My knowledge of AI is limited but I have worked with neural nets a bit. The key point is you don't 'program' any behaviour at all. Instead you create a blank net, throw an input at it and then tell it what the correct output should be. It 'learns' outputs for inputs by adjusting the lengths of the connections between the nodes in a similar manner to the way the synapses grow and shrink in the human brain. Later you can throw the same input at it and it should produce the same output.that we know there needs to some basic code before the A.I can't learn its own
AI isn't programmed, it's trained. That's what allows it to make best guesses when faced with an input it hasn't encountered before. It's also what allows it to make mistakes. Train a Hopfield network to recognise a captial U. Then give it a capital V as input. It will probably tell you that V was a U because it's never been trained to recognise the difference and they look similar so it will make a best guess.




Reply With Quote