Politique Internationale — How did you become a specialist in autonomous cars?
Jean-François Bonnefon — This goes back a long way. Twenty years ago, I did my thesis in psychology on artificial intelligence (AI), looking at how human intelligence can program machines. Then I returned to more traditional psychological questions. Until, a few years ago, AI began to gain real momentum, which has continued ever since. For me, it was a bit of a revelation, in the sense that moral and technological issues have never interacted so much. A field like the autonomous car illustrates this mix: how can the human mind cope with leaving so much room for manoeuvre to the machine? We’re opening up a field of thought that’s both gigantic and extremely stimulating.
P. I. — Apart from autonomous cars, is there any other field in industry and/or services where the issue of delegating trust is so prevalent?
J.-F. B. — The car is a specific machine in many respects. It operates in an open environment, which makes the safety of passengers, road users and pedestrians particularly crucial. It would take too long to list all the factors involved in collisions and accidents in general, as they are so numerous. What is unique, however, is time management: with a few notable exceptions, when danger arises, reaction time is extremely short. It’s so short that, in the case of the autonomous car, there isn’t even the leeway to bring the human into the loop and try to find a way out.
P. I. — Is the autonomous car coming tomorrow, the day after tomorrow or even later?
J.-F. B. — On this subject, caution is the order of the day. I remember that ten years ago, in 2015, very close prospects were evoked for the rise of the autonomous car. A decade has gone by and tangible advances are still slow in coming. But that doesn’t mean we can’t experiment on a fairly large scale, for example with cab fleets in the United States. Still on the other side of the Atlantic, batteries of tests have been carried out in Arizona, where weather conditions, with virtually no rain or fog, lend themselves well to such trials. In any case, the current situation is a bit like a snake biting its own tail. Implementing autonomous cars is one thing, but we still need an indisputable statistical tool that can verify the maximum reliability of large-scale vehicle fleets: that is the minimum required by a safety ethic. In other words, no fleet, no mass launch; and conversely, no launch, no fleet development.
P. I. — You’ve developed a particularly stimulating instrument nicknamed the Moral Machine. What is it?
J.-F. B. — This tool refers to the lack of references in the field of autonomous vehicles. With the team I lead at CNRS, we wanted to compensate for the inability of the social sciences to grasp the phenomenon. We knew that such an approach would be highly beneficial, provided it could be based on an ultra-popular, even viral, questionnaire. This is exactly what happened: our grid has already been filled in by almost 10 million people worldwide, providing an extraordinary database based on perceptions of the autonomous car. The results are very interesting: we can see, for example, that in almost all countries, women and children are the priority for protection. Above all, the approaches are sufficiently detailed to avoid the cultural stereotypes that inevitably threaten the study of a piece of equipment like the car. The questionnaire makes it possible to touch on very concrete points; for example, in regions of the world where people happily jaywalk, the perception of the autonomous vehicle is not the same as in countries where people wisely wait at traffic lights.
P. I. — What about the average citizen? What do they think of the autonomous car? Have they been questioned about it? Are they interested, passionate or indifferent?
J.-F. B. — Here again, we need to be cautious. It’s true that a great many surveys have been carried out all over the world, but it’s hard to come up with a rational interpretation. On the one hand, people lack the benchmarks to make accurate judgements; on the other, the price factor comes into play. If people realize one day that they have to pay an extra 20,000 euros for an autonomous car, I’m not sure they’ll be too keen. And let’s not forget all those who enjoy driving. In consequence, the acceptability of autonomous vehicles remains subject to a number of uncertainties. The only thing that remains – and this is borne out by trials – is the time saved on transport with this type of vehicle: thanks to AI, it moves more easily than a motorist. This argument is far from negligible.
P. I. — You mentioned the tests carried out in the USA, and the cab fleets that have taken the leap. Can the autonomous car make inroads in Europe, at a time when efforts on the Old Continent are focused on the development of electric cars?
J.-F. B. — Little by little, despite the obstacles I mentioned earlier, we can see an autonomous car ecosystem taking shape. Frankly, I don’t see why Europe should be left by the wayside. In fact, there’s no definite reason why it should.
