I was thinking about AI the other night and feel that I may have stumbled on a very simple way to determining the current AI technology level. The scale has only three points and our current best systems are, I believe, around level two. The scale requires one to think about where a piece of AI fits and is perhaps a little biased toward AI that interacts with the real world. Even so I believe it is widely applicable to most AI systems. Hereâs the scale:
- Level One: The AI is able to warn an operator that an error condition is occurring.
- Level Two: The AI is able to tell an operator that an error condition is occurring and be specific about the cause and maybe offer a simple solution.
- Level Three: The AI decides for itself an error condition is occurring and makes a reasoned decision about how to remedy the problem.
To understand this scale I offer this example. Imagine a robot that is doing the washing up at the sink. Its hands are in the water which is hot and causing the robot to warm up. If the robots core temperature gets to high it may malfunction.
A level one AI will have a sensor for its core temperature and will inform the user when it detects that itâs over heating. It makes no attempt to diagnose why itâs over heating and offers no ways to remedy the problem.
A level two AI will have multiple sensors and be able to determine that the hot water is what is causing it to over heat. The subtle difference here is that the robot may be standing in a warm room but it needs to be able to determine that it is the water that is causing it to overheat and not the room. Simply sensing that something is hot is not enough to qualify for level two AI â the link between the heat of the water and overheating needs to be made. Level two AI might offer the solution of removing its hands from the water.
A level three AI will not only be able to make the link between the hot water and over heating but will take remedial action to correct the situation after weighing up all possibilities. This weighing up is important because the robot removing its hands from the water while causing its core temperature to drop also results in a failure of itâs orders â to do the washing up. The AI system needs to be able to weigh up the importance of completing its task with the importance of self preservation.
I believe we have systems that are just about capable of level two behaviour but we have nothing that approaches level three. Certainly there are systems that can give level three like behaviour in very limited situations but real level three AI requires an excellent understanding of the world.