3 ms·
> 5) Intuitive leaps. e.g. Rubics cube. You're trolling, right? Intuitive leaps are the bread and butter of neural nets.
by visarga 9y ago
> 5) Intuitive leaps. e.g. Rubics cube.
You're trolling, right? Intuitive leaps are the bread and butter of neural nets.
- justinjlynn 9y agoIn fact, one might argue that's all they do -- depending on how one defines "intuition".
- goatlover 9y agoIn a vary narrow sense, sure. But humans are great at generalizing across different domains, and finding new connections and patterns.
- arichard123 9y agoNo, I wasn't trolling. I think we have a different concept of what an intuitive leap is. A neural net, as far as my understanding goes, needs training in the field that it makes decisions in. I think you believe that process itself contains intuitive leaps. Neural nets trained to walk develop weird walking styles, and I think that you think that is evidence of an intuitive leap. If that's what you mean, I would disagree. 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.