P. I. — By definition, an autonomous car is more than human. Does this mean that humans delegate all their trust to machines, or do they have the means to control them?
J.-F. B. — In this field, as in many others, the question of delegating trust is paramount. It goes without saying that the machine is never left completely alone in the wild, without any control. That said, the functioning of the autonomous car suggests two main directions. In the first case, the vehicle has almost complete control over operations, that is, a maximum level of trust. In the second case, we’re closer to a joint operation between man and machine. This second scenario is the most complicated to grasp, for a number of reasons. On a technical level in particular, how should human beings and AI work together? Which processes make the most sense? How can the technology make the most of both protagonists? The legal aspect is also a source of great complexity: in the event of a collision, for example, who should be held primarily responsible? It’s easy to get bogged down in endless discussions. That’s why there’s an urgent need for clarity. We are well aware that nothing will be formalized definitively for at least another ten years, but the implementation of autonomous vehicles will inevitably require a regulatory arsenal. The task ahead is immense, in keeping with the density of automobile traffic and its many ramifications: the road is made up of freeways, but also and above all of a plethora of smaller roads, in other words, a myriad of details to be examined.
P. I. — When it comes to regulation, case law often plays an important role. Incidents have presumably already occurred, with lessons to be learned.
J.-F. B. — One night in Arizona, an autonomous car – with an operator on board – collided with a person who died in the accident. The question of responsibility immediately arose. But as we dug deeper into the case, things became even more complicated: we realized that the victim had been hit while crossing a highway; we also discovered that he was under the influence of narcotics. And there’s more: at the time of the accident, the operator had been glued to her smartphone for half an hour, watching an episode of The Voice. As for the car, it was unable to brake autonomously. A whole series of data thus came to bear on the case, which became a knot that was impossible to untie.
P. I. — You work on psychology and morality. The approach to the autonomous car is full of equations that are close to intellectual, intimate and emotional functioning. How do you clear your way through this field?
J.-F. B. — Let’s start by saying that artificial intelligence is revolutionizing psychology. Let’s take the case of the autonomous car: if an artificial intelligence foresees the risk of hitting a group of children, should it perform a manoeuvre to sacrifice the passengers? This scenario is just one of millions of possibilities, since the circumstances of an accident can vary so widely. The inflation of threats on the road leads to quasi-philosophical issues: is one life superior to another? How far should technological progress go? Is everything possible, even the unexpected? The classic answer that comes to mind is that “there is no right answer”. But that is too easy to settle for. We need to be as precise as possible in our approach to the autonomous car and its interaction with human psychology. And we have to be aware that we can’t build the future of the automotive industry by spending months, even years, studying the rarest accident scenarios.
P. I. — Getting back to regulation, and therefore responsibility, can we agree, despite the complexity of the issues, on the duties of each player? Between automakers, AI suppliers, on-board operators...
J.-F. B. — At first, automakers said they would take full responsibility in the event of an accident. Then, in 2016, a Mercedes manager stated that the priority was to protect the occupants of the autonomous car. His words set off a firestorm, with a host of detractors accusing artificial intelligence of prioritizing the comfort of the wealthy. The answer to the question of duties and responsibilities can only be a collective one. If everyone works in their own corner, there’s a risk of multiplying the number of legal contortions. On the contrary, experts, lawyers, industrialists and all the players involved must be able to work together to define a global operating scheme.
P. I. — Every day, we see just how difficult it has become for cars, motorbikes, cyclists and scooters to coexist. Doesn’t the introduction of the autonomous vehicle add another layer of antagonism?
J.-F. B. — Should we see it as a challenge? It underlines a little further the need to move forward within a framework.
P. I. — Perhaps we should have started with this question: is the development of autonomous cars good for the planet?
J.-F. B. — Ecological concerns are obviously not absent from the issues surrounding the autonomous car. Especially since, yes, this vehicle is good for the planet. On the one hand, it is destined to run on electric power; on the other hand, it has been estimated that artificial intelligence is capable of reducing by 10% the flow of traffic linked to individualized driving styles. A 10% reduction means fewer CO2 emissions.