Return Styles: Pseud0ch, Terminal, Valhalla, NES, Geocities, Blue Moon. Entire thread

There's just too much

Name: Anonymous 2012-06-13 5:41

so many things to learn
not enough time, life running out
the code, a black morass of details mingling infinitely
will i ever understand?

Name: Anonymous 2012-06-13 16:17

>>4

You are thinking of AGI, which means artificial general intelligence. There will never be a quick way of creating an AGI because the smarter it is, the more concepts it has to handle. Why? Because that's how you define intelligence. Or am I wrong?

AI is just a professional way of saying program to do a certain task or set of tasks. When people talk about teaching an AI, it means they are using a natural way to evolve a program so it reaches a particular fitness for a given task. If the fitness function is designed so that the program evolves (converges) as quickly as possible and the fitness value said function agrees with your definition of fitness for the problem, it is a good AI.

Neural networks or support vector machines are just different structures for the I/O system (feedforward or feedback) and the evolution process to take place. You could as well use something like a cellular automaton, but personally I have found that simply using whatever CPU instructions you have available is the best, fastest and overall nicest way.
What matters is the fitness function. You design the fitness function for the algorithm above, be it neural networks or whatever you want. But the "intelligence" of the program comes only from a good fitness function. You also need some sort of recombination/evolution algorithm, but that is a trivial thing to implement and often you find good ones accidentally, since evolution tends to just "spring up" from nowhere.

Newer Posts
Don't change these.
Name: Email:
Entire Thread Thread List