of war. The causes and effects of particular wars are not of as much interest when we look at ants. In ants, there is an ecosystem and there are inhabitants, and these generate particular forms of ant society, and in some ant species these societies periodically engage in and are shaped by wars. Instead of thinking about causes and effects, the focus is on the co-evolution of war and other aspects of ant societies: wars are just one aspect of a bigger system. This approach is not unheard of among those who study human war of course: think of historian and sociologist Charles Tilly’s aphorism “War made the state and the state made war”, but the perspective of war as a system is more obvious in the world of insects.

The second theme is a perspective of “convergent evolution” between ant and human societies. Here we don’t mean a biological determinism, but evolution in a broader sense. Here is a beautiful paragraph by Canadian author Ronald Wright, invoked by John Gowdy and Lisi Krall writing about the origins of ants’ ultrasocial behaviour:

What took place in the early 1500s was truly exceptional, something that had never happened before and never will again. Two cultural experiments, running in isolation for 15,000 years or more, at last came face to face. Amazingly, after all that time, each could recognize the other’s institutions. When Cortés landed in Mexico he found roads, canals, cities, palaces, schools, law courts, markets, irrigation works, kings, priests, temples, peasants, artisans, armies, astronomers, merchants, sports, theatre, art, music, and books. High civilization, differing in detail but alike in essentials, had evolved independently on both sides of the earth.

The similarities between Tenochtitlán and Madrid were not just metaphors or similes. The two are examples of a single thing – a city – which has developed independently, and is hosted in, different societies. A common underlying dynamic generates similar outcomes even in these two different environments. So, we would argue, war is a single thing: a system that has developed independently, hosted in both ant and human societies.

With those themes in mind, let’s take a tour around the study of war. Approaches to understanding war take place at three levels: the international system of nation states, the nature of the states that make up this system, and the nature of the individuals that make up those states.

3 So what?

We finish by returning to the question of “so what?” We’d like to suggest two possibilities, one fairly specific, and one more general.

1 Introduction

Entomologists have long looked to human societies and events to help them understand the world of insects. They have spoken for more than a century of “wars” between insect colonies¹ and, to describe what happens during these wars, they have imported a rich military vocabulary. Terms such as “armour,” “weapon,” “battle,” “warrior,” “soldier”, “phalanx”, “bivouac”, “kamikaze”, and “mercenary” all appear in insect research.

The traffic in the opposite direction has been much lighter. It is true that “insect wars” have become increasingly familiar to us through television and YouTube, often narrated by the mellow tones of David Attenborough, but scholars of human wars still view their subject as unique to our species.

As just one example: October 2020 saw the publication of Margaret MacMillan’s book War: How Conflict Shaped Us,² based on her 2017 Reith Lectures. As epigraph for the Introduction she chose a quotation from Svetlana Alexievich: “War remains, as it has always been, one of the chief human mysteries”.³ For Chapter 1 she chose a quotation from Frederic Manning: “War is waged by men; not by beasts, or by gods. It is a peculiarly human activity.”⁴

We can think of several reasons why scholars of human war have largely ignored insect wars:

• Caution. Schooled in the humanities or social sciences they may be reluctant to venture into the study of animal behaviour, and especially reluctant to project human forms, emotions and ideas onto other creatures, portraying these creatures in ways that suit particular human agendas.

• The spectre of “biological determinism” and social Darwinism. The sociobiological application of natural selection to explain aspects of human societies caused great controversy in the 1980’s and leads uncomfortably close to saying “war is in our genes”. If war is claimed as a widespread reality in nature, fashioned unalterably into the bodies and minds of living beings, is this to say that human beings are fated to be entangled in war forever?

• The obvious distance between individual humans and insects on the evolutionary tree. Can creatures as unintelligent and tiny as ants really participate in actual war? Surely this is all just a matter of convenient metaphor? If we were to find war among chimpanzees, our evolutionary cousins, that would be one thing, but among insects?

• The question, “so what?” Even if we conclude that insects do engage in war, of what possible use is that to those of us concerned with human history and human action, which are surely varied and complex enough by themselves?

This essay takes seriously the idea that speaking of “insect wars” is not just metaphor or imagery. It argues that insect war and human war can usefully be thought of as instances of the same thing, developing in response to similar forces. It makes the following claims:

• Definitions of “war” either fit insect wars or can be extended to do so without damage to their meaning. There are no reasonable grounds for the assertion that wars are uniquely human events.

• Acknowledging that “insects fight wars” focuses attention on war as a system, emphasizing that wars (as discrete events) are inextricably linked to the organization of societies so as to fight wars effectively.

• Beyond the basic definition of war, many questions that we ask about human wars are also asked of insect wars and many concrete features of human war can also be found in insect wars. We believe there are opportunities for learning in both directions, and suggest some possible avenues.

2 What makes a war?

It has long been recognized that ants display behaviour that looks like war. Here is Henry David Thoreau in 1854:

One day when I went out to my wood-pile, or rather my pile of stumps, I observed two large ants, the one red, the other much larger, nearly half an inch long, and black, fiercely contending with one another. Having once got hold they never let go, but struggled and wrestled and rolled on the chips incessantly. Looking farther, I was surprised to find that the chips were covered with such combatants, that it was not a duellum, but a bellum, a war between two races of ants, the red always pitted against the black, and frequently two red ones to one black. The legions of these Myrmidons covered all the hills and vales in my wood-yard, and the ground was already strewn with the dead and dying, both red and black ¹⁰.

Is this really a “war” or is Thoreau just seeing a surface similarity? Identifying what counts as war and what does not is not just a matter for dictionaries. In most societies intentional killing is forbidden under normal circumstances, but is permitted, even required, and often lionized if it takes place in war. Over millennia, humans have repeatedly separated war from the everyday activities of society, often by means of symbolic acts and ceremonies, so that the norms of the battlefield do not spill over into civilian society.

Margaret MacMillan identifies, in the early pages of War, a set of attributes associated with war. Here we show that conflicts between insect societies have these attributes.

3 War as a system

When we talk of human wars, we do not seek simply to classify events as “war” or “not-war”; we want to know how wars come about, what effects they have, how they can be avoided or limited, and how they are won and lost. Addressing these questions leads us inevitably to the worlds of international relations, social institutions, ethics, human nature, the nation state, technological innovation, law, diplomacy, and more. Wars are not merely events that erupt at random times and places: they are the visible, above-water part of a much bigger iceberg, inextricably connected to these other realms of nation states, ethics and so on, and these realms seem uniquely human. Together, we may call wars (the events) and those other heterogeneous realms that they are connected to, a war system. We may think of the system as being in a “hot state” when a war is being fought, and in a “cold state” between wars. The violence might not be there, but all the elements that it is connected to are still in evidence.

MacMillan does not use the term, but she treats war as a system in that she traces how war (events) shape and are shaped by other entities. Her book’s subtitle is How Conflict Shaped Us, and she pursues the theme throughout Chapter 1. She quotes historian and sociologist Charles Tilly’s well-known aphorism that “War made the state and the state made war”²³ and traces the growth of centralized state power in Europe by the 18th century (p23) to the wars of that time.

As MacMillan emphasizes, the war system is not tied to any particular form of political organization; it has made its home in many different human societies throughout history. When feudalism is replaced by capitalism, with attendant political transformations, war may be transformed but it marches on. When capitalism is displaced or modified by socialistically inclined states, war changes but continues. None of this means it cannot be constrained or defanged, but it does suggest that human war itself is a complex, adaptable entity with its own nature.

But surely this picture of a war system is quintessentially about us as humans? Even if ant conflicts fall under a definition of “war” can we learn anything from them about the questions we really want to ask? We suggest that the answer is “yes”. The formicidae do not hand out medals to victorious heroes or sit at negotiating tables any more than they wave tiny flags as they rush into battle, but many of the connected phenomena that form the human war system have their own parallels in the insect world. War, we suggest, is a system in nature, which is hosted in both ants and humans.

War is an example of convergent evolution: the independent appearance of similar complex features in different contexts. An example of convergent biological evolution is the eye, which has developed independently along multiple separate branches of the evolutionary tree. We can say that eyes are hosted in many species. The city is an example of convergent social evolution, which became hosted in independent civilisations, as this passage from Ronald Wright, quoted by Gowdy and Krall, beautifully illustrates:²⁴

What took place in the early 1500s was truly exceptional, something that had never happened before and never will again. Two cultural experiments, running in isolation for 15,000 years or more, at last came face to face. Amazingly, after all that time, each could recognize the other’s institutions. When Cortés landed in Mexico he found roads, canals, cities, palaces, schools, law courts, markets, irrigation works, kings, priests, temples, peasants, artisans, armies, astronomers, merchants, sports, theatre, art, music, and books. High civilization, differing in detail but alike in essentials, had evolved independently on both sides of the earth.²⁵

Using the word “eyes” to name the octopus’s organ of vision is not just a metaphor for mammalian eyes, and use of “city” to describe Tenochtitlán is not just a metaphor for European cities. In both cases, a common underlying dynamic generates a similar outcome in different environments. In both cases, the processes are broadly evolutionary, in the sense that we would not say that cities cause roads or that roads cause cities, but that each is necessary for and accompanies the other. In the same way, the system of war is caused by and causes large-scale complex societies, be they human or insect.

Wars in humans and ants are not identical in all cases, any more than an octopus eye is identical to a human eye, but many essential features of the war system are displayed in many of those environments in which it is hosted, be they human or ant.

4 Territory: the politics of nation states

As MacMillan observes (above), wars take place between organized societies which “have existed over considerable time, usually in their own territory”. Humans have been fighting wars since we began farming, which produced fixed valuable resources—hence territories—that were worth fighting over, and also sustained communities large enough to mobilize mass, organized, violence.³⁵

Nation states are defined, in part, by their territory, and political scientists seek explanations of war between nation states at three levels: the individual, the societal, and the systemic.³⁶ A neo-realist school sees the behaviour of states as arising from self-interested actions in a world of similar entities jostling for survival in an anarchic (that is, ungoverned) system. These neo-realists “are interested in how structural pressures—as opposed to the complexities of psychology or domestic politics—shape international outcomes”.³⁷

Ant colonies may be the ultimate neo-realist political unit. Unencumbered by domestic politics or moral considerations, colonies act out of self-interest, with survival as the only metric of success. They have discovered ways to protect their territories from attackers and to expand their territories when the opportunity presents itself.

Weaver ants live in the tree canopies in the “Old World” tropical rainforests of Africa, Asia, Australia, and the Western Pacific.³⁸ They build nests by weaving together leaves using larval silk. Their colonies consist of many nests, and may have populations in the millions. Weaver ants are “territorially dominant”: unlike many more selective ant species they feed on many kinds of insect and other prey, and so it makes sense to exclude all competitors from the entire area surrounding and between their nests.³⁹ Each colony marks out whole trees or large tree limbs as its own territory, and these territories cover the available canopy so densely as to form a “mosaic” of non-overlapping domains,⁴⁰ much as nation state territories fill the map of the world, and each colony defends its own territory against other colonies of their own species and against other competitive species.

Weaver ants mark the edges of their territory with brownish spots containing territorial pheromones unique to each colony,⁴¹ and older colony members actively patrol and monitor this boundary, raising the alarm in case of encroachment. As with other animals, weaver ants often fight more aggressively defending their own territory than in attack,⁴² and so the area outside the boundary may become a strip of “no ant land” which other colonies actively avoid. Boundaries may be stable for a number of years, during which time weaver ants exemplify the old human adage “if you want peace, prepare for war”.

But weaver ants also show us that this prescription is good only as long as the environment remains stable. Wars break out when the environment receives a shock: when a colony dies, or when a particularly good season leads to population growth. And these wars can be brutal: one colony may take over an entire tree, and may even destroy much or all of the neighbouring colony.

In Central America, the ant “Pseudomyrmex ferruginea F. Smith” demonstrates an even more dramatic case of territoriality.⁴³ It has developed a symbiotic relationship with a particular species of acacia tree. The tree provides suitable domiciles and a regular supply of food for the ants, while the ants have taken on a strong territorial sense coupled with extreme aggressiveness, and developed an array of weapons to protect their tree. The ants are inseparable from their territory and will attack just about anything that encroaches on it, be it ant, mammal, bird or plant. Colonies fight fiercely with other ant colonies, of the same or different species, for exclusive possession of the tree, and will not hesitate to exterminate their opponents’ communities. The colony may rest content with the single tree it has come to regard as its own, but in some cases, it becomes expansionistic, occupying up to 20 trees.

Political scientists treat nation states as acting with intent. They (or their leadership) take decisions, and weigh the costs and benefits of their actions. A “strategy” at the level of a nation state is the result of some form of deliberation. From a game theory perspective, they are assumed to act “rationally”. Ant colonies have no such deliberative qualities, no leadership to decide on a strategy, and no hierarchy to implement it. They arrive at their behaviour through millennia of trial and error. That rational decision making and evolution can lead to the same strategic outcomes has been known since Maynard Smith used game theory to understand animal behaviour such as altruism.⁴⁴

Like political scientists, entomologists seek answers at three levels: the individual, the political unit, and the ecosystem to which these units belong. Ant colonies and nation states each act as coherent political units, even if their decision-making mechanisms and social ordering could hardly be more different. Territorial war is not unique to humans: it is a complex phenomenon that has emerged in both human and insect ecosystems, under common pressures and for similar reasons.

5 Us and them: the sociology of identity

Nation states are more than areas on a map: they must also be politically cohesive if they are to succeed, and this requires a clear sense of who is part of the nation and who is not, particularly in times of war. Discourses of common origin and shared blood relationship are common (terms like “motherland”) even when there is no historical or material basis for it. During a war, each side may produce propaganda to emphasize how different “we” are from “them”, further deepening the divide.

Individuals face a trade-off between pursuing their individual ends and contributing to the strength of the society of which they are a part. What sociologists call in-group/out-group distinction is associated with increased co-operation among community members in the face of external competition; “conflict with an out-group increases the cohesion of a well-defined in-group”.⁴⁵ The “scapegoat hypothesis” or “diversionary theory of war” suggests that leaders may deliberately create or maintain external conflicts to sustain compliance within their state.⁴⁶

The in-group/out-group distinction is also important when it comes to actual combat: effective war-fighting demands that soldiers distinguish “us” from “them”, not just by their uniforms, but by how they react to those uniforms. Episodes such as the famous Christmas Truce of 1914, in which soldiers realise their common humanity across the us/them divide, threaten the effective pursuit of war.

The ability to distinguish members of our own group from those who are not, and to make appropriate behavioural adjustments, is widespread in animals. Maintaining the distinction even where very large communities are involved is much rarer. In most primates, for example, animals recognize in-groups by recognizing the individuals that belong to it. In larger societies, individual recognition is not possible; ants share with humans the ability to form what Moffett calls “anonymous societies”: individuals recognize members of the same society even if they do not recognize (and may never have met) the individual itself. ⁴⁷

For ants, the cue to distinguish members of one colony from another is a complex cuticular hydrocarbon (CHC) profile on the waxy surface of the hard outer cuticle.⁴⁸ When ants meet, they sense each other’s CHC profile through their antennas. The CHC profile is sometimes called an “odour” although the sense involved is somewhere between taste, scent, and touch.

Members of a single-queen, single-nest colony are close genetic relatives. Membership as read by odour matches genetic relatedness and “colony recognition” is the same as “kin recognition”. In these cases, the odour acts as a sign to indicate deeper ties, marking who to collaborate with, and who to confront.

But for some ant species, the CHC profile and the “in-group/out-group” distinction has become divorced from relatedness. Many species of wood ant construct extended colonies with multiple nests and queens, in which each ant is only distantly related to others,⁴⁹ yet colony group membership is still defined by CHC profile. If nests were isolated, the odour would grow to differ among nests over time, but worker ants move back and forth between nests. Ants exchange odour when they meet through grooming (licking) and trophallaxis (mouth-to-mouth transfer of CHCs),⁵⁰ and in this way the CHC profile is spread among members. There have even been suggestions that worker ants carry brood and other workers between nests to maintain a common odour.⁵¹

The CHC profile that a colony maintains this way can be thought of as a “cultural” identity – a distinguishing characteristic that serves not only as a marker of identity but defines it, and which is actively maintained through exchange throughout the members of a colony.⁵² The colony is making a cultural investment in maintaining a common identity among its members: a social infrastructure of cohesion.

An extreme case of ingroup/outgroup recognition among ants is that of the “unicolonial” Argentine ant. A hundred years ago, humans inadvertently transported Argentine ants on ships from their native habitat to new homes where they have become invasive. In these new environments they have formed huge multi-nest “supercolonies”, and these colonies have engaged in wars.

