Cars and artificial intelligence: The seven most important questions
When things learn and make decisions, they become more like us. Artificial intelligence plays a key role in autonomous driving.
Cars will become more “social,” since they will communicate with everything in their surroundings: with other cars, for instance, to avoid congestion and accidents; with other means of transport; with the infrastructure around them, including traffic lights, street lamps or restaurants. And it will also be possible for people who cannot drive, like the visually impaired or children, to use them.
Only through interaction can there be intelligence. The individual components work together in a way similar to how people function: sensors provide computers with data just as our senses send information to the brain, where they are processed and transformed into inputs. In people, these signals are passed on to arms, legs and organs. In a self-driving car, the central computer sends its instructions to the brakes, steering wheel or air-conditioning.
They compare the current situation with historic data – with a large, extremely large quantity of historic data. When a pattern emerges, they act in the same manner that proved correct in the past. What results is a new current situation which is in turn compared with historic data. As soon as a discrepancy with a past pattern is identified, a more appropriate pattern is sought – and on and on.
From data. The more data, the better. The systems are trained when they've been fed enough data; this is done by confronting them with a wide range of challenges – from snow in September to simulations in which children run into the street from between parked cars. If the results aren't satisfactory, additional data is added.
Even if all possible ways in which cars could crash into each other were calculated down to the last centimetre, artificial intelligence would not know what to do if an airplane suddenly started to land on the highway. In such situations, the system attempts to reach a stable status that doesn’t harm anyone. When encountering an approaching airplane, therefore, it would probably: move to the right, stop, wait – as a human driver would likely do as well.
Yes, digital technology is already so advanced: Hundreds of thousands of test kilometres have been successfully driven. The concept we call “deep learning” plays a critical role here: The overlapping of many layers of data – the wealth of experience of many networked vehicles, so to speak – enables the software to also recognise and classify developments that are practically incalculable.
Every technology can fail at some point. Overall, however, experts forecast that autonomous driving will lead to a drastic reduction in the number of accidents – and therefore in the number of people hurt or killed in road traffic as well.
Source: TOGETHER.net – Volkswagen AG