You're trolling, right? Intuitive leaps are the bread and butter of neural nets.
You're trolling, right? Intuitive leaps are the bread and butter of neural nets.
You train it to walk, and it won't understand how to climb. You train it to climb on a climbing wall and it won't understand that it can't climb on glass, or string, or a waterfall without additional training.
Whereas you and I wouldn't need to try and fail to climb a waterfall. We would just know it isn't going to work. We would also understand, without having tested it, that those climbing skills would enable us to climb a tree, or a rope net, or ladder.
We might even be able to make a judgement about how well we could climb each before disaster.
If I were wrong on this, and I could see an AI trained to do X and then apply it to Y then it would cause me to review my beliefs on AI.