The Argentine ant has established supercolonies on several continents. A community discovered in western Europe⁵³ stretches 6000 km from northern Italy, through the south of France to the Atlantic coast of Spain—with billions of ants—perhaps even a trillion—occupying millions of nests.⁵⁴

Colony members maintain the ability to identify members of the same colony across continents,⁵⁵ while still identifying outgroup members when neighbouring supercolonies collide.⁵⁶ The largest known ant war on earth is taking place among two Argentine ant colonies in North America. “Each month millions of Argentine ants die along battlefronts that extend for miles around San Diego, where clashes occur with [the”Large Colony” and] three other colonies in wars that may have been going on since the species arrived in the state”.⁵⁷ Researchers observing the conflict concluded that “[O]ver 15 million workers would have died over the six-month period of the study”.⁵⁸ Move a”Large Colony” Argentine ant 800 kilometres from San Francisco to San Diego and she will still be “home”, accepted by surrounding ants. On the other hand, move an ant a few centimetres across an invisible (to humans) border in the outskirts of San Diego, and she will probably be killed by the members of the neighbouring Lake Hodges colony.⁵⁹

The anonymous nature of the ingroup/outgroup distinction, so clear in ants, emphasizes that during war the enemy is defined as a group, not as individuals. This anonymity is one feature that makes killing in war different from killing in peacetime. The killing is not personal, it is required by the mass nature of the conflict. At the same time, the individual combatant is disposable from the point of view of the community, entering hostilities with combatants from another colony, at great risk to themselves.

6 Warriors: the psychology of combat

Ants and humans have both, in their own ways, identified techniques for building successful communities in the face of competition from neighbours. Clear boundaries of territory and membership play important roles in limiting unnecessary and accidental outbreaks of war. Still, wars do break out, and when it does, the behaviour of community members changes. Behaviour that is damaging to the community in normal times (killing, aggression) is now necessary.

Human societies have long struggled with the need to develop effective warfighting abilities, while limiting that organized violence to deal with external threats, without overflowing into the daily life of peacetime society. Rituals before going to war, ethical codes for soldiers, heroic myths to emulate all prepare warriors to be ready to kill, and ready to die. They also separate war-time from peace-time norms, with varying degrees of success.

Ant societies face the same challenges. Mustering a collective response to a threat demands the rapid raising of an alarm, alerting others to the presence of an enemy. It must assemble nestmate responders and set in motion a change in their behaviour, so that they are disposed to aggression.

“Recruitment” is the term for communication bringing nestmates to a location where activity is required. Colony members recruit others for a range of activities, a prominent one being response to threats such as floods, attacks by large animals (bears, anteaters), or attacks by other ants. Recruitment in the face of threat must be fast, and some species of ants respond so quickly to a flooding domicile that “one or two workers can mobilize a large fraction of the colony in 30 seconds or less and even initiate colony immigration”.⁶⁵

Recruitment typically involves chemical signals and mechanical gestures. It is more complex than a simple cry of “alarm!”: faced with combat, a weaver ant uses a short-range recruitment pheromone to summon immediate reinforcements, but also returns to the nest, depositing a different pheromone to mark a trail back to the site of conflict, and “jerks its body at passing ants to alert them to the ongoing combat”. Weaver ants may use different pheromones when recruiting for other purposes, such as to occupy a previously unoccupied space for the nest.⁶⁶

The “decision” to recruit also has subtleties. We have already seen that wars among wood ants break out only at times of the year when nutrition is scarce. Ants may also behave differently depending on whether the queen is nearby,⁶⁷ and some fighting ants can assess group sizes, recruiting nestmates only when they outnumber opponents.⁶⁸

Once mobilized, risk-taking and self-sacrifice are common in warfare, in ants as in humans. Students of formic behaviour have referred to this as evidence of “extreme altruism.” A remarkable example is the “suicide warriors” of Camponotus.

Two huge glands, filled with toxic secretions, run from the bases of the mandibles all the way to the posterior tip of the body. When the ants are pressed hard during combat…they contract their abdominal muscles violently, bursting open the body wall and spraying the secretions onto the foe.⁶⁹

7 The militarized state

Despite war’s comparative rarity, it has often precipitated long-lasting changes in the organization of societies. The demands of war require the state to take a central role, and that role has not been given up with the end of conflict.

In ants, war is one of the forces driving the formation of new “castes”. Among ants, function is often connected to physical type or state and the term “caste” has been used for a long time, for better or worse, to express this convergence. “A caste is any set of a particular morphological type, age group, or physiological state (such as inseminated versus barren) that performs specialized labour in the colony”.⁷⁵

Among the workers, who are all female, the most common morphological types are minor, media and major, which have differing body size and other specialized physical attributes. The difference in size between minor and major nestmates reaches an apex in marauder ants, where the major ants may weigh five hundred times more than a minor, and where the minor may ride on the head of the major.⁷⁶

War is “a strong and recurrent pressure” driving the development of distinct castes in a species, “including the evolution of a functionally specialized and large soldier caste”.⁷⁷ As an aside: The nature of division of labour is still controversial. Deborah Gordon argues that “‘Division of labour’ is a misleading way to describe the organization of tasks in social insect colonies, because there is little evidence for persistent individual specialization in task.”⁷⁸ Robert Jeanne responds that “Division of labour is not a process or a misleading concept”.⁷⁹ Either way, the topic is lively.⁸⁰

Consider three ant species of the genus Camponotus, which live in the southern United States and are subject to predatory raids by army ants. One of those species responds to raids by flight, evacuating the nest and carrying their brood with them. The other two species respond by fighting, with the majors taking a leading role. All three species have “major” and “minor” castes, but in the one that flees the majors are only slightly larger than the minors. In those that fight, the majors are much larger. The researchers conclude that the army ant raids are sufficient to produce this specialization.⁸¹

Across humans and ants, war has driven changes in society to enable it to put all its weight behind the achievement of a single goal, demanding increasingly specialized roles and communication techniques, and changing the relationship among members of society. One has achieved this goal through a centralized state and the other through decentralized coordination.

8 Technology and innovation

The militarized state is the organization of society so as to be able to fight wars effectively. But innovation is not limited to social organization: there is also a long history of war-driven technological innovations in both ants and humans, most obviously in the development of more lethal weapons and more effective defences.

Ants have not only developed a distinct soldier “caste” in some species, but have also equipped them with specialized weaponry, designed to be effective against specific groups of foes. Science journalist Erich Hoyt⁹³ wrote about formic war in his book, The Earth Dwellers, and he distinguishes “mandibular techniques” from “chemical warfare.” Under the first category (see p. 206) he says:

“Hatchet-shaped mandibles can slice through the chitinous exoskeleton or chop off the legs of an opponent.” (Examples include many Pheidole species as well as leaf cutters.)“Saber- or hook-shaped mandibles with pointed tips that turn in at the end.” “When an enemy ant gets its head or body caught inside these jaws, the result is instant death.” (Examples include various of the “swarm-raiding and column-raiding army ants, and some of the notorious slave-making species.”) “Trap mandibles, as used by the dacetine tribe and some of the ponerines, consist of elongated mandibles that snap shut convulsively, impaling prey or enemies on the sharp teeth at or near the tip.”

Hoyt notes that “almost all ants use some chemical warfare,” the weapons of which he distinguishes as “stings, sprays and secretions.”

The sting is one of the defining features of ants, and is particularly effective against threatening vertebrates. As ants became more abundant, their greatest threat shifted from vertebrates to other ants and some species lost their sting, or modified it into a “chemical weapons delivery system”, enhancing their ability “to attack and fight off and win battles against other ants”.⁹⁴

We have already seen the “kamikaze” weapons of Camponotus ants, but chemical weapons can also be used for defence. In northern California the native winter ant has been able to preserve itself against the Argentine ant while other species have succumbed; its success is partly thanks to a secretion they apply directly to the body of their invasive foe, killing it within an hour, four times out of five.⁹⁵

Technological innovation is not limited to weapons. Leafcutter ant soldiers have developed a biomineral armour overlaying their exoskeleton which provides additional protection in battles and against disease.⁹⁶ The 119 different species of turtle ants have been known as “an iconic example of caste specialization” since Darwin. Turtle ants make their homes in pre-existing cavities⁹⁷ and when the nest is threated the elaborately armoured heads of their soldier caste act as living barricades. Some cavities have entrances much larger than the head of one soldier and among the species that live there, soldier heads are square or dome-shaped and several soldiers will cooperate to block the entrance. In contrast, for species that live in cavities with smaller entrances, a single soldier will block it by locking its disc or dish-shaped head into place.⁹⁸

9 Limited war and ethics

War is a costly enterprise, even for the victor, who may be left weakened by the extremities of combat. Humans throughout history have searched for ways to avoid war where possible, and to limit the devastation of war by devising rules for its conduct. Yet MacMillan quotes Pancho Villa as saying “limiting war is a silly idea: war is not a game”. From a different perspective, White House Counsel Alberto Gonzalez was quoted as saying that in the War on Terror, the Geneva Conventions are “quaint”.

Is the search for ways to limit war a pipe dream? Here, ants may offer signs of hope: they have found ways to limit the frequency and intensity of their wars, and nobody could accuse them of being quaint or silly.

10 Asymmetric and unilateral violence

Both ants and humans take part in raids and invasions to plunder the resources of other communities. Here we focus on the case where the “resources” to be acquired are the inhabitants themselves (often taken in the brood stage), to be used as either food or labour by the invading colony. The killing of other animals as food is called predation; the acquisition of other ants for use as labour is called slavery. Both have been identified as distinct systems with their own dynamics, but both overlap with war.

We have already met examples of each. For both there is a spectrum of mutuality: to the extent the targets fight back in an organized manner, we may call these events “wars”, but there are also cases where the target does not fight back. In such cases, mass, organized lethal violence may be carried out by one political unit against another, but it will not be mutual.

11 Discussion

In this article a definition of war has been proposed that is intentionally broad, and that casts the net widely enough to include species other than homo sapiens. The tradition of conceiving of war as a system, combined with this broad definition of war, permits the uncovering of war as a system in the natural world.

Drawing lessons from ants in a simplistic sense (“consider the ant, thou sluggard”) holds no attraction for scholars today, but we can still be stimulated to see and to think in new ways when we study systems common to humans and to life forms other than the human. The consideration of formic war allows old questions to be asked in new ways. Although it is beyond this article’s scope to pursue this topic in detail, we conclude with some examples of questions that may benefit from the perspective outlined in this paper.

12 Conclusions

Appreciating that qualities once thought to be uniquely human are found elsewhere in nature is often a valuable step in itself. Understanding that intelligence and language are not limited to humans has helped us better appreciate our place in the natural world. Perhaps there is something to be gained by appreciating that our complex societies too, with their achievements and tribulations, are not as unique as we once thought. Accepting that war is not limited to humanity is one small piece with the other adjustments we have made regarding our place in nature.

Ant wars show that human ingenuity is not essential to the development of social complexity, and humans are not the only species capable of developing the set of competences that warfare involves.

Treating war as a peculiarly human activity is particularly inaccurate because most of the wars fought on this planet have been insect wars and most of the deaths from war have been insect deaths. Almost all the fatalities of war are insects. Some insect wars are of long duration, with cumulative fatalities in the billions.

The variety of behaviours among ants is astonishing, and more subtleties and novelty continue to be discovered by myrmecologists’ careful observations and experiments. An obvious entry point to the idea of treating war as a system in nature is to ask whether we can use ant warfare to predict the outcome of human warfare. We consider this unlikely: as with other highly non-linear problems the range of possibilities is too great. Instead, we see potential for ant behaviour to provide a source of inspiration or “hypothesis generation” for those studying war in a human context, and hope that our suggestions and attempts in this essay may prompt others.

13 Bibliography

Adams, Eldridge S. “Boundary Disputes in the Territorial Ant Azteca Trigona: Effects of Asymmetries in Colony Size.” Animal Behaviour 39, no. 2 (February 1, 1990): 321–28. https://doi.org/10.1016/S0003-3472(05)80877-2.

Adams, Eldridge S., and Nicola J.R. Plowes. “Self-Organizing Conflicts: Group Assessment and the Spatio-Temporal Dynamics of Ant Territory Battles.” Behavioural Processes 162 (May 2019): 119–29. https://doi.org/10.1016/j.beproc.2019.01.009.

Adams, Rachelle M. M., Joanito Liberti, Anders A. Illum, Tappey H. Jones, David R. Nash, and Jacobus J. Boomsma. “Chemically Armed Mercenary Ants Protect Fungus-Farming Societies.” Proceedings of the National Academy of Sciences 110, no. 39 (September 24, 2013): 15752–57. https://doi.org/10.1073/pnas.1311654110.

Allies, Anthony B., Andrew F. G. Bourke, and Nigel R. Franks. “Propaganda Substances in the Cuckoo Ant Leptothorax Kutteri and the Slave-MakerHarpagoxenus Sublaevis.” Journal of Chemical Ecology 12, no. 6 (June 1, 1986): 1285–93. https://doi.org/10.1007/BF01012348.

Anderson, Carl, and Nigel R. Franks. “Teams in Animal Societies.” Behavioral Ecology 12, no. 5 (September 1, 2001): 534–40. https://doi.org/10.1093/beheco/12.5.534.

Arquilla, John. “Perils of the Gray Zone: Paradigms Lost, Paradoxes Regained.” PRISM 7, no. 3 (2018): 118–29.

Bakker, Theo C. M. “Editorial: Special Issue on Integrative Analysis of Division of Labor.” Behavioral Ecology and Sociobiology 70, no. 7 (July 1, 2016): 993–94. https://doi.org/10.1007/s00265-016-2132-0.

Bellany, Professor Ian. “Fighting Asymmetric Wars: An Application of Lanchester’s Square‐law to Modern Warfare.” The RUSI Journal 147, no. 5 (October 1, 2002): 72–76. https://doi.org/10.1080/03071840208446819.

Brady, Patrick. “Between Peace and War: Gray Zone, Bright Line, or Dialectic? | RealClearDefense,” December 7, 2020. https://www.realcleardefense.com/articles/2020/12/07/between_peace_and_war_gray_zone_bright_line_or_dialectic_652193.html.

Brady, S. G. “Evolution of the Army Ant Syndrome: The Origin and Long-Term Evolutionary Stasis of a Complex of Behavioral and Reproductive Adaptations.” Proceedings of the National Academy of Sciences 100, no. 11 (May 27, 2003): 6575–79. https://doi.org/10.1073/pnas.1137809100.

Brandt, M., Jürgen Heinze, Thomas Schmitt, and Susanne Foitzik. “Convergent Evolution of the Dufour’s Gland Secretion as a Propaganda Substance in the Slave-Making Ant Genera Protomognathus and Harpagoxenus.” Insectes Sociaux 53 (August 1, 2006): 291–99. https://doi.org/10.1007/s00040-006-0871-z.

Brooks, Rosa. How Everything Became War and the Military Became Everything. Simon & Schuster, 2016. https://www.simonandschuster.com/books/How-Everything-Became-War-and-the-Military-Became-Everything/Rosa-Brooks/9781476777870.

Buschinger, A., and U. Maschwitz. “Defensive Behavior and Defensive Mechanisms in Ants.” In Defensive Mechanisms in Social Insects (Hermann, H. R., Ed), 95–150. New York: Praeger, 1984.

Christensen, Charlotte, and Andrew N. Radford. “Dear Enemies or Nasty Neighbors? Causes and Consequences of Variation in the Responses of Group-Living Species to Territorial Intrusions | Behavioral Ecology | Oxford Academic.” Behavioural Ecology 29, no. 5 (October 2018): 1004–13.

Dejean, Alain, Bruno bullet, Bruno Corbara, Jérôme bullet, Jerome Orivel, Maurice bullet, and Maurice Leponce. “Rainforest Canopy Ants: The Implications of Territoriality and Predatory Behavior,” January 1, 2007.

D’Ettorre, P., and J. Heinze. “Sociobiology of Slave-Making Ants.” Acta Ethologica 3, no. 2 (April 23, 2001): 67–82. https://doi.org/10.1007/s102110100038.

Ellis, S., and E. J. H. Robinson. “Polydomy in Red Wood Ants.” Insectes Sociaux 61 (2014): 111–22.

Ettershank, G., and J. A. Ettershank. “Ritualised Fighting in the Meat Ant Iridomyrmex Purpureus (Smith) (Hymenoptera: Formicidae).” Australian Journal of Entomology 21, no. 2 (1982): 97–102. https://doi.org/10.1111/j.1440-6055.1982.tb01772.x.

Franks, Nigel R., and Lucas W. Partridge. “Lanchester Battles and the Evolution of Combat in Ants.” Animal Behaviour 45, no. 1 (1993): 197–99. https://doi.org/10.1006/ANBE.1993.1021.

Frederic Manning. The Middle Parts of Fortune: Somme and Ancre, 1916. Vintage Classics, 2014. http://gutenberg.net.au/ebooks02/0200261.txt.

