Weak AI

What we call AI today is generally pretty weak artificial intelligence. There are still many tasks that cannot be solved efficiently. The traveling salesman problem is a case in point. The task is to choose an order for visiting several places in such a way that no station except the first is visited more than once; the traveling distance of the traveling salesman should be as short as possible and the first station should be identical to the last one. Mathematically, such a thing is called a combinatorial optimization problem, and such problems can be solved non-deterministically in polynomial time (simply speaking, this takes – too long).

Today's AI systems also have problems with things that come easily to a small child, for example, like intuitive physics (gravity, pressure, motion). Walking is actually quite easy to learn, just don't ask yourself how, and better don't try to teach that to computers either (the company Bosten Dynamics succeeds very well – the videos of their walking humanoid robots make most people's hair stand on end).

AI systems also usually learn only one specific task (e.g., recognize what is in the picture), and the effort of learning is enormous. Text-to-speech synthesis is also currently still one of the challenges.