general Robots and mankind's future

[Splashee 28,537]

32 minutes ago, Pentium100 said:

AI (or rather artificial neural networks) is kind-of based on how a brain operates, however, right now it is too simple, because it would take a huge amount of processing power to have the complexity of a human brain (or even dog brain, though I think they managed to simulate some fish brain).

Because of that, we see some weird problems, for example, you can take the image of a kitten, give it to an AI and it will identify the image as a "kitten". Now change a few pixels imperceptibly and the AI will say that it's a gun, even though humans would have hard time telling the difference between the original and altered picture. This is because the AI has limited processing power and has not seen a lot. In comparison, over your life, you have probably seen many examples of kittens and cats, and even more examples of things that are not kittens or cats, in addition, you probably have an understanding of what a "kitten" is. The AI can only say that "this collection of pixels is similar enough to the collections of pixels previously identified as a kitten".

I have seen a TED talk where they trained an AI to identify some kind of fish. They had good success rates and decided to see which part of the image the AI uses to identify the fish. The answer - the hands of the fisherman (because most pictures show to the AI were of people posing with the fish).

 

Interesting, I always like watching TED stuff that is given to me.

 

It seems like people are trying to use AI to do human things, where the AI should be used for much simpler tasks. Once the people that work with AI have understood that you cannot use this kind of technology on such complex things, and allow AI to work on a smaller level, like almost on binary level, we might see some "real AI" start developing. That's my take on it.