Reply to post: Re: Neural net - nobody really understands how they work.

Q: How does hydrogen turn into a metal? A: Hang on a second, I need to train my AI supercomputer first

Paul Kinsler

Re: Neural net - nobody really understands how they work.

Actually, we know exactly how they work - it's specified by the architecture, the weights, the thresholding function, the back-propagation algorithm, the inputs, training data, etc (or, if you prefer, the code that implements it). The problem is, except for very very simple implementations, there is far too much information to understand, and the system is far too complicated to comprehend how - heuristically - we might get from some given input to a given output [1].

It seems to me that it's more like an "Artificial Intutition" than an "Artificial Intelligence". It seems to work quite well in many cases, but there is no clear reasoning explaining why or why not, and so you never really know whether you can (or should) trust its answers (or, if you prefer, its proposals, guesses, etc) it or not [2]. But it's probably better than randomised guessing, ... except when it isn't.

.

[1] Which is, of course, why we get computers to do it for us.

[2] Thank goodness people are never like that. :-)

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

SUBSCRIBE TO OUR WEEKLY TECH NEWSLETTER

Biting the hand that feeds IT © 1998–2020