Giraud, Tatiana, Jes S. Pedersen, and Laurent Keller. “Evolution of Supercolonies: The Argentine Ants of Southern Europe.” Proceedings of the National Academy of Sciences 99, no. 9 (April 30, 2002): 6075–79. https://doi.org/10.1073/pnas.092694199.

Gordon, Deborah M. “From Division of Labor to the Collective Behavior of Social Insects.” Behavioral Ecology and Sociobiology 70, no. 7 (July 1, 2016): 1101–8. https://doi.org/10.1007/s00265-015-2045-3.

Gowdy, John, and Lisi Krall. “The Economic Origins of Ultrasociality.” Behavioral and Brain Sciences 39 (ed 2016). https://doi.org/10.1017/S0140525X1500059X.

Guénard, Benoit. “The Global Ant Biodiversity Informatics (GABI) Database: Synthesizing Data on the Geographic Distribution of Ant Species (Hymenoptera: Formicidae).” Myrmecological News 24 (March 2017): 83–89.

Heinze, Jürgen, Susanne Foitzik, Andreas Hippert, and Bert Hölldobler. “Apparent Dear-Enemy Phenomenon and Environment-Based Recognition Cues in the Ant Leptothorax Nylanderi.” Ethology 102, no. 3 (1996): 510–22. https://doi.org/10.1111/j.1439-0310.1996.tb01143.x.

Herbers, Joan M. “Watch Your Language! Racially Loaded Metaphors in Scientific Research.” BioScience 57, no. 2 (February 1, 2007): 104–5. https://doi.org/10.1641/B570203.

Hölldobler, B., and E. O. Wilson. “Colony-Specific Territorial Pheromone in the African Weaver Ant Oecophylla Longinoda (Latreille).” Proceedings of the National Academy of Sciences 74, no. 5 (May 1, 1977): 2072–75. https://doi.org/10.1073/pnas.74.5.2072.

Hölldobler, Bert. “Territoriality in Ants.” Proceedings of the American Philosophical Society 123, no. 4 (August 30, 1979): 211–18.

———. “Tournaments and Slavery in a Desert Ant.” Science 192, no. 4242 (May 28, 1976): 912–14. https://doi.org/10.1126/science.192.4242.912.

Hölldobler, Bert, and Edward O. Wilson. Journey to the Ants: A Story of Scientific Exploration. Belknap Press, 1998. https://www.hup.harvard.edu/catalog.php?isbn=9780674485266.

———. The Ants. Harvard University Press, 1990.

Hölldobler, Bert, and Edward O Wilson. The Leafcutter Ants: Civilization by Instinct. W. W. Norton & Company, 2011.

Hölldobler, Bert, and Edward O. Wilson. The Superorganism: The Beauty, Elegance, and Strangeness of Insect Societies. W. W. Norton & Company, 2008. https://wwnorton.com/books/9780393067040.

Hölldobler, Berthold K, and Edward O Wilson. “Weaver Ants.” Scientific American, 1977, 146–54.

Hoyt, Erich. The Earth Dwellers: Adventures in the Land of Ants. Reprint edition. New York: Simon & Schuster, 1997.

Janzen, Daniel H. “Coevolution of Mutualism Between Ants and Acacias in Central America.” Evolution 20, no. 3 (1966): 249–75. https://doi.org/10.2307/2406628.

Jeanne, Robert L. “Division of Labor Is Not a Process or a Misleading Concept.” Behavioral Ecology and Sociobiology 70, no. 7 (July 1, 2016): 1109–12. https://doi.org/10.1007/s00265-016-2146-7.

Keeley, Lawrence H. War Before Civilization. Oxford University Press, USA, 1996.

Lafleur, Laurence J. “Tolerance in Ants.” The American Naturalist 76, no. 762 (January 1, 1942): 85–93. https://doi.org/10.1086/281016.

Lamon, Brent, and Howard Topoff. “Avoiding Predation by Army Ants: Defensive Behaviours of Three Ant Species of the Genus Camponotus.” Animal Behaviour 29, no. 4 (November 1, 1981): 1070–81. https://doi.org/10.1016/S0003-3472(81)80060-7.

Langen, T. A., Frédéric Tripet, and Peter Nonacs. “The Red and the Black: Habituation and the Dear-Enemy Phenomenon in Two Desert Pheidole Ants.” Behavioral Ecology and Sociobiology 48, no. 4 (September 1, 2000): 285–92. https://doi.org/10.1007/s002650000223.

Lawick-Goodall, Jane van, and David A. Hamburg. “Recent Developments in the Study of Primate Behavior.” Bulletin of the American Academy of Arts and Sciences 27, no. 7 (1974): 36–48. https://doi.org/10.2307/3823704.

Lazar, Seth. “War.” In The Stanford Encyclopedia of Philosophy, edited by Edward N. Zalta, Spring 2020. Metaphysics Research Lab, Stanford University, 2020. https://plato.stanford.edu/archives/spr2020/entries/war/.

LeBrun, Edward G., Robert M. Plowes, Patricia J. Folgarait, Martin Bollazzi, and Lawrence E. Gilbert. “Ritualized Aggressive Behavior Reveals Distinct Social Structures in Native and Introduced Range Tawny Crazy Ants.” PLOS ONE 14, no. 11 (November 22, 2019): e0225597. https://doi.org/10.1371/journal.pone.0225597.

Leonhardt, Sara Diana, Florian Menzel, Volker Nehring, and Thomas Schmitt. “Ecology and Evolution of Communication in Social Insects.” Cell 164, no. 6 (March 10, 2016): 1277–87. https://doi.org/10.1016/j.cell.2016.01.035.

Levy, Jack S. “The Causes of War and the Conditions of Peace.” Annual Review of Political Science 1, no. 1 (1998): 139–65. https://doi.org/10.1146/annurev.polisci.1.1.139.

Li, Hongjie, Chang-Yu Sun, Yihang Fang, Caitlin M. Carlson, Huifang Xu, Ana Ješovnik, Jeffrey Sosa-Calvo, et al. “Biomineral Armor in Leaf-Cutter Ants.” Nature Communications 11, no. 1 (November 24, 2020): 5792. https://doi.org/10.1038/s41467-020-19566-3.

Lieshout, E. van, M. A. Elgar, and E. van Wilgenburg. “Conflict Resolution Strategies in Meat Ants (Iridomyrmex Purpureus): Ritualised Displays versus Lethal Fighting.” Behaviour 142, no. 6 (January 1, 2005): 701–16. https://doi.org/10.1163/1568539054729150.

Lopez, Anthony C., and Dominic D.P. Johnson. “The Determinants of War in International Relations.” Journal of Economic Behavior & Organization 178 (October 2020): 983–97. https://doi.org/10.1016/j.jebo.2017.09.010.

Mabelis, A. “Aggression in Wood Ants (Formica Polyctena Foerst., Hymenoptera, Formicidae).” Aggressive Behavior 10, no. 1 (1984): 47–53. https://doi.org/10.1002/1098-2337(1984)10:1<47::AID-AB2480100107>3.0.CO;2-N.

MacMillan, Margaret. War: How Conflict Shaped Us. Allen Lane, 2020.

Martin, Kit, and Pratim Sengupta. “Multi-Agent Simulations of Intra-Colony Violence in Ants.” Springer, 2020. https://prism.ucalgary.ca/handle/1880/112121.

Maschwitz, Ulrich, and Heinz Hänel. “The Migrating Herdsman Dolichoderus (Diabolus) Cuspidatus: An Ant with a Novel Mode of Life.” Behavioral Ecology and Sociobiology 17, no. 2 (July 1, 1985): 171–84. https://doi.org/10.1007/BF00299249.

Maschwitz, Ulrich, Karla Jessen, and Eleonore Maschwitz. “Foaming in Pachycondyla: A New Defense Mechanism in Ants.” Behavioral Ecology and Sociobiology 9, no. 1 (August 1, 1981): 79–81. https://doi.org/10.1007/BF00299857.

Moffett, Mark W. “Ants & the Art of War.” Scientific American 305, no. 6 (2011): 84–89.

———. “Supercolonies of Billions in an Invasive Ant: What Is a Society?” Behavioral Ecology 23, no. 5 (September 1, 2012): 925–33. https://doi.org/10.1093/beheco/ars043.

———. The Human Swarm: How Our Societies Arise, Thrive, and Fall. Basic Books, 2019. https://www.basicbooks.com/titles/mark-w-moffett/the-human-swarm/9781541617292/.

Morris, Lyle, Michael Mazarr, Jeffrey Hornung, Stephanie Pezard, Anika Binnendijk, and Marta Kepe. Gaining Competitive Advantage in the Gray Zone: Response Options for Coercive Aggression Below the Threshold of Major War. RAND Corporation, 2019. https://doi.org/10.7249/RR2942.

Moseley, Alexander. “The Philosophy of War.” In Internet Encyclopedia of Philosophy. Accessed February 21, 2021. https://iep.utm.edu/war/.

Newey, Philip S, Simon K A Robson, and Ross H Crozier. “Weaver Ants Oecophylla Smaragdina Encounter Nasty Neighbors Rather than Dear Enemies” 91, no. 8 (2010): 7.

Nygaard, Sanne, Haofu Hu, Cai Li, Morten Schiøtt, Zhensheng Chen, Zhikai Yang, Qiaolin Xie, et al. “Reciprocal Genomic Evolution in the Ant–Fungus Agricultural Symbiosis.” Nature Communications 7, no. 1 (July 20, 2016): 12233. https://doi.org/10.1038/ncomms12233.

O’Donnell, Sean. “Ant Wars.” Serious Science (blog), August 25, 2016. http://serious-science.org/ant-wars-6652.

Parent, André. “Auguste Forel on Ants and Neurology.” Canadian Journal of Neurological Sciences / Journal Canadien Des Sciences Neurologiques 30, no. 3 (August 2003): 284–91. https://doi.org/10.1017/S0317167100002754.

Pereira, MC, ELB Firmino, RC Bernardi, IC Guimarães, CAL Cardoso, and WF Antonialli-Junior. “Dear Enemy Phenomenon in the Ant Ectatomma Brunneum (Formicidae: Ectatomminae):,” 2019, 9.

Powell, Scott. “A Comparative Perspective on the Ecology of Morphological Diversification in Complex Societies: Nesting Ecology and Soldier Evolution in the Turtle Ants.” Behavioral Ecology and Sociobiology 70, no. 7 (July 1, 2016): 1075–85. https://doi.org/10.1007/s00265-016-2080-8.

Powell, Scott, and E. Clark. “Combat between Large Derived Societies: A Subterranean Army Ant Established as a Predator of Mature Leaf-Cutting Ant Colonies.” Insectes Sociaux 51 (January 11, 2004): 342–51. https://doi.org/10.1007/s00040-004-0752-2.

Powell, Scott, Shauna L. Price, and Daniel J. C. Kronauer. “Trait Evolution Is Reversible, Repeatable, and Decoupled in the Soldier Caste of Turtle Ants.” Proceedings of the National Academy of Sciences 117, no. 12 (March 24, 2020): 6608–15. https://doi.org/10.1073/pnas.1913750117.

Reeve, H. Kern, and Bert Hölldobler. “The Emergence of a Superorganism through Intergroup Competition.” Proceedings of the National Academy of Sciences 104, no. 23 (June 5, 2007): 9736–40. https://doi.org/10.1073/pnas.0703466104.

Regnier, F. E., and E. O. Wilson. “Chemical Communication and ‘Propaganda’ in Slave-Maker Ants.” Science 172, no. 3980 (April 16, 1971): 267–69. https://doi.org/10.1126/science.172.3980.267.

Rudolph, Kathleen P., and Jay P. McEntee. “Spoils of War and Peace: Enemy Adoption and Queen-Right Colony Fusion Follow Costly Intraspecific Conflict in Acacia Ants.” Behavioral Ecology 27, no. 3 (January 1, 2016): 793–802. https://doi.org/10.1093/beheco/arv219.

Schadlow, Nadia. “Research & Debate—It’s a Gray, Gray World.” Naval War College Review 73, no. 3 (2020): 7.

———. “War and the Art of Governance.” The US Army War College Quarterly: Parameters 33, no. 3 (August 1, 2003). https://doi.org/10.55540/0031-1723.2168.

Schelling, Thomas C. “An Astonishing Sixty Years: The Legacy of Hiroshima (Nobel Prize Speech).” Nobel Foundation, 2005. https://onlinelibrary.wiley.com/doi/abs/10.1111/nejo.12236.

Schneirla, Theodore Christian, and T. C. Schneirla. Army Ants: A Study in Social Organization. W. H. Freeman, 1971.

Schultz, Ted, R., Ulrich G. Mueller, Cameron R. Currie, and Stphen A. Rehner. “Reciprocal Illumination: A Comparison of Agriculture in Humans and Fungus-Growing Ants.” In Insect-Fungal Associations: Ecology and Evolution, 149–90. Oxford University Press, 2005. http://www.sbs.utexas.edu/Muelleru/pubs/SchultzEtAl2005RecipIllum.pdf.

Smith, John Maynard. Evolution and the Theory of Games. Cambridge: Cambridge University Press, 1982.

Sorrells, Trevor R., Leah Y. Kuritzky, Peter G. Kauhanen, Katherine Fitzgerald, Shelby J. Sturgis, Jimmy Chen, Cheri A. Dijamco, Kimberly N. Basurto, and Deborah M. Gordon. “Chemical Defense by the Native Winter Ant (Prenolepis Imparis) against the Invasive Argentine Ant (Linepithema Humile).” PLOS ONE 6, no. 4 (April 19, 2011): e18717. https://doi.org/10.1371/journal.pone.0018717.

Souza, Jorge L P, and Carlos A R Moura. “Predation of Ants and Termites by Army Ants, Nomamyrmex Esenbeckii (Formicidae, Ecitoninae) in the Brazilian Amazon” 52, no. 2 (2008): 5.

Steiner, Florian. “How Many Ant Species Are There on Earth?” Myrmecological News Blog (blog), June 15, 2018. https://blog.myrmecologicalnews.org/2018/06/15/how-many-ant-species-are-there-on-earth/.

Stoker, Donald, and Craig Whiteside. “Blurred Lines: Gray-Zone Conflict and Hybrid War—Two Failures of American Strategic Thinking.” Naval War College Review 73, no. 1 (2020): 38.

Sturgis, Shelby J, and Deborah M Gordon. “Nestmate Recognition in Ants (Hymenoptera: Formicidae): A Review.” Myrmecological News 16 (January 2012): 101–10.

Svetlana Alexievich. The Unwomanly Face of War. Penguin Random House, 2018. https://www.penguinrandomhouse.com/books/540744/the-unwomanly-face-of-war-by-svetlana-alexievich/.

Thomas, Melissa L., Christine M. Payne-Makrisâ, Andrew V. Suarez, Neil D. Tsutsui, and David A. Holway. “When Supercolonies Collide: Territorial Aggression in an Invasive and Unicolonial Social Insect.” Molecular Ecology 15, no. 14 (November 14, 2006): 4303–15. https://doi.org/10.1111/j.1365-294X.2006.03038.x.

Thoreau, Henry David. “Chapter 12: Brute Neighbors.” In Walden; or, Life in the Woods. Boston: Ticknor and Fields, 1854. https://americanliterature.com/author/henry-david-thoreau/essay/the-battle-of-the-ants.

Torres, Candice, M. Brandt, and Neil Tsutsui. “The Role of Cuticular Hydrocarbons as Chemical Cues for Nestmate Recognition in the Invasive Argentine Ant ( Linepithema Humile ).” Insectes Sociaux 54 (January 11, 2007): 363–73. https://doi.org/10.1007/s00040-007-0954-5.

Van Wilgenburg, Ellen, Candice W Torres, and Neil D Tsutsui. “The Global Expansion of a Single Ant Supercolony.” Evolutionary Applications 3, no. 2 (March 2010): 136–43. https://doi.org/10.1111/j.1752-4571.2009.00114.x.

Wheeler, William Morton. Ants: Their Structure, Development, and Behavior. The Columbia University Press, 1910.

———. “Ecological Relations of Ponerine and Other Ants to Termites.” Proceedings of the American Academy of Arts and Sciences 71, no. 3 (1936): 159–243. https://doi.org/10.2307/20023221.

“Wheeler, William Morton (1865-1937) - AntWiki.” Accessed January 31, 2021. https://www.antwiki.org/wiki/Wheeler,_William_Morton_(1865-1937).

Whitehouse, Mary E. A., and Klaus Jaffe. “Ant Wars: Combat Strategies, Territory and Nest Defence in the Leaf-Cutting Ant Atta Laevigata.” Animal Behaviour 51, no. 6 (June 1, 1996): 1207–17. https://doi.org/10.1006/anbe.1996.0126.

