We want self-driving cars on our roads, even if they cause accidents

If we want to save as many human lives as possible in the traffic, we should start replacing our vehicle fleet to self-driving cars as soon as they are as safe as the average driver. Bureaucrats and politicians don’t have the information nor the incentive to get this right, instead they will want to avoid headlines by restricting self-driving cars, which means people who could have been saved will die. 

When should we allow self-driving cars on our roads? The intuitive answer is something along the lines of “when they are safe enough”. The question then become what we mean with “safe enough”. That depend on the goal. If our goal is to minimize the number of car crashes caused by self driving cars, then the answer is simple. Never. This is of course not an optimal policy, because if we care about saving human lives and minimizing damages, then it doesn’t matter if it’s the error of a machine or a human at fault.

If we care about road safety in general, then we must accept the fact that people will be killed by self-driving cars. If a self-driving car is e tenth as dangerous as a human driver, then for any death caused by a self-driving car, there are on net nine people who are saved from dying on the road. Because the human driver who would have killed them wasn’t steering the car they were in. If all lives are valued equally, then we would want that self driving car to be out on the road, killing that one person.

The safety of self-driving cars improve with time as technology advance, and as soon as self-driving cars as good as human drivers, we should start to replace our vehicle fleet. And when I say better than human drivers, I mean the average driver on the road, not the best human drivers. If self-driving cars are more likely to replace bad drivers than good drivers, then they should be allowed sooner. And if allowing self-driving cars can speed up the innovation in traffic safety then they should be allowed even sooner. For many people, this is not intuitive.

Unfortunately, there is a difference between what should happen and what actually will happen. There is as a matter of fact reason to suspect that the people who have power to influence the decision of when to allow self-driving cars will fail to get it right. I don’t doubt that the bureaucrats and politicians who make regulation on self-driving cars have good intentions, and wish to see fewer people killed on the road. But all deaths are not treated equally.

An accident caused by a self-driving car will get a lot of media attention. When a pedestrian was killed by an autonomous car in Arizona in march 2018, it became known across the globe. The people who are saved by self-driving cars however will not show up in any statistic and will motivate no headlines.

It’s well established in psychology research that media coverage influence our perception about the frequency of events more than statistics. The nobel laureate Daniel Kahneman referred to this as the availability heuristic. Politicians who want to be reelected, and the bureaucrats under them who wish to keep their jobs, tend to want to avoid being blamed for deaths in headlines. On the issue of deciding when self-driving cars are “safe enough”. It’s easy enough to see what direction they might err. Even if these people understand the logic I’ve written of here, they have incentive to treat accidents caused by self-driving cars as much worse than accidents caused by human drivers. As a consequence, self-driving cars will come to the market way too late.

This is not a new problem at all. The government authority in charge of regulating pharmaceuticals in the USA frequently get critique for being too strict with approving new drugs to become available to the public. It’s the same story as with the self-driving cars. Any death caused by an unsafe drug gets headlines. All the people who die because the drug that could save them isn’t available due to bureaucracy don’t get headlines. So they don’t count.

The people who have the ability to assess when self-driving cars are safe enough are the people working on developing them, and if they care about their brand, they will not make that decision lightly. They would have the good incentives to make sure their self-driving vehicles actually are “safe enough”. Instead, they now have to prioritize meeting the regulatory standards set up by the bureaucrats. If we care about traffic safety, we should allow self-driving cars on the streets right now.

Can game developers get people to play more and have a better gaming experience by restricting playing?

A while back I read the widely known book by nobel laureate Daniel Kahneman “Thinking fast and slow”. In it he describe a psychological heuristic called peak-end rule. It describes the observation that people’s perception about an activity depend heavily on how they experience its end rather than say the sum of or the average moment of the experience.

Kahneman and his associates published a study in 1993 with the titled “When More Pain Is Preferred to Less: Adding a Better End”. In a wikipedia article, the experiment they conducted is described in the following way:

Participants were subjected to two different versions of a single unpleasant experience. The first trial had subjects submerge a hand in 14 °C water for 60 seconds. The second trial had subjects submerge the other hand in 14 °C water for 60 seconds, but then keep their hand submerged for an additional 30 seconds, during which the temperature was raised to 15 °C. Subjects were then offered the option of which trial to repeat. Against the law of temporal monotonicity, subjects were more willing to repeat the second trial, despite a prolonged exposure to uncomfortable temperatures.”


The participant’s remembered the second trial more favourably even though they experienced unpleasing temperatures for a longer period. This finding have very practical applications in many areas such as business, health care and even dating. The ending of a customer interaction will be highly influential for whether or not the customer will come back to your business in the future. By making the ending of a painful medical treatment slightly less painful, patients will have a better experience, and perhaps have an easier recovery. If you want to maximise your chances of getting that second date, don’t argue over whether or not you should split the bill after the restaurant visit on the first date.

Applied to game economics, this means player’s opinion of a video game will be highly influenced by the ending of the gaming sessions, rather than the overall playing experience. Now what makes someone stop playing a certain game? The economists way of thinking about this is to think about decision making at the margin.

Suppose we are talking about a game that isn’t continuous but consists of subgames such as battlefield, or dota where players play matches. After each match, the player will make an inner calculation whether to play another game or not. As long the marginal benefit is higher than the marginal cost, they will continue playing. This means there’s a great chance that the last match that they play each session is the one that gives least satisfaction (if there’s decreasing marginal utility from playing video games). Considering the peak-end rule, this is not at all optimal.

If a gaming session is disrupted by an exogenous factor (for example because the player’s parent demands they go to bed because it’s weekday and they have school tomorrow), the player might get a more positive view of the game than if the player is (a grown up and) allowed to play however much they want and so play until they grow tired or bored.

If game developers deliberately create game session disrupting elements in the game, it’s possible that total amount played will increase. In other words, even though each session is cut down before the player themselves chooses to stop playing, increased number of gaming sessions may lead to a larger total playing time – and a better playing experience.

Creating session disruption elements in a game might be a sensitive thing to do however, and different individuals may respond differently to such mechanisms. In freemium games, only allowing players to play for a limited time (for free) can be one way of doing this. Including natural waiting periodsis another way (in Simcity BuildIt, players have to follow specified time schedules to harvest/farm/cash in on resources and so continue playing).

This is a hypothesis still. But a testable one! As I’ve written before, video games are the perfect environment for social science randomized controlled trial experiments. Here’s one thing that would be cool to test.