Around the pool of the CleanTechnica worldwide virtual headquarters on top of Mount Elonpus, we did celebrate Zach’s 300 Tesla Jedi Masters masterpiece. This led to a discussion about the safety of Full Self Driving (FSD). It is often said that FSD needs to be 10 times better or safer than a human driver. “Better or Safer” was at the heart of this recent discussion — they are not the same.
A thought exercise: We define 10 times safer as causing 1/10th the number of accidents and casualties as human drivers. We can simply count the number of accidents over 10 billion miles driven by an autonomous robot driver in all types of traffic, right? That is a fair method, isn’t it?
In this exercise, we replace all drivers in the USA with FSD robots that are certified to be 10 times safer. What will happen is a complete disaster.
Let’s say that the 40,000 yearly traffic deaths will be down to only 4,000. Sounds great, until you look at the accidents. Most, if not all, will be robot-driver faults most humans would never make. They will not be acceptable to the public.
The AI solves problems humans cannot solve and does not do the stupid things that cause most accidents, like getting distracted or being drunk. The robot will make mistakes a human, with his/her far greater experience, will never make. Those 4,000 deaths from stupid FSD faults will result in at least 4,000 wrongful death lawsuits. As well as even more stories about the dangers of FSD. This is clearly not the way to go. Let’s now look at “10 times better.”
When thinking of being a 10 times better driver — better in solving difficult situations — we get a completely different picture. In that case, we can remove all the victims of DUI, texting, dozing off while driving, disciplining the kids in the back seat, or plain stupid behavior like reckless speeding. That is about 90% of all accidents.
The other ~10% of accidents are the result of the driver not being fast enough, not being experienced enough, having only 2 eyes (with none in the back of his head), and other normal human limitations that prevent the driver from resolving a complex situation in time to avoid an accident.
Again, we can replace all drivers in the USA with these 10-times-better robotdrivers. There will be fatalities, but how many we don’t know. It will be a fraction of the 4,000 that human drivers could not avoid.
This time the accidents will be in situations a human driver also would have failed. Because a 10 times better AI driver is a robot that can solve all problems a human can solve and then some. The robot is never distracted. It has 360 degree vision. It can monitor multiple difficult situations at the same time. It reacts in milliseconds. As soon as the accident is unavoidable, it will call 911.
What it cannot do is get out of the vehicle and help the victims.
This is the difference between 10 times safer and 10 times better. In my uneducated opinion, I would be okay with introducing FSD when it is only 2-3 times better than a human driver. That would be enough to avoid most accidents.
Perhaps the robot driver should not drive everywhere initially. Geofencing it to areas away from situations that are too complex, like Place Charles de Gaulle in Paris during rush hour in the tourist season. But letting it drive on more than that silly road in Michigan between Ann Arbor and Detroit.
What is left is how to decide the robot AI is a better driver than most humans. It is tough in the USA, where there is no standardized, difficult evaluation of driving skills.
It is simpler in Western and Central Europe, where a driver’s license is much harder to obtain. You have to drive 30–45 minutes in big city traffic, where it is always like rush hour, without a single mistake. The examiner is very experienced in judging the driving. The candidate should not be too aggressive nor too timid, should see and anticipate problems, should interact with other drivers in a clear and polite way, etc.
Have a few dozen of these examiners spend a few days each with the AI robot driver and they can give a very, very clear report on the capabilities and shortcomings of the AI as a driver.
Now, in the case of an accident, we can say: “Accidents happen. Even AI could not prevent this.”