Wills, Bill D., Scott Powell, Michael D. Rivera, and Andrew V. Suarez. “Correlates and Consequences of Worker Polymorphism in Ants.” Annual Review of Entomology 63, no. 1 (January 7, 2018): 575–98. https://doi.org/10.1146/annurev-ento-020117-043357.

Wilson, Edward O. “Slavery in Ants.” Scientific American 232, no. 6 (1975): 32–40.

Wright, Ronald. A Short History of Progress. Toronto: House of Anansi Press Limited, 2004. https://houseofanansi.com/products/a-short-history-of-progress.

Acknowledgements

I would like to thank the editors for their invitation and guidance, and the other contributors who took part in the Toronto workshop for their inspiration and expertise. I acknowledge helpful conversations with John Slee and Lynne Supeene.

Bibliography

Essential texts concerning the mechanisms and consequences of sorting.

• Geoffrey C. Bowker and Susan Leigh Star, Sorting Things Out: Classification and Its Consequences (The MIT Press, 1999)

• Wendy Nelson Espeland and Michael Sauder, Engines of Anxiety: Academic Rankings, Reputation, and Accountability (Russell Sage Foundation, 2016)

• Bernard E. Harcourt, Against Prediction (University of Chicago Press, 2006)

• Jane Jacobs, The Death and Life of Great American Cities (New York: Random House, 1961)

• Donald MacKenzie, An Engine, Not a Camera: How Financial Models Shape Markets (MIT Press, 2007).

• James C. Scott, Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed (Yale University Press, 1998)

• Thomas C. Schelling, Micromotives and Macrobehavior (W.W. Norton and Company, 1978)

2 Workarounds

An adversarial attack is an input crafted intentionally to force a model to make a mistake. A workaround is the opposite: an input crafted intentionally to force a model to correct a mistake.

A schematic example is shown below. The data points are shown with their “true” values of blue or orange, and the background shade marks the regions categorized by the model as “blue” or “orange”. The model has a high accuracy: most of the actually-blue dots lie in the categorized-as-blue region, and most of the actually-orange points lie in the categorized-as-orange region. But even the best machine learning systems are statistical engines with non-zero error rates, so you can see a few incorrectly classified examples in the diagram too.

The adversarial example (red arrow) changes the input for a correctly-classified “orange” data point so that it is incorrectly classified as blue. The workaround (green arrow) changes the input for an “orange” data point that is incorrectly classified as “blue” so that it is classified correctly.

Adversarial examples may be novel, but workaroudns have become routine in the automated workplace. Automation shifts decision-making from “front line” workers to computerized systems, but as these systems have entered the workplace, many professionals have learned a new and often unacknowledged skill: they have become experts at making automated decision systems work properly.

As just one example, legal scholar Jennifer Raso explored the changing jobs of Ontario welfare case workers when an automated decision system was introduced in a prize-winning paper. Instead of making their own judgments about client needs, the role of case workers is now to enter data into an algorithmic scoring system (not deep learning), which makes entitlement decisions according to well-defined rules. She writes about how social service case workers responded:

While new technologies may attempt to deskill and decentre front-line decision-makers, transforming them into data entry clerks, caseworkers learn how to expertly translate and input client data to produce decisions that more closely match their interpretation of clients’ needs and welfare laws.

The importance of workarounds will be familiar to anyone who has read James C. Scott’s Seeing Like a State. Heavily designed schemes to improve the human condition assume, and impose, a simplified and idealized representation of working and living that neglects the unruly irregularities of real life. These systems rely on a certain level of workaround to function: people prepared to bend the official rules in order to get their work done. There is a reason that the “work to rule” has been successful in labour disputes as a form of protest.

The American healthcare system is no exception. The rigidities of automated billing systems recently produced a memorable essay by Atul Gawande entitled Why Doctors Hate Their Computers, documenting many of the challenges experienced by physicians as new decision support systems have been introduced, and the ingenious but often costly workarounds they have had to undertake in order to deliver high quality care to their patients.

Will workarounds be needed for the next generation of deep learning medical systems? History suggests that they will. There is a long history of technical innovations that are designed around idealized and simplified models of behaviour, and which underestimate the complexities of “edge cases” in real life. Witness the over-enthusiasm around self-driving cars a couple of years ago, that is now being confronted with the messiness of reality.

From a researcher’s point of view, adversarial examples are fascinating cases that raise a multitude of new questions: so much so that several papers per day are now published on the topic. Workarounds, in contrast, are boring. They raise few questions. They are merely machine learning residue—what’s left over after the algorithm has made its best attempt, hopefully cleared up by the next iteration of the model.

From an end user’s point of view, things look different. Faced with getting their job done and a decision system that seems to have taken a bad decision, they are not in a position to improve the model: they can either follow often a maze of twisty passages in an attempt to get an exception or they can fudge the input and move on. Workarounds are far more interesting than adversarial examples, for all their exoticism.

3 Two sides of the same coin

The previous section presented adversarial attacks and workarounds as opposites, whcih holds true in a laboratory environment where the “correct” result for each individual data point is known and particularly in the classic cases where the “correct” result is, to a human, obvious and uncontroversial. In real-world environments things are different. Machine learning systems are statistical beasts, and however accurate they may be over a large data set, statistics cannot tell us whether any one individual data point is handled correctly or not. In many cases the “true” value for an individual case is not available at all. This difference between the world of the laboratory and that of production environments is noted by the Harvard / MIT researchers:

[T]he ground truth in medical diagnoses is often ambiguous, meaning that for many cases no individual human can definitively assign the true label between, say, “benign” and “cancerous” on a photograph of a mole.

The figure below is the same as the previous one, but now the two individual data points are coloured grey, showing that we do not know their “true” value. All of a sudden, instead of being opposites, there is no difference between the adversarial example and the workaround.

(The diagram is misleading in an important way: in a two-dimensional space only a few of the dots are close to a boundary, and that boundarey is with just one other class. It would be reasonable to think that the problem is irrelevant for most inputs that lie comfortably within one region or another. In the many-dimensional space in which machine-learning models operate, it turns out that all dots are close to boundaries with all other classes: the situation for machine learning is therefore much more severe than the diagram suggests.)

Surprisingly, machine learning gives no definition to distinguish unambiguously between an adversarial example and a workaround. A standard definition in papers is something like “an input to a machine learning model that an attacker has intentionally designed to cause the model to make a mistake”, but assigning “intent” is outside the scope of machine learning, and a “mistake” is, as we have seen, often not decidable in any individual case.

Some machine learning practitioners have gone so far as to define “attacks” as any attempt by a subject to influence the outcome of a machine learning system, but that makes an attacker of anyone who has put effort into writing an appealing resume in the hopes of getting a job. Adversarial examples and workarounds are both just optimization attempts: intentional changes to inputs in order to generate desired outputs.

But don’t take all this from me: if you want to hear an expert opinion on the definition problem, see this keynote talk by David Evans (especially the first ten minutes) from which much of the last few paragraphs is taken.

When it comes to medicine, physicians face incentives to engage in workarounds as well as in adversarial attacks. The Science article is not completely silent on the topic of workarounds: it references a survey carried out by the American Medical Association from almost 20 years ago, described in a titled Physician Manipulation of Reimbursement Rules for Patients: Between a Rock and a Hard Place, but the remainder of the Science article focuses almost exclusively on the problem of adversarial attacks by healthcare providers.

The “physician manipulation” article outlines the frustrations physicians face when trying to provide care for their patients:

In some cases, physicians may feel trapped between professional obligations to advocate for their patients and conflicting contractual obligations to follow coverage rules. It has been suggested that some insurers are “gaming” patients and physicians—tricking them into paying for covered services by routinely denying coverage but then approving services that are subsequently appealed, knowing that time and other constraints will prevent some appeals.

Any attempt to tackle the threat of adversarial attacks will also penalize workarounds: designing proper rules for machine learning systems starts with recognizing the problems, and an exclusive focus on adversarial attacks fails to do so.

4 Governing machine learning

When it comes to interventions, the Harvard / MIT researchers make two recommendations. The first is to procrastinate: to avoid stifling innovation by prematurely enforcing demands for robustness. The second is to increase IT-driven supervision of medical practitioners, storing “hashes” of images to check that they are not “tampered” with during processing.

To take the second suggestion first: the proposal will crack down on “adversarial attacks” but will also prevent workarounds. As such, it removes room for doctors’ discretion and judgement. Good physician performance will be measured by accurate data entry rather than creative problem solving, and such measures would prevent physician pushback against any payment-minimizing moves by insurance companies. It reduces the role of physician from a professional responsible for diagnosis and prescription (which most would see as their primary roles), to that of a managed and supervised data entry technician. Any assessment of medical machine learning must address the potential consequences of deskilling among physicians.

The billing system, whether or not it is driven by deep learning, is a site of tension, at which all parties seek to find their best outcomes. To treat the problems around billing systems as security problems in which the healthcare provider is the most likely attacker is to adopt the perspective of the insurance company or the vendor that provides them with a deep learning system.

In fairness, the authors do point out that the American medical claims approval process is plagued by many competing financial interests, “providers [i.e. doctors and hospitals] seeking to maximize and payers [insurance companies] seeking to minimize reinbursement”, but for the bulk of the article, and in the recommendations, the problem of “attacks” by insurance companies is neglected.

A more productive approach would be to adopt a game-theoretical perspective, where strategic considerations by all parties lead to some equilibrium outcome. As I’ve written elsewhere, most machine learning systems of any interest are incentive incompatible, and medical systems are no different. The subjects who provide the inputs and the consumers of the outputs have different and conflicting interests.

In such an arrangement costly governance measures are not just likely, but inevitable, and that leads us to what’s wrong with the first recommendation: procrastination. To encourage deployment without proper costing of governance is to underestimate implementation costs: it’s building in debt, committing institutions that deploy these systems to unknown but significant future costs, and painting a false picture of efficiency gains. It tilts the cost comparison between physician-diagnosis and machine-diagnosis in favour of the machine (or, more accurately, its vendor), and more-or-less guarantees buyer’s remorse.

There is no one-size fits all solution, of course. The right form of governance for EHR-based methods would be different to that for medical image diagnostics. But a prerequisite for any progress must be to acknowledge all the incentives in the system, to attempt to estimate the true costs of implementation, and not to overlook the humble but essential role of the workaround.

2 2008: Beginnings

Let’s look at three companies which each started just about ten years ago and which had a similar business model in common, even though the people involved did not know each other at all. There was no name for it at the time, but there was something in the air…

Let’s start with Airbnb, which you will probably know is the controversial short-term rental or holiday accommodation company, where people rent out residential homes for use by tourists and other visitors.

In 2007, newly graduated design students Brian Chesky and Joe Gebbia were looking for ways to make their end-of-month rental payments in San Francisco when an industry conference came to town. They bought some inflatable airbeds and offered accommodations to attendees looking for a cheap place to stay. They were inundated with requests and realized there may be a market for this kind of thing, and so “Airbed & Breakfast” was born. ¹

Since then, the story has been one of hard work and growth. Running up the limit on multiple credit cards to finance the very beginnings, they got an early investment from Paul Graham’s Y-Combinator fund. Struggling to get the site to take off, they went out to their biggest city (New York) and got the hosts to have professional photos taken of their rooms to make them more appealing; the bookings increased, and professional photography continues to be the most effective way for a host to attract guests. Other maneuvers included a breakfast-cereal pitch around the Democratic convention in Chicago and a widely criticized email campaign via Craigslist.

After the early tribulations, the now-trio of Chesky, Gebbia, and Nathan Blecharczyk found themselves at the head of a phenomenon. In 2011 Airbnb had 50,000 listings; in 2012 that number more than doubled to 120,000; by the end of 2013 it was 550,000; in 2018 it is nearly 5 million listings. ² For comparison, the largest hotel company in the world by number of rooms is the InterContinental Group with 775,000 rooms. ³

Just as Airbnb was getting started, two entrepreneurs were attending a conference in Paris, and had a hard time finding a cab. These two were Travis Kalanick, who was to go on to be the controversial CEO of Uber, and Garrett Camp. They didn’t need to worry about paying rent, as both were already rich from earlier technology ventures. Wouldn’t it be great, they thought, if you could get a cab just by pressing a button on the newly released iPhone.

In its early days, Uber worked with limo companies to provide drivers, but then they realized that a competitor named Lyft was using amateur drivers rather than professionals, and Uber did not want to be left out. So they set up Uber X, which is Uber Pop in most of Europe, and opened up their business to independent drivers.

Uber’s growth has been even more dramatic than that of Airbnb. The company has raised investment money like no other, raising tens of billions of dollars in investment.

Our third example is less well known now, but it was often talked about as another archetype of the Sharing Economy. Here is how the company’s web site described the founding story.

It was a cold night in Boston in February of 2008 when Leah Busque realized she was out of dog food for her 100-lb yellow lab, Kobe. Leah thought to herself, “Wouldn’t it be nice if there was a place online I could go to connect with my neighbors—maybe one who was already at the store at that very moment—who could help me out?”

From this experience, TaskRabbit (formerly RUNmyERRAND), an online and mobile marketplace that connects neighbors to get things done, was born . . . Neighbors helping neighbors—it’s an old school concept reimagined for today. ⁴

One of the typical tasks that people helped with was assembling Ikea furniture. TaskRabbit had early successes and attracted a lot of attention, but in the end it did not become the kind of global company that Airbnb and Uber have become. Instead, after several years as an independent company, it sold itself to Ikea.

Most people’s problems don’t turn into multi-billion dollar companies. Timing --- whether by luck or by skill – was essential to their success. Here is a set of things that had been put in place in the few years before 2008, each of which was needed for Airbnb and Uber to take off:

• Broad-band internet. Many homes had computers, and many people were increasingly comfortable with the Internet as a part of their daily lives.

• Online payments. People were also getting used to using credit cards on the Internet to pay for their purchases at Amazon, eBay, Expedia and other early successes.

• Mobile computing. Beyond that, we were entering the era of the smartphone. The iPhone was released in 2007 and Uber, for one, could not have been invented before the iPhone.

• Social media. In the early years, pseudonyms and anonymity were the norm on the Internet. As the famous cartoon caption went: “On the Internet, nobody knows you’re a dog”. But starting with Facebook in 2006, many people were getting used to the idea of using real names. TaskRabbit and Uber’s competitor Lyft both hooked into Facebook’s user identification system as a part of their business model, and Airbnb and others used Facebook accounts as proof of identity.

• Rating systems. The use of ratings, particularly for Netflix movies and Amazon books, but also through sites such as Yelp and Travelocity, had spread and the five-star rating that is now ubiquitous was becoming common.

• Cloud computing. Starting in 2006 Amazon started renting out computer usage from its own massive data centres at bargain prices, starting the era of cloud computing. This meant that new companies could start up without buying expensive hardware: they could use Amazon for their computing needs and get started much more cheaply.

• Crowdsourcing. The basic computing architecture of the Sharing Economy is the “platform” where the company itself runs the software, but those who provide the content come from outside. The success of Wikipedia showed this could be done, and eBay showed you could build a valuable company that way. YouTube was and continues to be a remarkable example of this crowdsourced structure.

Even though these were all present, the Sharing Economy brought with it the shock of the new --- pressing a button on your phone and seeing a car appear seemed like magic, and booking to stay in somebody’s house half way around the world opened up new possibilities for travel. The novelty of the experiences gave the new services an aura of cool, and many people immediately thought: this is the future.

3 2010 – 2015: Rise

Airbnb and Uber were not founded as Sharing Economy companies, because the name did not exist. In the years 2010 to 2012 a set of other companies were founded using a similar business model and commentators looked for names to describe this trend. “Collaborative economy”¹, “circular economy”, ⁵ and “the mesh”⁶ were all put forward, but in the end the Sharing Economy won.

If there is a one-sentence description of the Sharing Economy, it may be Internet platforms for real-world services. The first few years of the century had seen internet platforms for retail (Amazon and eBay), for social media exchanges and advertising (Google and Facebook), and for media (Netflix, Apple, YouTube). Of particular interest were eBay and YouTube: crowdsourced platforms where the content seemed to appear by magic. The Sharing Economy takes this model, but instead of uploading videos the platform is used for people who provide real-world services.

How far can crowdsourced Internet platforms go? Can their economic benefits be applied to the service economy too? That’s what the Sharing Economy phenomenon asks us.

A 2017 report for the World Economic Forum gives a three-point definition.⁷ First is “The use of digital technologies to match buyers and sellers,” which is similar to the point above. But they add two other components. One is “Capitalizing on idle capacity”: the service providers are generally offering something they already own, but which they are not using right now. So the service providers are not employees of the platform owners, they are independent actors who are making use of their own assets. To go back to our three stories: perhaps they have a car, and can drive for Uber; perhaps they have some spare time and can run some errands for TaskRabbit; perhaps they have a spare room and can put it on Airbnb. The other additional point is “trust verification”, to which I’ll return.

