New technology and business models bring new opportunities and risks. New risks should be examined in a different way to older, familiar ones. Take robot cars, for example.
“Google’s self-driving car caused its first crash”, “Fatal Tesla Self-Driving Car Crash Reminds Us That Robots Aren’t Perfect” , “Google’s self-driving car was just involved in ANOTHER crash“. Accidents involving robot cars are big news. Is this because robot cars are a major threat to road safety? It’s more probably because such accidents are so rare. And, above all, because all new things beyond our individual control are viewed as particular threats. Cars are also something that arouse strong passions and opinions, particularly among men, in this ‘world’s fastest nation’. To many people, a private car is much more than a way of getting from point A to B.
Of course, only few robot cars are on the roads, but Tesla’s own statements reveal that, prior to the above-mentioned accident, their cars had driven 210 million kilometres in traffic, while on ‘autopilot’, without causing any damage. How many of we drivers could make the same boast? Robot cars are often justified on the basis of safety. For example, it has been claimed that ”94 percent of road accidents are caused by human error, and it is said that driverless technology will drastically lower, if not eliminate this factor.” Robot car developer Matti Kutila from VTT confirms that robot cars are mainly about avoiding collisions. On the other hand, statistics show that robot cars are involved in collisions more often than human drivers (USA 2015). Which of these is the ‘alternative truth’?
The truth is that events are not easily and clearly foreseeable in complex adaptive systems such as traffic. If one type of accident is avoided, others can occur. What is more, the current robot car models are not intended for full use in traffic. Even Tesla’s robot cars are intended for routine sections of motorway. The speed of Google cars, on the other hand, is limited to 55 km/h. Decades lie ahead, before the journey to full automation that works in all weather conditions and traffic cultures is completed. Many developers of robot cars around the world (including VTT) have run into the problem that, while 85% of traffic situations can be handled by computers, 15% remain unaccounted for. There are situations, such as certain weather conditions, which occur just once a month. Every improvement by a tenth-of-a-percent is a hard-won battle. For example, the fact that a robot car effectively avoids collisions by braking or swerving could lead to collisions with other road users. Since people will always be present in traffic, robot cars need to take account of the human factor in order to reduce the number of accidents. Even in normal situations, people’s abilities and behaviour differ widely and are, to some extent, difficult to anticipate. On the other hand, we should not forget that people are also needed when robotic cars hit trouble: What would happen, for example, if people tried to push a robot car out of a snowbank?
It also has to be expected that deliberate and even suicidal attempts will be made to obstruct robot cars, as occurs even currently in traffic. Even the hacking of robot cars has to be accounted for, given the number of cyber attacks made on various information systems. This is particularly noteworthy considering that a malfunctioning car would pose a direct threat to many people and a large number of cars using the same software could be made to malfunction simultaneously, and without warning. Luckily, car developers and the authorities are aware of the risks and are also developing robot car safety.
When viewed as problems, new technology and business models also raise time-worn questions. A robot car too must make difficult choices, which has led to sensational headlines: ”Why Self-Driving Cars Must Be Programmed to Kill”. Such a situation might be as follows: Which will I hit: an oncoming lorry or a child running onto the road, if I cannot avoid a collision? This question is basically as old as humanity (if we substitute, say, mammoths for trucks). However, a general public discussion of these ethical issues has begun, alongside the emergence of robot cars and artificial intelligence. There have been demands for public supervision of artificial intelligence, even as technology developers have made their own initiative to improve their practices. Mercedes-Benz has solved the problem associated with ethical choices in such a manner that the car always primarily chooses to protect the passenger. However, many people believe that pedestrians should be primarily protected, even if they would hardly buy a car that failed to prioritise passenger safety.
Risks and opportunities go hand in hand
New risks can always emerge when developing something new. On the other hand, many positive outlooks and expectations are linked to new developments. In addition to new kinds of road safety challenges, there is the risk that robot cars will not fulfil expectations. Various pressures can easily arise in the gap between different expectations and managed development. However, in managing new risks we should advance one step at a time, while gathering information. The development of robot cars seems to be progressing in accordance with this principle. Development work is taken to an advanced stage in laboratories and controlled test environments before robot cars are introduced into traffic. They are currently being tested in traffic, for example, within limited areas in Finland, Sweden and in this year also in London.
Traffic has a major impact on society, from both the positive and negative perspective. That is why it is strongly controlled by legislation, particularly with regard to safety. For example, according to the Vienna Convention on Road Traffic a vehicle must have a driver. Developers of robot cars respond to this by pointing out that the convention does not say that the driver has to be in the vehicle. This idea can be stretched further, by stating that the convention does not, perhaps, forbid the same driver from controlling several vehicles. Legislators may view this as a loophole which needs to be filled. On the other hand, it would make more sense for all concerned if car developers, the authorities, other experts and the ‘general public’ developed regulations and controls that take the best possible account of the risks.
So why is more expected of a robot car than from a person, or is it? If a robot driver were treated like a human driver, it would have to pass a theoretical and practical driving test. With a little luck, even a lower-end robot driver could pass such a test and obtain a driving licence. However, as I mentioned above, a robot driver faces quite different challenges to a human driver. This means that we should assess robot drivers differently.
We have already considered the risks associated with the new technology, and their control, by referring to examples and the development of robot cars. We must begin by clarifying how much we know about the risks and then feeling our way towards learning more further development. We should neither allow the fear of risk to impede development, or become so carried away with development that we make disastrous decisions. Risk management is an activity that ensures success.
VTT has studied the assessment of new and emerging risks. The findings have been summarised in a publication that is available, in Finnish, online: Uutta riskien arviointiin – Tietopohjan merkitys ja uudistamisen keinot. Getting and evaluating information and knowledge is essential in modern risk assessment – as presented in the following illustration of a modern risk assessment process.
Senior Scientist Marinka Lanne also participated in the study of new risks.