The idea behind the study was to see how the public felt self-driving cars should make what are effectively moral decisions. Specifically, in a dilemma where there was a choice to be made between, say, killing one or other person or groups of people, how should the car decide? As a concept, Bonnefon makes it clear this is a descendent of the classic 'trolley' problem where participants are asked to decide, for example, whether or not to switch the points so a tram that is currently going to kill five people will be switched to a track where it will kill one person who wouldn't otherwise have been harmed.
The many variants of the trolley problem are hard enough to deal with, but as Bonnefon makes clear, there are more complex determinants when dealing with autonomous vehicles. Apart from anything else, we would have to tell the car how to decide between killing its passenger(s) and a pedestrian. Would anyone buy a car if they were conscious that it would take the choice to intentionally sacrifice them where necessary?
Bonnefon and his colleagues decided to go beyond the usual scope of a study like this (which often just asks a few hundred students) to try to address humanity at large with a simple online system that presented each participant with a short set of preferences between, say, killing a jaywalking child or mounting the pavement and killing an elderly person who just happens to be passing.
In part, this is the story of how such a project works, giving rare insights into what researchers do, down to attempting to get into the big name journals and the fun of getting a bad peer review. This is really interesting, though in an attempt to make the story more approachable, Bonnefon does give us too much detail of the characters of his colleagues and where and how they met. Central to the book, though, is what might be done about autonomous vehicles. We are told of a German commission's findings on the subject, which is then contrasted with the findings from the study. (The main disparity is that the commission felt that cars should not be discriminatory, where the public at large thought they should be ageist.)
There are some omissions. When the results were published, one of the criticisms Bonnefon and colleagues got was that they asked people to make choices based on information that autonomous vehicles could never access - for example, choosing between a professional and a homeless person, a distinction the car couldn't possibly make. This seems a bizarre thing to do - surely the study should have been limited to choices that a car could actually make, and Bonnefon does not adequately defend the decision to include these extras.
The book also omits one of the key factors that could influence self-driving car uptake. Bonnefon spends quite some time on the idea that if the cars were only 10 per cent safer than a human driver, then anyone who thought they were at least 10 per cent better than average (which is most people) wouldn't want one. However, he didn't go into the potential problem that if autonomous cars are killing thousands of people, that would be thousands of real families seeing the negative side (and probably suing manufacturers), but the lives saved are statistical, so have nowhere near the same weight. It seems likely that for statistical benefits to outweigh damage caused (as is the case with a vaccine) autonomous vehicles would have to save many more lives than they kill.
There is a very interesting mention of the highly negative reaction received by a car manufacturer's representative, who pointed out that it may be better to definitely save a passenger when weighed against just a possibility of saving a pedestrian, but again Bonnefon doesn't adequately address this very reasonable view. The same issue comes up with many of the analyses of the trolley problem I've seen, where participants will usually switch a trolley to a different track with a remote switch, killing an individual, but won't push an individual off a bridge to their death if it would stop the trolley. This is put down to the remoteness of throwing the switch, but what hardly ever gets mentioned is that there can be a lot more certainty about the effect of throwing the switch compared with pushing someone off a bridge and hoping they somehow stop the trolley.
I would, then, have liked to have seen more detail here - but it's an enjoyable and easy read that both opens a window on what it's like to run such a study and gets the reader thinking about this very real decision that would have to be made if autonomous vehicles are to be allowed free access to our roads (something I suspect won't happen for a long time to come).
Hardback:  |
| Bookshop.org |
Review by Brian Clegg
Comments
Post a Comment