With this framework, let’s look at some examples of companies that started up in those years of 2010 to 2012. This was the era when seemingly a million start-up plans were pitched as the “Uber for this” or the “Airbnb for that”. But I want to start with something that is the opposite. The Toronto Tool Library⁸ is a non-profit organization that lets people rent tools for home projects. The organization is geographically limited and community focused: it has no plans to drive major stores out of business or to become a global behemoth. But it matters to the story of the Sharing Economy because it draws on one of the early images that promoted the idea, which is the story of the power drill.⁹ Many of us have a hand drill that sits on a shelf and is hardly ever used and Rachel Botsman, who was one of the early promoters of the Sharing Economy and helped to articulate it as a new development, says: we don’t really want a drill, what we want is the hole it makes. So why not share a drill instead of owning it? Sharing is sustainable, it helps us get off the consumerist track of ownership, weighed down by our goods. The drill was the embodiment of using “idle capacity”. It has an appealing sustainability angle to it and also hints to a concept called “access over ownership”: we don’t have to buy in to the bourgeois and consumerist culture of ownership, we can tread more lightly on the earth by dipping in and out of these sharing arrangements instead. It was a catchy and attractive image.

In 2013 Airbnb was already taking off, and a natural extension of that idea was to share meals. Airbnb said “you don’t need hotels, you just need a platform where you can connect with other people.” Cookening, Feastly and other sites applied the same logic to restaurants. If you are travelling to a new city, why not eat like a local in their own house? Again, the appeal here is to authenticity, to personal connection rather than institutions and companies, and to the unusual combination of small and big that crowdsourcing requires: many small individual exchanges building a global platform.

Tools and meals are fairly low risk, banged thumbs and food poisoning aside, but finance is surely something different. Remember that we are talking about a period right after the financial crisis, when personal finances were stretched and banks were viewed with suspicion. So companies like Lending Club took a lead from charities such as Kiva, which focused on direct person-to-person donations, as well as from initiatives like the Grameen bank, which used social lending to pioneer the idea of microfinance, and whose founder Mohammed Yunus won the Nobel Peace Prize in 2006. The idea behind peer-to-peer lending was that individuals could lend to each other, combining social purpose and sound investment.

Here the question of trust is at its most severe. What happens if I take a loan from you and don’t pay it back? How do you assess credit? The early players in this industry like Lending Club and Prosper were experimenting with a variety of approaches, and we’ll see how that worked out later on.

And to finish, let’s go back to household services like TaskRabbit. A lot of companies started in this area. One is Homejoy, which was Google’s venture into housecleaning – an eyebrow-raising idea in itself. Another was Handy, which looked to broaden from cleaning to other services like plumbing and carpentry.

And there were lots of others. A short list would include DogVacay, which was often written about to show the quirky and personal aspect of the market. Instacart and Deliveroo highlighted the food delivery area, which is still going strong. BlaBlaCar is a French service for taking rides from city to city. ZipCar was important as one of the first carsharing companies, and it preceded most of the others here.

In 2013 to 2015 various lists of Sharing Economy companies were published, and they often had hundreds of examples. But for some analysts, this was just the beginning. PwC is one of the most consistently upbeat forecasting agencies around the Sharing Economy, and have forecast markets sizes well into the hundreds of billions of dollars in a few years time. It looks like the Sharing Economy is taking off.

With those specific examples in mind, let’s take one more look at a definition.

First, we have the platform, and that’s the Sharing Economy company. Perhaps it is Homejoy, or DogVacay, or Lending Club. It matches service providers to consumers. It doesn’t take on providing the service itself, but it does handle payments, identity verification, and ratings. Rating systems were often called out as one of the main innovations of the Sharing Economy. New York Times columnist Thomas Friedman lauds “Airbnb’s real innovation—a platform of ‘trust’—where everyone could not only see everyone else’s identity but also rate them as good, bad or indifferent hosts or guests. This meant everyone using the system would pretty quickly develop a relevant ‘reputation’ visible to everyone else in the system.” ¹⁰ Friedman was writing just a couple of weeks after his New York Times stablemate David Brooks described “How Airbnb and Lyft Finally Got Americans to Trust Each Other”: “Companies like Airbnb establish trust through ratings mechanisms . . . People in the Airbnb economy don’t have the option of trusting each other on the basis of institutional affiliations, so they do it on the basis of online signaling and peer evaluations.” ¹¹

On the consumer side we have seen that the Sharing Economy fits into and help to shape a trend of particularly young people as valuing experiences over ownership, and choosing access over ownership. Your books and music may already be on devices: why buy a car when you can take an Uber? Everything you need is on your phone or your computer.

And then there are the service providers, who are presented as partners rather than employees. The word that used early on was “micro-entrepreneurs”: not really companies, not really private individuals, but somewhere in between. And by making the most of their unused assets they were being green, building community, and making a little money.

4 Business models and strategy

Not all sharing economy business models are novel, or successful, but when it comes to strategy there are some distinctive features for this sector.

The appeal of platforms to investors lies in the idea of increasing returns to scale. There is a popular phrase that you may have seen. The biggest hotel company -- Airbnb -- owns no property. The biggest taxi company -- Uber -- owns no cars. This mimics the way that Facebook and YouTube own no content. Each is as big as the biggest company in their market. The value, it seems, is in the platform.

One of the benefits of this model is that if a company sees demand, it can grow very quickly and globally. Just as other Internet companies have spread around the world, so can Airbnb and Uber. If a hotel chain wants to add another 100 rooms it has to build a hotel, but for Airbnb it’s just another row in a database. So on the supply side there are benefits here, sometimes labelled as a “zero marginal cost” business structure.

There are increasing returns on the demand side as well as on the supply side, and they go under the name network effects. Renters want to be where travellers are, and travellers want to be where renters are. So just like social media, every person who takes a trip on Airbnb, and every person who adds a listing, makes it more attractive for everyone else.

One of the things about increasing returns to scale is that it makes it difficult for new companies to enter the market: it favours a natural monopoly, or at least oligopoly. Facebook, Amazon, Netflix have few real competitors. And this industrial organization makes profits possible.

Zero marginal costs and network effects: put the two together and you can see why investors are enthusiastic. Can Airbnb and Uber do for services what Amazon has done for retail, what Wikipedia did for the encyclopedia, and Apple and Spotify have done for music? Is the market in these services a winner-take-all? The large scale of venture capital investment shows that a lot of people are prepared to bet that it is.

There is a real pressure among startups to grow quickly, and one of the strategies that gained prominence early was a piece of advice given to Airbnb by their early investor Y Combinator: do things that don’t scale. ¹²

When a company starts, it doesn’t have the benefit of network effects so it can’t rely on them. Instead it needs something different to set itself apart and get growth started. For Airbnb that was visiting hosts in its first significant market, and helping them present their listings in an attractive way on the site. Paying for a photographer was a big step forward: it made a big difference to those early listings and got the idea of homesharing started.

But of course in business everyone wants a winning formula, and applying it blindly is rarely successful. Another Sharing Economy company, Washio, set out to do the same as Airbnb but for laundry: the company was going to “demolish laundry”. ¹³ They gave each customer cookies, and packaged the laundry with sleek black bags. But it turned out that the ideas of the Sharing Economy don’t really work for laundry. To start with the cost side: most of the cost is involved in actually doing the laundry and digital technology doesn’t help with that. On the demand side: just because other people are doing their laundry through Washio doesn’t make a big difference to me, so network effects were limited. And because of that, the strategy was just losing money on every sale. Washio ceased operations in 2016.

There is another aspect of business strategy that was central to the growth of the Sharing Economy. I’ve touched on it a couple of times. Many of these companies did not just see themselves as businesses, they were also a movement: brands with a purpose.

The companies fostered a community of so-called micro-entrepreneurs and leaned heavily on ideas of democratization. Instead of relying on faceless corporations we can now provide services to each other, in a “peer to peer” fashion. You don’t need a bank to get a loan, you don’t need a hotel to take a holiday, you don’t need a taxi to get across town, you just go to the Internet and click or tap. The idea appeals to those already invested in new technologies: the old rules don’t apply, we have a better way now. Airbnb CEO Brian Chesky said that most of the reasons for regulation in the accommodation and transport spaces were consumer protection, and reputation systems could carry out that task better than cities. ¹⁴

So the Sharing Economy shared a moment with the growth of Wikipedia, with the Arab Spring uprisings in 2011 and the idea of Facebook revolutions and social media as a democratizing force in politics. The internet was no longer this place you went to when you sat at a desk, it was moving into the real world, and taking its values with it.

But from the beginning there were tensions built into the Sharing Economy, and the name captures the problem. Is this about a community or a business?

On the one hand it’s about sharing: it’s social, it’s community-focused, and personal. On the other hand, it’s an economy: it’s about market transactions and prices, where scale is the key to profit and perhaps to monopoly rents.

The tensions were captured in an organization set up in 2013 called Peers. It called itself a grassroots movement set up to clear out old-fashioned regulations and clear a path for the new sharing economy companies. But it was founded by Airbnb, whose goal was to make money.

In the same way, the companies were touted as democratizing, but were set up with the goal of making money for investors, including some of the richest people on the planet.

It’s an attitude that has been called the Californian Ideology, ¹⁵ which is that you can have both. You can be rebellious, anti-establishment, and against hierarchy; but you can also make a big pile of money while doing it. You can be ethical, and live a life of purpose, and you can be rich beyond your wildest dreams.

But some early examples showed that you cannot have everything, and when it comes to choosing between community and profit, commercial enterprises will choose profit.

5 Fall?

I hope that gives an idea of the ideas behind the Sharing Economy. It’s been ten years now, and there has been a perhaps-inevitable shakeout.

For many of the “Uber for X” companies, like Washio, the promise of new efficiencies has proved hard to achieve. The unpleasant surprise was perhaps told best by Sarah Kessler, who went back to look at the story of the drill. ¹⁶ She showed that many of those companies who were looking to follow the Sharing Economy template were not growing fast. Going across town to get a drill from someone you don’t know when you can buy one for $50 was too much of a problem.

We have already seen that Washio went bust because its business model did not reflect reality. Laundry, it turns out, is laundry. In the same way, the cleaning company Homejoy was set up with $30 million from Google Ventures, but any efficiency it gained came at the cost of its cleaners, who were reported in one revealing article as living under bridges. ¹⁷ The cost of cleaning is what you pay the cleaners. The Internet only helps with that if you can push costs, like the time to travel, onto the cleaners. And obviously they did not want to bear that cost, so Homejoy had a hard time keeping staff.

Adding to the problem is that once a homeowner has found a good cleaner, they probably want the same one again, so the matching problem is not a huge benefit, and if the cleaner is really a micro-entrepreneur, then the homeowner can just make a deal with them directly and cut out Homejoy’s portion of the cost. Homejoy went out of business a few years to.

Other companies had to carry out what technology companies call a “pivot”: change their business model to something different. TaskRabbit has sold itself to Ikea, and Instacart has partnered with grocery chains, so both are now a source of cheap labour for orthodox businesses rather than a peer-to-peer alternative. Perhaps the biggest pivot was in the “peer-to-peer lending” market. The companies struggled with high rates of default, and ended up changing the name to what they call “marketplace lending”. Most of the loans for these companies now come from regular financial institutions, rather than individuals. The role of the platforms has changed to be that of an automated loan officer: algorithmically assessing the credit-worthiness of applicants, and supplying those assessments to the financial institutions who can decide whether or not to make the loan.

What does this tell us? For many companies, the Internet has not been the game changer they thought. For others, the Internet has worked, but they have had to move from the Sharing Economy model. And that leaves us focusing again on the survivors.

6 2018: Full circle?

In 2018 it seems like we are nearly back to where we started: left with two successful industries– ridesharing and short-term rentals. There is Airbnb and there is Uber.

But while the overall phenomenon of the Sharing Economy may be fading, these two are still growing rapidly. A look at their success is still valuable because of their size and impact, and also because some of the aspects of their business is a first look at the next wave of innovation, in which the service providers themselves are algorithms instead of people. The world of autonomous vehicles and AI services will be platform-based businesses that face many of the same challenges as the Sharing Economy.

It seems like Airbnb and Uber are two cases where the Sharing Economy model worked: zero marginal costs and network effects combining to give runaway success. But for Uber at least, it may be too early to say that.

Uber is often described as a company like Amazon, or Google, or Apple: one that has changed the nature of the world in which we live. It is easy to forget that Uber is still not a profitable company, and as financial quarter after quarter goes by with large losses (over $1 Billion in 2018 Q3 ¹⁸), it is becoming increasingly doubtful whether its business model is sustainable. ¹⁹

Uber’s challenge is to keep both drivers and riders happy. If it raises prices it loses riders and if it keeps them low then driver retention becomes a challenge. Much of the early enthusiasm for the company assumed that this balancing act was solvable because of the huge efficiency increases that Uber was bringing to the industry, helping riders and drivers find each other and eliminating idle time for drivers. The enthusiasm peaked with a 2014 report in the Washington Post suggesting that some Uber drivers were earning $90,000 a year, ²⁰ but that story quickly vanished and now the stories are far more often about drivers earning less than minimum wage, or facing financial difficulties because they are unable to keep up payments on a leased car. ²¹

So far Uber has solved its problem by throwing money at it. Uber subsidises rides by providing enticements to the drivers, burning through billions of dollars of capital while looking with increasing desperation for a profitable business model.

Recent reports from New York City suggest that Uber and Lyft cars are empty for at least as much of the time as the traditional “yellow cabs”.²² This is not the case everywhere, but it suggests that efficiency in its current market is not likely to drive profitability. Maybe squeezing more out of drivers will be a way to do it? Uber is investing in dynamic pricing, driver incentives, and a myriad of other techniques for taking more out of every rider’s dollar ²³, but we should not be too quick to assume they will succeed. It remains possible that Uber is a blip.

Meanwhile, Airbnb is profitable, but that is a low bar to reach for a company valued at $31 Billion, whose repeated attempts to break into “whole trip” planning and business travel have been unsuccessful.

Still, both these companies have shown that rapid and global growth is possible. They have done so at the cost of continual controversy in cities around the world where they have become popular. There are many different controversies, but many have a common origin. We highlighted the need for Sharing Economy companies to have a “zero marginal cost” model, but that model relies on the company itself not having significant new expenses for each service provider or each transaction. And it has become clear that “zero marginal cost” is not a feature of the technology alone, it also depends on the laws that govern the company’s actions.

Sharing Economy companies present themselves in their terms and conditions as a marketplace. Uber’s user agreement, for example, says “Your ability to obtain transportation… through the use of The Services does not establish Uber as a provider of transportation, logistics or delivery services or as a transportation carrier”. The assertion is that the exchange itself, and anything that may go wrong with it, is between the consumer and the service provider, not the consumer and Uber. Other sharing economy companies present themselves in the same way. Financially, this gets them off the hook for liability and other problems, and they have relied on a particular American law to sustain this position. Section 230 of the “Communications Decency Act” includes this phrase:

No provider… of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider.

These sentences were written with web sites like YouTube in mind. The hosting platform itself was protected from liability for the actions of the users on the site. If someone uploads illegal videos to YouTube, YouTube is at least partially protected from responsibility.

Whether Section 230 to the Sharing Economy is still in dispute. The European Court of Justice recently ruled that Uber may not qualify for this kind of protection:

The service provided by Uber is more than an intermediation service… The ride-booking company must be classified as ‘a service in the field of transport.’ ²⁴

In 2016 a London tribunal judge was even more direct:

The notion that Uber in London is a mosaic of 30,000 small businesses linked by a common ‘platform’ is to our minds faintly ridiculous. ²⁵

The end result will not be a simple outcome, as it depends on laws in different jurisdictions and will vary from company to company. But the impacts are huge: all kinds of disputes surrounding these companies focus on their responsibility for some aspects of their business that they are trying to avoid. Here is a brief list:

• Are Uber drivers employees? Their flexibility suggests not, but that is not the only criterion for employment: many jurisdictions also weigh independence while on the job as a criterion, and Uber drivers do not set their own prices, do not get to select riders, and have many small controls over how they do their job. Many would say that full-time Uber drivers, which is a large part of the overall business, should qualify for many of the benefits that employees get. This would add significantly to Uber’s costs, and present them with even more financial problems. ²⁶

• In the same way, Airbnb maintains that it is not responsible if its hosts break the law or break their tenancy agreements in offering their space for rent, but the company also makes it very difficult for cities to go after hosts because Airbnb does not, in general, share host information with the cities where it operates. ²⁷ Again, the outcome of this dispute will have a real effect on Airbnb’s cost structure.

