Self driving cars and the limits of AI

Had the benefit to recently drive a modern car, German manufacturer, enjoying assisted driving.

We are mid terms toward complete autonomous driving: this car in particular reads from the environment, offers parking sensors, is able to detect traffic in front and can assist you stay your course, suggesting you might be veering off track while turning corners or simply in driving in lanes. It also offers the ability to set the speed at the desired level and select distancing options to the traffic in front.

Unfortunately all this driving assistance doesn’t play out all too well.

When the onboard computer decides you are going off track, it suggests a correction through an hardening or gentle counter steering of the power assisted wheel that is at times rather useless – not enough to actually set the right turn radius – or fully undesirable, as when you are changing lanes to overtake, and the systems tries to keep you in check.

Other discomforting findings include the inability to anticipate. The sensors will only detect a car when it is in range, obviously not before it leaves its own lane to overtake. What is obvious to the average driver in terms of situational awareness(“this guy in front is going to come out to overtake anytime soon, so I’d better adjust speed now and leave him space”) is much less obvious for this car’s assistance system.

On the other hand you also know if you are driving on a speedway that the guy in front is quite close but will also get out of the way soon, so there is no need to reduce speed to his level. He will not be there much before you enter the security distance of separation.

This is where the machine and AI system are still falling short. The ability to play multi level intelligence. As I have argued in other posts in this blog, the trick accomplished by our mind is – yes – to integrate information through our sensors and make sense of it. At the same time we are able to play the battle, define the tactic and follow the strategy at the same time. Our mind is wired to play at different levels at the same time: the game, the rules of the game, the rules of the rules of the game and the same. Every time it toggles up and down this hierarchy of thinking it does so in the swiftest possible manner, without us being aware and making it seem an effortless task.

Machines have to follow different algorithms, each targeted at solving a specific problem and finding a way to orchestrate everything is no ease feat.

One of the reasons why machines eventually won on humans at go (see other article on my blog) was precisely this ability to move up and down the hierarchy of control systems, even though this particular part was partly monitored and assisted by a human.

The ability to “meta” that is to transition to the next level, from rules to meta rules, from meta rules to meta meta rules is so profoundly linked to the essence of human intelligence (and hardcore mathematics .. G?del theorem any takers see other blog article) that I have the feeling it is the key to conscience, whatever that might be and to machine finally becoming self aware, whatever that might entail.

For the time being – luckily???– we are only left with hope and ill functioning assisted driving vehicles.

For more tech and business articles please check out my blog