Name: Anonymous 2018-11-08 17:25
So I looked at these AI researchers, using "deep learning" and hiring people to create training data sets:
https://www.bbc.com/news/technology-46055595
and I keep wondering, do they really believe human minds works like that? Because it is obvious, that people use logic to construct objects: it has head lights and, 4 wheels and 4 doors, therefore it is a car. And no training data set will ever cover all cars. The AI has to use logic and make conclusion, while deep learning is relevant only to very basic small parts, like form or texture.
The methods these AI researchers currently use will never allow them to build self driving cars and you can even prove that mathematically: there are infinite number of possible car designs, so you AI has to capture the essence of what is car, not irrelevant parts about it, like its color or form. The problem is similar to how they used early perceptrons, without realizing that single-layer perceptron cant handle non-linear problems.
https://www.bbc.com/news/technology-46055595
and I keep wondering, do they really believe human minds works like that? Because it is obvious, that people use logic to construct objects: it has head lights and, 4 wheels and 4 doors, therefore it is a car. And no training data set will ever cover all cars. The AI has to use logic and make conclusion, while deep learning is relevant only to very basic small parts, like form or texture.
The methods these AI researchers currently use will never allow them to build self driving cars and you can even prove that mathematically: there are infinite number of possible car designs, so you AI has to capture the essence of what is car, not irrelevant parts about it, like its color or form. The problem is similar to how they used early perceptrons, without realizing that single-layer perceptron cant handle non-linear problems.