• While Brian Chesky was confident that his rating systems could solve consumer protection problems, it has turned out that there are many that ratings cannot address. ²⁸ Whether it is providing a usable complaint system, ensuring vehicle safety, ensuring the safety of Airbnb apartments, so far the Sharing Economy companies have avoided many of the costs that other providers of commercial services have to undertake. To the extent that they have to take on the cost of things going wrong for consumers, the zero-marginal cost model will vanish.

• As all the transactions on the platform go through the Internet, both companies have set up shell structures so that they are routed through subsidiaries in tax-friendly jurisdictions. ²⁹ By not paying into the tax base of the cities where they operate, the companies are essentially parasitic on cities.

• There are also many aspects of running cities that governments care about, from road congestion to housing affordability, where there have been continual disputes about the responsibility of the platform companies. Airbnb has long promoted a vision of openness and authenticity, promising that you could “live like a local”, but in some of their biggest markets the locals have not been so happy about it: Paris neighbourhoods have been losing population, Barcelona has said “bring refugees, keep out tourists”. Managing tourism effectively is a matter of balance, and cities are becoming increasingly assertive in their dealings with Airbnb.

The lesson from all these disputes is that the business model of new technology companies is not dictated solely by the technology itself, it also reflects the responsibilities we expect companies – particularly large global companies – to accept in order to operate.

It has been interesting to watch cities become more assertive over time. When Airbnb and Uber appeared on the scene full of brash promises, they were intimidated by the challenge to get on board with the future or be left behind, especially when the platforms could mobilize thousands of happy customers and service providers to lobby on their behalf. But now cities are realizing that some problems are shared, and that they have choices beyond “take it or leave it”. Major tourist cities are now regularly talking to each other about their successes and failures when dealing with Airbnb, for example, and out of this is growing a new set of rules that balance the interests of tourists with those of the citizens in the places they visit.

If the “zero marginal cost” model is shaped by laws as much as by technology, it raises the question of who builds the future? The debates around the Sharing Economy mirror earlier debates that have grown around technological innovation, from the industrial revolution on. Changes are one part technological, and one part social, and getting the social arrangements right is just as important for a healthy future as getting the technology right. Those in the 19^th century who fought for child labour protections or the forty hour working week or the vote were just as much a part of building the future as the new industrialists and inventors.

It’s the same today. Technology will be a part of the changes that shape the lives of our children, but so too will the legal and social changes that surround them. So an understanding of history, of economics, of society and what a good society needs is just as important as developing a new twist on an artificial intelligence algorithm when it comes to shaping the future. Educators in particular play an important role in ensuring the next generation has the critical faculties and understanding of society that is going to be needed to ensure they have a healthy planet and a healthy relationship with our computational future.

The future of transport and travel is not being made only by the technology companies, it is being made in city government offices around the world. It will be made not just by today’s computer science students, but by our economics students and business students as well.

Aarhus

Survey dates: 2016-10-28 (2258 listings), 2016-11-26 (1900 listings), 2017-01-21 (2167 listings), 2017-02-21 (2295 listings), 2017-03-30 (2323 listings), 2017-04-18 (2398 listings), 2017-04-28 (2360 listings), 2017-05-15 (2437 listings), 2017-06-19 (2802 listings), 2017-07-28 (3142 listings)

Download zip file

Alpes Maritime

Survey dates: 2016-09-06 (24868 listings), 2017-01-28 (25888 listings), 2017-02-26 (26707 listings), 2017-04-24 (26057 listings), 2017-06-19 (28554 listings), 2017-07-12 (32343 listings)

Download zip file

Amsterdam

Survey dates: 2014-05-24 (5645 listings), 2014-08-30 (6840 listings), 2015-01-19 (7693 listings), 2015-03-17 (7807 listings), 2015-12-16 (9850 listings), 2016-05-31 (13360 listings), 2016-08-04 (15314 listings), 2016-12-15 (17018 listings), 2017-01-18 (16871 listings), 2017-02-17 (16542 listings), 2017-03-27 (16362 listings), 2017-04-24 (17037 listings), 2017-05-12 (17201 listings), 2017-06-15 (17795 listings), 2017-07-22 (18723 listings)

Download zip file

Anchorage

Survey dates: 2016-05-13 (447 listings), 2017-03-10 (728 listings), 2017-04-12 (764 listings), 2017-05-22 (842 listings), 2017-06-24 (932 listings), 2017-07-20 (969 listings)

Download zip file

Asheville

Survey dates: 2016-09-10 (757 listings), 2016-12-26 (799 listings), 2017-03-10 (783 listings), 2017-04-12 (788 listings), 2017-05-14 (810 listings), 2017-06-24 (848 listings), 2017-07-20 (854 listings)

Download zip file

Asturias

Survey dates: 2017-07-02 (1225 listings), 2017-07-26 (1327 listings)

Download zip file

Auckland

Survey dates: 2016-08-19 (4590 listings), 2016-12-26 (6103 listings), 2017-01-16 (6230 listings), 2017-02-12 (6507 listings), 2017-03-25 (7384 listings), 2017-04-13 (7889 listings), 2017-04-22 (8010 listings), 2017-05-10 (8487 listings), 2017-06-14 (8847 listings), 2017-07-19 (8815 listings)

Download zip file

Barcelona

Survey dates: 2014-05-20 (10441 listings), 2014-09-04 (10413 listings), 2015-01-14 (12467 listings), 2015-04-29 (10468 listings), 2015-11-06 (14012 listings), 2016-05-30 (17154 listings), 2016-07-25 (19145 listings), 2016-08-28 (18953 listings), 2016-10-22 (16383 listings), 2016-12-10 (15148 listings), 2017-01-18 (17153 listings), 2017-02-18 (17269 listings), 2017-03-27 (17385 listings), 2017-04-15 (17448 listings), 2017-04-24 (17572 listings), 2017-05-12 (17775 listings), 2017-06-15 (18558 listings), 2017-07-23 (18838 listings)

Download zip file

Belgium

Survey dates: 2016-12-30 (15711 listings), 2017-01-29 (15761 listings), 2017-02-28 (15695 listings), 2017-04-26 (15980 listings), 2017-06-19 (17093 listings), 2017-07-14 (17651 listings)

Download zip file

Berlin

Survey dates: 2014-05-18 (7789 listings), 2015-01-19 (11553 listings), 2015-07-04 (12903 listings), 2015-11-07 (14132 listings), 2015-11-30 (14922 listings), 2016-05-01 (16135 listings), 2016-05-29 (8490 listings), 2016-08-30 (15680 listings), 2016-10-09 (16540 listings), 2016-12-15 (18551 listings), 2017-01-01 (18684 listings), 2017-01-17 (18312 listings), 2017-02-16 (18473 listings), 2017-03-26 (18999 listings), 2017-04-23 (19528 listings), 2017-05-11 (19681 listings), 2017-06-15 (20415 listings), 2017-07-21 (21864 listings25 (2443 listings), 2017-02-12 (2565 listings), 2017-03-11 (2580 listings), 2017-04-10 (2634 listings), 2017-05-08 (2645 listings), 2017-06-10 (2721 listings), 2017-07-10 (2954 listings)

Download zip file

Cambridge MA

Survey dates: 2016-09-30 (1442 listings), 2017-03-10 (1694 listings), 2017-04-12 (1713 listings), 2017-05-14 (1765 listings), 2017-06-24 (1825 listings), 2017-07-20 (1809 listings)

Download zip file

Florence

Survey dates: 2015-10-11 (5309 listings), 2015-11-02 (6175 listings), 2015-12-20 (6583 listings), 2016-01-07 (6698 listings), 2016-08-05 (8553 listings), 2016-09-12 (8525 listings), 2016-12-26 (8702 listings), 2017-01-24 (8725 listings), 2017-02-22 (8753 listings), 2017-04-04 (8945 listings), 2017-04-29 (9198 listings), 2017-05-19 (9215 listings), 2017-06-22 (9394 listings)

Download zip file

Frankfurt

Survey dates: 2016-03-27 (1580 listings), 2016-08-21 (1717 listings), 2016-12-26 (1944 listings), 2017-01-23 (1910 listings), 2017-02-22 (1968 listings), 2017-04-02 (1998 listings), 2017-04-28 (2134 listings), 2017-05-18 (1957 listings), 2017-06-22 (1995 listings)

Download zip file

Geneva

Survey dates: 2014-11-26 (702 listings), 2016-08-09 (666 listings), 2017-01-23 (1663 listings), 2017-03-10 (1676 listings), 2017-04-12 (1738 listings), 2017-05-14 (1782 listings), 2017-06-24 (1849 listings), 2017-07-20 (1954 listings)

Download zip file

Girona

Survey dates: 2017-05-13 (393 listings), 2017-06-04 (403 listings), 2017-06-23 (410 listings)

Download zip file

Granada

Survey dates: 2017-07-06 (3565 listings), 2017-07-25 (3595 listings)

Download zip file

Groningen

Survey dates: 2016-03-18 (382 listings), 2016-03-28 (386 listings), 2016-12-26 (533 listings), 2017-01-24 (588 listings), 2017-02-22 (607 listings), 2017-04-04 (610 listings), 2017-04-29 (648 listings), 2017-05-19 (645 listings), 2017-06-22 (706 listings)

Download zip file

Helsinki

Survey dates: 2016-05-07 (1971 listings), 2016-12-26 (2742 listings), 2017-01-23 (2733 listings), 2017-04-02 (2792 listings), 2017-04-28 (2932 listings), 2017-05-18 (2936 listings), 2017-06-22 (3191 listings)

Download zip file

Houston

Survey dates: 2017-04-30 (7758 listings), 2017-05-08 (7712 listings), 2017-06-11 (7677 listings), 2017-07-13 (7559 listings)

Download zip file

Hungary

Survey dates: 2017-02-24 (4507 listings), 2017-04-28 (10517 listings), 2017-05-19 (10605 listings), 2017-06-21 (11141 listings), 2017-07-18 (12139 listings)

Download zip file

Iceland

Survey dates: 2016-07-29 (4768 listings), 2016-09-24 (5005 listings), 2016-10-25 (4993 listings), 2016-11-29 (5180 listings), 2016-12-22 (5573 listings), 2017-01-27 (5629 listings), 2017-02-26 (5820 listings), 2017-04-23 (6255 listings), 2017-05-14 (6496 listings), 2017-06-18 (6876 listings), 2017-07-12 (7117 listings)

Download zip file

Jersey City

Survey dates: 2016-05-26 (1284 listings), 2016-12-26 (1538 listings), 2017-03-10 (1561 listings), 2017-04-12 (1647 listings), 2017-05-14 (1694 listings), 2017-06-24 (1808 listings), 2017-07-20 (1798 listings)

Download zip file

Joshua Tree

Survey dates: 2016-05-27 (227 listings), 2016-11-06 (234 listings), 2017-03-10 (154 listings), 2017-04-12 (160 listings), 2017-05-14 (171 listings), 2017-06-24 (174 listings), 2017-07-20 (178 listings)

Download zip file

Kenya

Survey dates: 2017-05-12 (4420 listings), 2017-07-23 (4764 listings)

Download zip file

Kitchener Waterloo

Survey dates: 2016-05-22 (262 listings), 2016-09-22 (274 listings), 2016-10-28 (282 listings), 2016-12-10 (346 listings), 2016-12-25 (326 listings), 2017-02-12 (353 listings), 2017-03-11 (343 listings), 2017-04-10 (349 listings), 2017-05-08 (366 listings), 2017-06-10 (382 listings), 2017-07-10 (388 listings)

Download zip file

Lake Tahoe

Survey dates: 2017-05-08 (4365 listings), 2017-06-29 (4342 listings), 2017-07-25 (4403 listings)

Download zip file

Las Vegas

Survey dates: 2017-02-25 (5268 listings), 2017-03-10 (5261 listings), 2017-04-12 (5520 listings), 2017-05-14 (5691 listings), 2017-06-24 (5823 listings), 2017-07-20 (5830 listings)

Download zip file

Lenox

Survey dates: 2017-06-25 (81 listings)

Download zip file

Lincoln County

Survey dates: 2016-04-27 (435 listings), 2017-03-11 (810 listings), 2017-04-12 (492 listings), 2017-05-15 (878 listings), 2017-06-24 (941 listings), 2017-07-20 (1190 listings)

Download zip file

Linkoping

Survey dates: 2016-08-02 (57 listings), 2017-03-11 (50 listings), 2017-04-12 (50 listings), 2017-05-15 (56 listings), 2017-06-24 (67 listings), 2017-07-20 (70 listings)

Download zip file

Lisbon

Survey dates: 2015-03-18 (5652 listings), 2016-03-20 (8968 listings), 2016-06-02 (10272 listings), 2016-09-12 (10998 listings), 2016-12-26 (11668 listings), 2017-01-21 (11654 listings), 2017-02-21 (11792 listings), 2017-03-30 (12045 listings), 2017-04-18 (12287 listings), 2017-04-27 (12499 listings), 2017-05-15 (12797 listings), 2017-06-19 (13232 listings), 2017-07-27 (13578 listings)

Download zip file

London

Survey dates: 2013-12-21 (13054 listings), 2014-05-13 (13237 listings), 2015-01-17 (21298 listings), 2016-01-09 (34414 listings), 2016-08-07 (48515 listings), 2016-09-22 (47472 listings), 2016-12-26 (53313 listings), 2017-01-21 (53947 listings), 2017-03-31 (55846 listings), 2017-04-28 (57826 listings), 2017-06-19 (60561 listings), 2017-07-28 (64144 listings)

Download zip file

Los Angeles

Survey dates: 2014-09-01 (10027 listings), 2014-09-18 (11317 listings), 2014-10-25 (11780 listings), 2015-04-06 (13899 listings), 2015-08-03 (15790 listings), 2015-08-27 (16649 listings), 2015-09-09 (16968 listings), 2015-10-13 (17420 listings), 2015-11-16 (18665 listings), 2015-12-10 (19709 listings), 2016-01-11 (18863 listings), 2016-02-10 (19311 listings), 2016-03-12 (19717 listings), 2016-04-12 (20951 listings), 2016-05-14 (21986 listings), 2016-06-14 (21996 listings), 2016-07-13 (23071 listings), 2016-08-15 (23793 listings), 2016-09-13 (24744 listings), 2016-10-14 (25347 listings), 2016-11-17 (24633 listings), 2016-12-17 (30735 listings), 2017-01-12 (29517 listings), 2017-02-10 (29316 listings), 2017-03-10 (29252 listings), 2017-04-06 (29955 listings), 2017-05-03 (30696 listings), 2017-06-05 (30829 listings), 2017-07-08 (32146 listings)

Download zip file

Lyon

Survey dates: 2016-05-14 (7346 listings), 2016-10-02 (8106 listings), 2016-12-27 (9816 listings), 2017-03-11 (9494 listings), 2017-04-12 (9707 listings), 2017-05-15 (9856 listings), 2017-06-24 (10186 listings), 2017-07-20 (10409 listings)

Download zip file

Madrid

Survey dates: 2014-08-29 (4790 listings), 2015-02-21 (5393 listings), 2017-03-11 (12615 listings)

Download zip file

Mallorca

Survey dates

Download zip file

Melun

Survey dates: 2017-03-24 (37 listings)

Download zip file

Menorca

Survey dates: 2016-04-19 (1769 listings), 2016-04-20 (1757 listings), 2016-11-06 (1866 listings)

Download zip file

Miami

Survey dates: 2015-08-17 (6947 listings), 2015-10-03 (7057 listings), 2015-10-18 (7414 listings), 2015-11-19 (7676 listings), 2015-12-12 (8349 listings), 2016-01-13 (8553 listings), 2016-02-12 (8855 listings), 2016-03-14 (9468 listings), 2016-04-12 (7034 listings), 2016-05-15 (7244 listings), 2016-06-15 (7162 listings), 2016-07-14 (7249 listings), 2016-08-16 (7212 listings), 2016-09-14 (7213 listings), 2016-10-15 (6563 listings), 2016-11-18 (6856 listings), 2016-12-21 (9811 listings), 2017-01-13 (7810 listings), 2017-02-12 (7845 listings), 2017-03-12 (7664 listings), 2017-04-07 (7586 listings), 2017-05-04 (7948 listings), 2017-06-04 (9608 listings), 2017-07-09 (8824 listings)

Download zip file

Montreal

Survey dates: 2014-08-30 (4424 listings), 2015-01-15 (6216 listings), 2015-09-02 (9236 listings), 2015-10-03 (9234 listings), 2016-06-29 (12476 listings), 2016-07-26 (12884 listings), 2016-08-20 (12587 listings), 2016-09-21 (12021 listings), 2016-10-27 (12444 listings), 2016-11-26 (11489 listings), 2016-12-08 (11518 listings), 2016-12-25 (13814 listings), 2017-02-11 (13729 listings), 2017-03-10 (13877 listings), 2017-04-10 (14342 listings), 2017-05-07 (14738 listings), 2017-06-10 (15286 listings), 2017-07-09 (15625 listings)

