A “self driving” Tesla killed it’s passenger because it didn’t understand a tractor trailer blocking it’s full field of vision. That accident – though a combination of several factors – again underlines the fact that “self driving” is nothing but a pipe dream.
Update: A self-driving UBER car killed a pedestrian in Tempe, Az. (March 19, 2018)
A few months back I had the opportunity to drive a state of the art 5 series BMW. Packed with sensors it allowed some partial “self driving” . Just push a button and the car regulates it’s speed itself – applying brakes or acceleration depending on the traffic ahead. It was cool and took plenty of work load from me – especially in a boring stop and go traffic situation. But the very next day brought ice and snow overwhelming the sensors. The computer was unable to read the traffic signs or measure the distances thus rendering the BMW dumb as a brick again. The human senses had no problem and I was able to navigate the car without problems.
So – I started to think about the differences between the human senses and our brains compared to what the car “sees” and what it is able to compute. I dug into what is known about the current technology and how a driving computer tries to “understand” the world.
The computer works by recognizing objects and calculating possible motions. It then figures out how and where to move. While this most certainly works most of the time, the computer does not understand the differences between objects that – in it’s field of view – look the same.
A human driver not only sees an “object” – our intelligence also recognizes the differences. Most human driver would understand both situations quite differently. But cars – even with very sophisticated computer brains – do not “understand”. They might be able to tell a human from a trash can – but fail to incorporate experience and intelligence into their decision making.
I don’t think that we are going to see self driving cars all around us very soon. Sensors are easily fooled by foul weather and idiots with lasers or high intensity flash lights. But the most important reason: One needs at least some amount of intelligence to drive a car. Just reacting is not good enough and will lead to a number of problems in the future.
Though everybody speaks about “Artificial Intelligence”. In all of my more than 25 years of experience in high tech software engineering and -development I haven’t seen anything close to the “brain power” of my dog. Because “intelligence” is much more than knowledge or the interpretation of sensor data. It is the combination of intuition, instinct, experience with learning and the ability to draw conclusions from very complicated and contradicting facts. Just look at the image above. An intelligent human would immediately understand the meaning of this traffic sign. A computer would have difficulties to even pattern recognize this stop sign. And it would most certainly not understand – not comprehend – the meaning under the current circumstance.
What is worse than something not functioning properly? Something that works great most of the time but may fail miserably in certain situations – and maybe kill you (or others). We should enjoy all the technological advances that make our life easier. But we should never -ever- rely on technology that just isn’t ready yet. And might not be ready for another 50 years.
So – let’s keep our hands on the wheel. And let’s make sure our law makers won’t fall for cuddly cars “self-” driving around in nice weather in carefully selected environments.