Testing autonomous vehicles, we are testing our trust in AI and taking a lesson in ethics

My article in Spider’s Web, 6 April 2019.
Autonomous cars are a vivid example of how technological progress can polarise our views. Personally, I believe they offer a chance to raise safety standards on the roads. And I am not surprised by enthusiasts who claim that in 15 years we will not be able to comprehend that people once drove cars.
For things to go in the right direction, however, we need universal standards for deploying the new technology. By investing huge sums, the manufacturers of autonomous vehicles have, in a sense, made a promise to society: broad acceptance of these cars on the roads could prevent tens of thousands of road deaths every year (in the United States alone, 30,000 to 40,000 people die in accidents each year).
Supporters of autonomous vehicles often cite repeated research findings showing that 90–95 percent of accidents are caused by the driver, not the technology.
Last year, in the American city of Tempe, a woman crossing the road was killed by an autonomous vehicle. Neither the car’s safety systems nor the driver inside reacted. The event triggered a flood of comments and showed how many perspectives we can take on it.
Some, a little maliciously, called the accident historic: artificial intelligence had claimed its first human victim. Opinions that defective technology produced by private, profit-driven companies should not be tested on public roads appeared instantly. Sceptics and critics argued that people would never accept autonomous vehicles, citing among other things a Reuters/Ipsos survey from January. Even before the Tempe accident, two thirds of Americans did not trust autonomous vehicles.
The accident also brought to light issues such as overworked test drivers (too many hours behind the wheel) and the decision to drop the second person in the cabin (for a long time, two drivers had always monitored the vehicle during test drives). The company testing the vehicle had also reportedly chosen a strategy of increasing ride comfort (by limiting frequent, sudden braking) at the cost of making the vehicle less sensitive to obstacles. Journalists also obtained internal reports showing that during tests the vehicles struggled to drive even one kilometre without driver intervention or braking.
Footage of the accident shows that the driver was not watching the road attentively but was busy with a phone. Paradoxically, this could confirm the thesis that the most dangerous thing on the road is the person behind the wheel. But that would probably be an oversimplification.
Autonomous vehicles are equipped with lidar sensors and radars. The equipment is meant to monitor the vehicle’s surroundings and detect any obstacles. However, there are no universal standards defining how such systems should work, which parameters should be taken into account and when the car should respond by braking.
As part of the ongoing investigation, the company admitted that the lidars, radars and cameras did register the person on the road (a point strongly emphasised by the manufacturer of the laser sensors). The problem therefore lay in how the car’s system was tuned – in the software and the machine learning algorithms. Settings with a higher tolerance for obstacles had been chosen.
This brings us to an important, perhaps the key, question in the whole situation: how should a vehicle’s safety system be “set”, and who should make the key decisions? The fact that there are many options does not make the problem simpler.
Since 2016, researchers at the MIT Media Lab have been studying autonomous vehicles. They launched the Moral Machine (moralmachine.mit.edu), a kind of game that tests our ethical choices related to autonomous vehicles. The website says it shows “moral dilemmas, where a driverless car must choose the lesser of two evils, such as killing two passengers or five pedestrians”. The platform shows participants various road scenarios, many of them involving different combinations of gender and age, and even animals crossing the road.
A few weeks ago the authors presented partial conclusions from more than 4 million responses. Participants were far more likely to want to protect the life of a child than that of an adult. But when they had to decide who should survive an extremely dangerous situation – the passenger, the driver or the pedestrian – the results were no longer clear-cut. For example, as many as 40 percent of participants said the vehicle should protect its passengers first, even at the cost of pedestrians’ lives.
Link to the full article (in Polish)
Related articles
– When will we stop being biological humans?
– Artificial intelligence is the new electricity
– Robots are waiting for judges
– Only God can count as fast: the world of quantum computers