Download zip file

Moscow

Survey dates: 2014-05-26 (1589 listings), 2014-08-24 (1843 listings), 2015-10-11 (3585 listings), 2015-11-02 (3819 listings), 2016-10-08 (5561 listings), 2017-03-11 (6201 listings), 2017-04-13 (6527 listings), 2017-05-15 (6789 listings), 2017-06-26 (6973 listings), 2017-07-21 (7200 listings)

Download zip file

Nairobi

Survey dates: 2017-05-12 (2516 listings), 2017-06-29 (2762 listings), 2017-07-25 (2837 listings)

Download zip file

Nantes

Survey dates: 2017-07-01 (2648 listings)

Download zip file

Nashville

Survey dates: 2016-05-22 (3012 listings), 2016-08-30 (3486 listings), 2016-10-10 (3603 listings), 2016-12-16 (4067 listings), 2017-03-11 (4077 listings), 2017-04-13 (4357 listings), 2017-05-16 (4463 listings), 2017-06-26 (4787 listings), 2017-07-21 (4804 listings)

Download zip file

New Orleans

Survey dates: 2015-01-16 (1905 listings), 2015-08-21 (2092 listings), 2015-10-03 (2667 listings), 2015-11-19 (2941 listings), 2015-12-13 (3041 listings), 2016-01-15 (3319 listings), 2016-02-15 (3697 listings), 2016-03-16 (3857 listings), 2016-04-12 (4177 listings), 2016-05-17 (4378 listings), 2016-06-16 (4069 listings), 2016-07-15 (4432 listings), 2016-08-17 (4482 listings), 2016-09-15 (4525 listings), 2016-10-16 (4449 listings), 2016-11-19 (4707 listings), 2017-01-13 (5067 listings), 2017-02-12 (5540 listings), 2017-03-12 (5663 listings), 2017-04-08 (5722 listings), 2017-05-04 (5855 listings), 2017-06-21 (3739 listings), 2017-07-10 (3862 listings)

Download zip file

New York

Survey dates: 2014-05-10 (18921 listings), 2014-08-31 (20777 listings), 2014-10-17 (22388 listings), 2014-12-02 (25405 listings), 2015-02-27 (26441 listings), 2015-03-14 (28552 listings), 2015-08-10 (30001 listings), 2015-09-10 (30059 listings), 2015-10-13 (29898 listings), 2015-12-04 (33926 listings), 2015-12-21 (36062 listings), 2016-01-21 (36003 listings), 2016-02-25 (34935 listings), 2016-03-24 (34678 listings), 2016-04-03 (34380 listings), 2016-04-20 (34661 listings), 2016-05-23 (35067 listings), 2016-06-23 (34205 listings), 2016-07-20 (35547 listings), 2016-08-24 (37769 listings), 2016-09-20 (35620 listings), 2016-10-23 (35795 listings), 2016-11-26 (37439 listings), 2016-12-26 (39079 listings), 2017-01-15 (38768 listings), 2017-02-17 (37878 listings), 2017-03-14 (38549 listings), 2017-04-21 (39745 listings), 2017-05-06 (40733 listings), 2017-06-12 (39167 listings), 2017-07-12 (41245 listings)

Download zip file

Newhaven CT

Survey dates: 2015-08-21 (369 listings), 2015-10-03 (475 listings), 2015-10-21 (431 listings), 2015-11-26 (567 listings), 2015-12-15 (586 listings), 2016-01-18 (604 listings), 2016-02-20 (632 listings), 2016-03-21 (678 listings), 2016-04-16 (473 listings), 2016-05-20 (481 listings), 2016-06-20 (489 listings), 2016-07-18 (444 listings), 2016-08-20 (431 listings), 2016-09-17 (444 listings), 2016-10-20 (448 listings), 2016-11-23 (471 listings), 2016-12-24 (407 listings), 2017-01-14 (359 listings), 2017-02-16 (373 listings), 2017-03-13 (370 listings), 2017-04-08 (388 listings), 2017-05-05 (420 listings), 2017-06-10 (414 listings), 2017-07-11 (428 listings)

Download zip file

Newport

Survey dates: 2016-04-26 (57 listings), 2016-12-26 (70 listings), 2017-03-11 (94 listings), 2017-04-13 (61 listings), 2017-05-16 (83 listings), 2017-06-27 (138 listings), 2017-07-22 (148 listings)

Download zip file

Nice

Survey dates: 2016-06-25 (8802 listings), 2016-10-06 (9288 listings), 2016-12-08 (9444 listings), 2017-01-21 (9511 listings), 2017-02-21 (9604 listings), 2017-03-30 (9962 listings), 2017-04-18 (10168 listings), 2017-04-28 (10229 listings), 2017-05-16 (10576 listings), 2017-06-19 (11193 listings), 2017-07-28 (11629 listings)

Download zip file

Oahu

Survey dates: 2015-08-17 (2481 listings), 2015-10-18 (2696 listings), 2015-11-19 (2855 listings), 2015-12-12 (3086 listings), 2016-01-14 (3193 listings), 2016-02-13 (3214 listings), 2016-03-15 (3190 listings), 2016-04-14 (3539 listings), 2016-05-16 (3665 listings), 2016-06-15 (3844 listings), 2016-07-14 (3953 listings), 2016-08-17 (4020 listings), 2016-09-14 (4233 listings), 2016-10-15 (4200 listings), 2016-11-18 (4662 listings), 2016-12-21 (5146 listings), 2017-01-13 (5051 listings), 2017-02-12 (5249 listings), 2017-03-12 (5346 listings), 2017-04-07 (5610 listings), 2017-05-04 (5863 listings), 2017-06-08 (6028 listings), 2017-07-09 (6163 listings)

Download zip file

Ottawa

Survey dates: 2016-05-16 (1496 listings), 2016-08-23 (1586 listings), 2016-09-21 (1624 listings), 2016-10-27 (1736 listings), 2016-12-07 (1897 listings), 2016-12-25 (1906 listings), 2017-02-12 (2003 listings), 2017-03-11 (2090 listings), 2017-04-10 (2257 listings), 2017-05-08 (2476 listings), 2017-06-10 (2919 listings), 2017-07-10 (3593 listings)

Download zip file

Palermo

Survey dates: 2016-09-14 (3635 listings), 2016-12-26 (3770 listings), 2017-03-11 (3872 listings), 2017-04-13 (3996 listings), 2017-05-16 (4086 listings), 2017-06-27 (4284 listings), 2017-07-22 (4434 listings)

Download zip file

Paris

Survey dates: 2013-11-27 (19004 listings), 2014-05-16 (19946 listings), 2014-09-21 (27326 listings), 2014-12-26 (28826 listings), 2015-02-28 (31341 listings), 2015-12-17 (40005 listings), 2015-12-31 (41761 listings), 2016-05-26 (55951 listings), 2016-08-10 (62213 listings), 2017-01-01 (62963 listings), 2017-01-19 (62043 listings), 2017-02-18 (61666 listings), 2017-03-28 (62087 listings), 2017-04-25 (63634 listings), 2017-06-16 (65727 listings), 2017-07-25 (70158 listings)

Download zip file

Peel

Survey dates: 2016-09-24 (838 listings), 2016-10-29 (875 listings), 2016-12-09 (919 listings), 2016-12-25 (896 listings), 2017-02-12 (927 listings), 2017-03-11 (930 listings), 2017-04-10 (972 listings), 2017-05-08 (1017 listings), 2017-06-10 (1013 listings), 2017-07-10 (1064 listings)

Download zip file

Philadelphia

Survey dates: 2015-08-20 (5990 listings), 2015-10-18 (7318 listings), 2015-11-19 (7177 listings), 2015-12-12 (7141 listings), 2016-01-14 (6895 listings), 2016-02-14 (7062 listings), 2016-03-16 (7026 listings), 2016-04-15 (6976 listings), 2016-05-16 (6813 listings), 2016-06-16 (6753 listings), 2016-07-15 (6921 listings), 2016-08-17 (6928 listings), 2016-09-15 (6904 listings), 2016-10-16 (6600 listings), 2016-11-19 (6875 listings), 2017-01-13 (7068 listings), 2017-02-12 (7152 listings), 2017-03-12 (7101 listings), 2017-04-07 (7090 listings), 2017-05-04 (7242 listings), 2017-06-08 (7391 listings), 2017-07-09 (7367 listings)

Download zip file

Pittsburgh

Survey dates: 2017-04-30 (850 listings), 2017-05-08 (892 listings), 2017-06-12 (924 listings), 2017-07-13 (879 listings)

Download zip file

Portland

Survey dates: 2014-10-28 (1748 listings), 2015-02-22 (1900 listings), 2015-08-22 (1917 listings), 2015-10-23 (2622 listings), 2015-10-26 (2722 listings), 2015-11-21 (2847 listings), 2015-12-13 (2952 listings), 2016-01-15 (2897 listings), 2016-02-15 (2970 listings), 2016-03-17 (3077 listings), 2016-04-13 (3128 listings), 2016-05-17 (3238 listings), 2016-06-17 (3383 listings), 2016-07-15 (3496 listings), 2016-08-18 (3661 listings), 2016-09-15 (3714 listings), 2016-10-17 (3626 listings), 2016-11-19 (3914 listings), 2017-01-14 (3860 listings), 2017-02-16 (3837 listings), 2017-03-12 (3748 listings), 2017-04-08 (3634 listings), 2017-05-05 (3690 listings), 2017-06-09 (3815 listings), 2017-07-10 (3950 listings)

Download zip file

Portland ME

Survey dates: 2016-07-31 (409 listings), 2017-03-12 (456 listings), 2017-04-13 (476 listings), 2017-05-16 (519 listings), 2017-06-27 (576 listings), 2017-07-22 (591 listings)

Download zip file

Porto

Survey dates: 2016-10-13 (4174 listings), 2016-12-12 (4461 listings), 2017-03-12 (4729 listings), 2017-04-13 (4971 listings), 2017-05-16 (5150 listings), 2017-06-27 (5561 listings), 2017-07-22 (5664 listings)

Download zip file

Prince Georges County

Survey dates: 2017-02-25 (1120 listings), 2017-03-12 (1121 listings), 2017-04-13 (1135 listings), 2017-05-16 (1149 listings), 2017-06-27 (1185 listings), 2017-07-22 (1203 listings)

Download zip file

Quebec City

Survey dates: 2017-05-17 (2022 listings), 2017-06-11 (2120 listings), 2017-07-10 (2153 listings)

Download zip file

Queenstown

Survey dates: 2017-03-28 (1171 listings), 2017-04-25 (1137 listings), 2017-05-10 (1205 listings), 2017-06-14 (1227 listings), 2017-07-18 (1210 listings)

Download zip file

Reykjavik

Survey dates: 2016-04-24 (2551 listings), 2016-06-06 (2799 listings), 2016-06-30 (2936 listings), 2016-08-08 (2813 listings), 2016-09-11 (2821 listings), 2016-11-02 (2807 listings), 2016-11-23 (2996 listings), 2016-12-25 (3036 listings), 2017-01-01 (3053 listings), 2017-01-21 (2969 listings), 2017-02-22 (3046 listings), 2017-03-31 (3091 listings), 2017-04-19 (3237 listings), 2017-04-28 (3178 listings), 2017-05-16 (3306 listings), 2017-06-19 (3406 listings), 2017-07-28 (3650 listings)

Download zip file

Richmond BC

Survey dates: 2017-01-13 (760 listings), 2017-02-12 (627 listings), 2017-03-13 (620 listings), 2017-04-12 (561 listings), 2017-05-12 (531 listings), 2017-06-14 (540 listings), 2017-07-13 (514 listings)

Download zip file

Rome

Survey dates: 2013-11-27 (1714 listings), 2014-05-19 (7810 listings), 2014-08-25 (9168 listings), 2014-11-28 (10778 listings), 2015-08-29 (15703 listings), 2016-01-06 (17900 listings), 2016-08-05 (23754 listings), 2016-09-11 (24061 listings), 2016-12-26 (25104 listings), 2017-01-16 (25275 listings), 2017-02-16 (25226 listings), 2017-03-25 (25430 listings), 2017-04-13 (25687 listings), 2017-04-22 (26000 listings), 2017-05-11 (26065 listings), 2017-06-14 (26206 listings), 2017-07-19 (26804 listings)

Download zip file

Saint Denis

Survey dates: 2017-03-10 (564 listings)

Download zip file

Saint Lucia

Survey dates: 2016-09-25 (642 listings), 2017-03-12 (718 listings), 2017-04-14 (743 listings), 2017-05-17 (761 listings), 2017-06-27 (799 listings), 2017-07-23 (835 listings)

Download zip file

Saint Malo

Survey dates: 2017-05-17 (1285 listings), 2017-06-29 (1408 listings), 2017-07-25 (1460 listings)

Download zip file

Salvador

Survey dates: 2017-07-01 (4308 listings)

Download zip file

San Diego

Survey dates: 2014-05-25 (1901 listings), 2014-11-03 (2632 listings), 2015-08-22 (3318 listings), 2015-10-03 (4014 listings), 2015-10-23 (4222 listings), 2015-12-03 (4482 listings), 2015-12-16 (4587 listings), 2016-01-19 (4721 listings), 2016-02-22 (4813 listings), 2016-03-23 (4954 listings), 2016-04-18 (5356 listings), 2016-05-22 (5764 listings), 2016-06-21 (6257 listings), 2016-07-19 (7056 listings), 2016-08-22 (7177 listings), 2016-09-19 (7394 listings), 2016-10-22 (7481 listings), 2016-11-25 (7839 listings), 2016-12-25 (8030 listings), 2017-01-15 (8021 listings), 2017-02-17 (8054 listings), 2017-04-09 (8225 listings), 2017-05-06 (8316 listings), 2017-06-11 (8643 listings), 2017-07-11 (9111 listings)

Download zip file

San Francisco

Survey dates: 2013-11-17 (215 listings), 2014-05-11 (4771 listings), 2014-08-24 (4771 listings), 2015-02-19 (5202 listings), 2015-06-05 (5305 listings), 2015-08-21 (5140 listings), 2015-10-21 (5587 listings), 2015-11-21 (7146 listings), 2015-12-14 (7399 listings), 2016-01-16 (7604 listings), 2016-02-17 (8549 listings), 2016-03-19 (8539 listings), 2016-04-15 (8051 listings), 2016-05-19 (7913 listings), 2016-06-04 (8669 listings), 2016-06-18 (7783 listings), 2016-07-17 (8037 listings), 2016-08-19 (7983 listings), 2016-09-17 (8076 listings), 2016-10-19 (8236 listings), 2016-11-21 (8407 listings), 2016-12-23 (8899 listings), 2017-01-14 (8508 listings), 2017-02-16 (8397 listings), 2017-03-12 (8286 listings), 2017-03-24 (8310 listings), 2017-04-08 (8319 listings), 2017-05-05 (8391 listings), 2017-07-10 (8344 listings)

Download zip file

Santa Barbara

Survey dates: 2017-01-15 (467 listings), 2017-03-12 (403 listings), 2017-04-14 (398 listings), 2017-05-17 (359 listings), 2017-06-27 (351 listings), 2017-07-23 (384 listings)

Download zip file

Sao Paulo

Survey dates: 2017-07-01 (11118 listings)

Download zip file

Savannah

Survey dates: 2016-05-13 (494 listings), 2016-12-26 (615 listings), 2017-03-12 (717 listings), 2017-04-14 (752 listings), 2017-05-17 (777 listings), 2017-06-27 (835 listings), 2017-07-23 (877 listings)

Download zip file

Seattle

Survey dates: 2015-09-01 (3223 listings), 2015-10-23 (3544 listings), 2015-11-28 (3715 listings), 2015-12-15 (3836 listings), 2016-01-19 (3871 listings), 2016-02-22 (3974 listings), 2016-03-22 (4048 listings), 2016-04-18 (4170 listings), 2016-05-22 (4413 listings), 2016-06-21 (4654 listings), 2016-07-19 (4935 listings), 2016-08-22 (5153 listings), 2016-09-18 (5253 listings), 2016-10-21 (5393 listings), 2016-11-24 (5654 listings), 2016-12-25 (5858 listings), 2017-01-15 (5717 listings), 2017-02-17 (5834 listings), 2017-03-13 (5816 listings), 2017-04-09 (5920 listings), 2017-05-06 (6062 listings), 2017-06-11 (6234 listings), 2017-07-11 (6399 listings)

Download zip file

Seoul

Survey dates: 2016-05-29 (8519 listings), 2016-10-02 (9969 listings), 2016-12-28 (10690 listings), 2017-01-15 (10905 listings), 2017-01-15 (10930 listings), 2017-02-11 (11121 listings), 2017-03-24 (11544 listings), 2017-04-12 (11619 listings), 2017-04-21 (11723 listings), 2017-05-10 (11516 listings), 2017-06-14 (11965 listings), 2017-07-17 (12406 listings)

Download zip file

Singapore

Survey dates: 2015-06-28 (2815 listings), 2016-06-11 (5796 listings), 2016-10-11 (6753 listings), 2017-03-12 (6828 listings), 2017-04-14 (7201 listings), 2017-05-17 (7212 listings), 2017-06-27 (7003 listings), 2017-07-23 (6834 listings)

Download zip file

Sri Lanka

Survey dates: 2016-04-27 (7107 listings), 2016-09-24 (8725 listings), 2017-04-28 (11568 listings), 2017-06-21 (12176 listings), 2017-07-19 (12400 listings)

Download zip file

Switzerland

Survey dates: 2016-04-28 (15587 listings), 2016-08-13 (18901 listings), 2016-10-30 (19775 listings), 2016-11-30 (21043 listings), 2016-12-23 (22104 listings), 2017-01-26 (22506 listings), 2017-02-24 (23195 listings), 2017-04-21 (24992 listings), 2017-05-20 (25674 listings), 2017-06-16 (26398 listings), 2017-07-11 (27696 listings)

Download zip file

Sydney

Survey dates: 2014-09-26 (5567 listings), 2015-01-20 (7424 listings), 2015-03-20 (7871 listings), 2016-01-29 (13134 listings), 2016-06-11 (17343 listings), 2017-01-25 (23859 listings), 2017-03-12 (23297 listings), 2017-04-14 (24006 listings), 2017-05-17 (24419 listings), 2017-06-28 (25266 listings), 2017-07-23 (25506 listings)

Download zip file

Tallin

Survey dates: 2016-07-08 (1378 listings), 2016-12-26 (1468 listings), 2017-01-24 (1483 listings), 2017-02-22 (1536 listings), 2017-04-04 (1574 listings), 2017-04-29 (1668 listings), 2017-05-19 (1708 listings), 2017-06-22 (1905 listings)

Download zip file

Toronto

Survey dates: 2013-12-14 (2617 listings), 2014-05-18 (2656 listings), 2015-02-18 (4831 listings), 2015-06-29 (6587 listings), 2015-07-25 (6586 listings), 2015-11-21 (7607 listings), 2015-12-13 (7969 listings), 2016-01-15 (7980 listings), 2016-02-16 (8352 listings), 2016-03-17 (8452 listings), 2016-04-13 (9149 listings), 2016-05-18 (9446 listings), 2016-06-17 (9595 listings), 2016-07-16 (10155 listings), 2016-08-18 (10700 listings), 2016-09-15 (11026 listings), 2016-10-17 (11301 listings), 2016-11-20 (11735 listings), 2016-12-17 (12801 listings), 2017-01-14 (12597 listings), 2017-02-16 (12679 listings), 2017-03-12 (12528 listings), 2017-04-08 (12755 listings), 2017-05-05 (13042 listings), 2017-06-09 (13161 listings), 2017-07-10 (13267 listings)

Download zip file

Turin

Survey dates: 2017-01-11 (3092 listings), 2017-03-13 (3205 listings), 2017-04-14 (3301 listings), 2017-05-18 (3403 listings), 2017-06-28 (3546 listings), 2017-07-24 (3662 listings)

Download zip file

Tuscany

Survey dates: 2016-08-02 (37295 listings), 2016-10-30 (38783 listings), 2017-05-04 (32850 listings), 2017-06-23 (24993 listings), 2017-07-20 (49286 listings)

Download zip file

Vancouver BC

Survey dates: 2015-04-25 (3036 listings), 2015-08-22 (3375 listings), 2015-11-26 (4595 listings), 2015-12-15 (4751 listings), 2016-01-17 (4731 listings), 2016-02-19 (4833 listings), 2016-03-20 (4916 listings), 2016-04-17 (5004 listings), 2016-05-20 (5209 listings), 2016-06-19 (5297 listings), 2016-07-18 (5611 listings), 2016-08-20 (5820 listings), 2016-09-17 (5926 listings), 2016-10-20 (5758 listings), 2016-11-22 (5611 listings), 2016-12-23 (5843 listings), 2017-01-14 (5619 listings), 2017-02-16 (5596 listings), 2017-03-13 (5588 listings), 2017-04-08 (5618 listings), 2017-05-05 (5781 listings), 2017-06-10 (5909 listings), 2017-07-10 (5994 listings)

Download zip file

Veneto

Survey dates: 2016-11-07 (11543 listings), 2017-01-24 (15153 listings), 2017-01-30 (15218 listings), 2017-02-22 (15508 listings), 2017-03-01 (15743 listings), 2017-04-03 (16300 listings), 2017-04-29 (16940 listings), 2017-05-18 (17148 listings), 2017-06-19 (17915 listings), 2017-07-16 (18455 listings)

Download zip file

Venice

Survey dates: 2015-12-20 (3161 listings), 2016-11-20 (5323 listings), 2016-12-27 (5727 listings), 2017-01-10 (5773 listings), 2017-01-23 (5801 listings), 2017-02-22 (5955 listings), 2017-04-02 (6120 listings), 2017-04-28 (6404 listings), 2017-05-18 (6402 listings), 2017-06-22 (6410 listings)

Download zip file

Versailles

Survey dates: 2016-09-30 (6003 listings), 2017-03-13 (6052 listings), 2017-04-14 (6271 listings), 2017-05-18 (6515 listings), 2017-06-29 (6756 listings), 2017-07-24 (7175 listings)

Download zip file

Victoria BC

Survey dates: 2016-07-08 (831 listings), 2016-11-06 (841 listings), 2016-12-10 (899 listings), 2016-12-25 (890 listings), 2017-02-12 (949 listings), 2017-03-11 (950 listings), 2017-04-10 (1004 listings), 2017-05-08 (1026 listings), 2017-06-10 (1063 listings), 2017-07-10 (1093 listings)

Download zip file

Vienna

Survey dates: 2015-07-24 (5274 listings), 2015-08-01 (5445 listings), 2017-01-11 (7790 listings), 2017-03-13 (7787 listings), 2017-04-15 (7840 listings), 2017-05-18 (8010 listings), 2017-06-29 (8715 listings), 2017-07-25 (9252 listings)

Download zip file

Warsaw

Survey dates: 2016-12-08 (3299 listings), 2016-12-14 (3389 listings), 2017-03-13 (3726 listings), 2017-04-15 (3863 listings), 2017-05-18 (4016 listings), 2017-06-29 (4325 listings), 2017-07-25 (4593 listings)

Download zip file

Washington

Survey dates: 2015-01-16 (2815 listings), 2015-08-22 (2800 listings), 2015-10-23 (3843 listings), 2015-11-26 (4081 listings), 2015-12-15 (4257 listings), 2016-01-18 (4245 listings), 2016-02-20 (4304 listings), 2016-03-21 (4290 listings), 2016-04-16 (4512 listings), 2016-05-20 (4756 listings), 2016-06-20 (4738 listings), 2016-07-18 (4960 listings), 2016-08-20 (5104 listings), 2016-09-18 (5186 listings), 2016-10-20 (5297 listings), 2016-11-23 (5975 listings), 2016-12-24 (7497 listings), 2017-01-14 (8885 listings), 2017-01-19 (9097 listings), 2017-02-16 (8475 listings), 2017-03-13 (8232 listings), 2017-04-08 (8061 listings), 2017-05-05 (8115 listings), 2017-05-11 (8085 listings), 2017-06-10 (8174 listings), 2017-07-11 (8237 listings)

Download zip file

Wellington

Survey dates: 2016-08-19 (858 listings), 2016-12-26 (1018 listings), 2017-01-16 (1039 listings), 2017-02-12 (1192 listings), 2017-03-25 (1338 listings), 2017-04-13 (1446 listings), 2017-04-22 (1455 listings), 2017-05-10 (1577 listings), 2017-06-14 (1894 listings), 2017-07-18 (1866 listings)

Download zip file

West Indies

Survey dates: 2016-09-28 (39195 listings)

Download zip file

Yellowknife

Survey dates: 2016-09-27 (67 listings), 2016-11-20 (78 listings), 2016-12-26 (84 listings), 2017-02-12 (93 listings), 2017-03-11 (89 listings), 2017-04-10 (83 listings), 2017-05-08 (93 listings), 2017-06-10 (95 listings), 2017-07-10 (88 listings)

Download zip file

York

Survey dates: 2016-11-21 (551 listings), 2016-11-29 (754 listings), 2016-11-30 (587 listings), 2016-12-04 (495 listings), 2016-12-15 (511 listings), 2017-01-04 (92 listings), 2017-01-18 (496 listings), 2017-02-10 (519 listings), 2017-02-17 (532 listings), 2017-03-27 (545 listings), 2017-04-14 (563 listings), 2017-04-24 (561 listings), 2017-04-21 (607 listings), 2017-05-12 (580 listings), 2017-05-22 (139 listings), 2017-06-15 (617 listings), 2017-07-22 (670 listings)

Download zip file

Zagreb

Survey dates

Download zip file

Zurich

Survey dates: 2016-01-11 (912 listings), 2017-04-17 (2410 listings), 2017-05-18 (2271 listings), 2017-06-22 (2392 listings)

Download zip file

Book

My second book What’s Yours is Mine: Against the Sharing Economy was published by OR Books in November 2015, but I am including the book and related articles here, making it a “long 2016”.

• US and UK edition (link), OR Books, November 2015

• Canadian edition (link), Between the Lines Press, February 2016

• German edition (Deins ist Meins): Verlag Antje Kunstmann

• Spanish edition: Lo tuyo es mío, Taurus, November 2016

• Simplified Chinese edition: (Ginkgo Book Co.)

• French language edition (France and Québec): Ce Qui est à Toi est à Moi, translated by Hélène Rioux, Lux Editeur

Book chapter

• “Trust and Platform Cooperatives” in Ours to Hack and to Own, edited by Trebor Scholz and Nathan Scheider.

Invited Essays

• The Sharing Economy’s Dirty Laundry, Jacobin, March 2016

• The Sharing Economy isn’t about sharing, Alternet, March 27, 2016 (excerpt)

• What’s Yours is Mine, AFR Weekend, Australian Financial Review, May 28, 2016 (excerpt)

• Airbnb is Facing an Existential Expansion Problem, Harvard Business Review digital article, July 11, 2016

• The downsides of sharing, Monocle (rejected)

Reviews of “What’s Yours is Mine”

• What’s Yours is Mine, Håkan Lindgren, Svenska DagBladet, Jan 11, 2016.

• The broken promise of the sharing economy, Ralf Grötker, Netopia, Feb 3, 2016

• Raw Deals, Ben Terrall, Counterpunch, Feb 12, 2016.

• What’s Yours is Mine, Saul of Hearts, Medium, Feb 25, 2016

• L’atto d’accusa contro la sharing economy, Fabio Chiusi, L’Espresso, 09 March 2016 (Italian)

• The Unforeseen Dangers of Uber and Airbnb, Robert E. Levine, The Spectator (UK), March 5, 2016

• Ist Carsharing das neue Lohndumping? (German), Welt, Oliver Rasche, April 2016.

• Book of the Day, Steven Poole, The Guardian, April 2, 2016.

• Tom Slee’s What’s Yours is Mine, Cathy O’Neil, April 20, 2016

• Die Sharing Economy - nur eine illusion?, Birgit Stratmann, Netzwerk Ethikheute,  (German)

• Sheared, Rohinton Medhora, Literary Review of Canada, May 2016

• Book slams Uber, Airbnb, and Sharing Economy, Thomas Walkom, Toronto Star, May 1, 2016

• What’s Yours is Mine, Peter Richardson, Truthdig, May 2, 2016.

• Teile und herrsche, Thomas Wagner, junge Welt, May 10, 2016

• What’s Yours is Mine, Liz Pelly, Maximum Rocknroll, June 2016

• What’s Yours is Mine, Howard Doughty, The Innovation Journal, 21 (1) 2016

• What’s Yours is Mine, Ruth Latta, The Compulsive Reader, March 25, 2016

• Le duel des critiques, BFM Business, (French, video).

Interviews

• What do we do about a problem like Uber, Andrew Orlowski, The Register, 15 Jan 2016

• Inside the sharing economy, Per Strömbäck, Netopia, 3 Feb 2016

• Every person a market, Chuck Mertz, This is Hell, 23 Feb 2016 (audio),

• The Sharing Economy is Not Your Friend, Drew Millard, Vice, Feb 2016.

• Sharing is not always caring, James Jackson, Waterloo Chronicle, 09 Mar 2016

• Geldwertes Teilen, DeutschlandRadio Kultur, 12 March 2016 (German)

• The technology is just a shiny shopfront, Barbara Speed, The New Statesman, 21 March 2016.

• “Diese Jobs schaffen ein neues Prekariat”, Nils Klawitter, Spiegel Online, 23 March 2016 (German)

• Teile und herrsche, Barbara Hofler, NZZ Am Sontag, 16 April 2016 (German, Switzerland)

• What’s Yours is Mine, Redeye Collective, Vancouver (audio), 29 April 2016

• Sharing the wealth, unevenly, The Agenda with Steve Paikin, TV Ontario, 4 May 2016

• Uber og Airbnb underminerer demokratiet, Assia Awad, djoefbladet, 26 May 2016 (Danish)

• Sharing economy platforms exist for one purpose only, which is to make money for their investors, Felix Simon, Medium, 6 June 2016

• 10 Questions on the sharing economy, The Agenda with Steve Paikin, TV Ontario, 14 June 2016

• Can’t find a home in Seattle? Airbnb rentals may be to blame, Gene Balk, The Seattle Times, 15 June 2016 (Seattle)

• Short-term rentals are reshaping real estate market in some cities, Tony Gonzalez, Marketplace, 20 Sept 2016

• The Sharing Economy Blues, David Hugill, Canadian Dimension, 23 September 2016

• Die Sharing Economie ist eine Blase, (German), Marc Kowalsky, Bilanz, November 2016.

• UK Tribunal Rules in Favour of Uber Drivers (audio), Redeye Collective, Vancouver, 22 Nov 2016

• The so-called sharing economy, Richard Aedy, The Money, Australian Broadcasting Corporation, December 2016, (audio)

Quoted, and Airbnb data

• Nobody’s Business: CCPA report on short-term rentals in Toronto, 22 September 2016

  • Airbnb cutting into available rentals in Toronto, study shows, Metro News, 22 Sep 2016

  • Airbnb cutting into available rentals in city, Tess Kalinowski, Toronto Star, 22 Sep 2016

  • Airbnb Is Sapping Toronto Of Rental Housing: Report, Jesse Ferreras, Huffington Post, 22 Sep 2016

  • (and others)

• Airbnb has made it nearly impossible to find a place to live in this city, Kristen V. Brown, Fusion, May 24. (Reykjavik)

• How do we research the sharing economy when the data can’t be validated?, Andrew Gelman, Monkey Cage Blog (Washington Post), June 6, 2016.

• Cities take aim at Airbnb, Sunny Dhillon, Globe and Mail, 11 July 2016

• Airbnb in Vancouver and its impact on affordable housing, Iain Marjoribanks, 1 June 2016,

• Airbnb open to market regulations in large cities like Vancouver, Mike Hager, Globe and Mail, 25 July 2016 (Vancouver)

• Iceland tourism and the mixed blessings of Airbnb, Andrew Scheivachman, Skift, 19 Aug 2016

• Small towns can’t handle the sharing economy tourism boom, Kristen V. Brown, Fusion, 28 August 2016.

• Airbnb et Uber, Nicholas Tittley, Radio Canada, September 2016. (audio)

• Lancé à Bordeaux, l’observatoire Airbnb veut créer un débat national, Xavier Ridon, Rue89 Bordeaux, 25 Oct 2016 (Bordeaux)

• New York deflates Airbnb, The Economist, 29 Oct 2016.

• The 14 th Global Forum on Tourism Statistics, 23-25 Nov 2016

• Consumo collaborativo: dall’anticapitalismo al rilancio del mercato, Carlo Lottiere, il Giornale, 7 Dec 2016 (Italian)

Report: How Airbnb hid the facts in New York City

Together with Murray Cox of Inside Airbnb,  I produced a report on Airbnb’s removal of 1500 listings in New York, just in time to make its public release of data look good.

• The report: my site, and Inside Airbnb

• List of media reports, including New York Times, The Guardian, and NBC.

• Airbnb admits it purged listings: Fusion,

Events

• Octopus Books (Ottawa)

• Last but not Least Feast, Words Worth Books and The New Quarterly (Kitchener-Waterloo)

• Word on the Street (Toronto)

• The City Seminar: The Ghost Economy, York University, November 4, 2016.