1
$\begingroup$

Some AI's, such as some chess players, are extremely well coded and have defeated humans in several matches. But I think that they won simply because computers can make calculations way faster than humans can not because they learned from their opponents.

If you put an AI against itself, who will win? Will the game continue indefinitely or will the game eventually finish because the AI plays randomly?

So, are machine learning and self-learning really possible?

$\endgroup$
2
$\begingroup$

Yes of course they are possible, I have built some! You are correct that there is an aspect of randomness to the process of machine learning but it is more accurate to describe this as trial and error. Each successive try in a machine learning system is evaluated against a goal and if it is an improvement or is closer to the goal, then this try is stored and some aspect that made it successful is incorporated into similar trys for this type of input. Therefore, machines learn by trying all possible combinations of a problem, albeit with some clever human described short cuts or "heuristics" to make the task easier.

$\endgroup$
  • $\begingroup$ But will machines learn by trying all possible combinations ? $\endgroup$ – BrnPer Jul 5 '17 at 7:55
  • 1
    $\begingroup$ @BrnPer There are different types of AI with different learning mechanisms. If a problem can be fully analyzed and all possible combinations considered (take TicTacToe for example), the AI can play in a perfect way. Most problems in modern AI research are not of that nature and therefore the AIs must learn to judge situations without knowing all possibilities. $\endgroup$ – Demento Aug 23 '17 at 22:33
1
$\begingroup$

Machine learning and self-learning are of course possible, and there're many successive cases!

You need to know this: machines won't think like humans. Machines form a statistical model and calibrate the model. A good model is a model that does what it's supposed to do accurately.

$\endgroup$
  • $\begingroup$ But will statistical models predict every decision? Will humans give all the models machine need or will machine make their own models? Because if it's self-learning, it can learn by itself...its some questions I've. Sorry if I wasn't clear enough $\endgroup$ – BrnPer Jul 5 '17 at 7:55
0
$\begingroup$

I like your focus on optimization (re: "One of my productive days was throwing away 1,000 lines of code";)

I think the problem with this question is it your supposition is incorrect. Check out Giraffe Chess for more info on self-learning. [Note that Giraffe Chess is a result, not a hypothesis, and Lai was subsequently tapped by DeepMind.]

I'd also recommend getting familiar with concepts like non-deterministic polynomial time and obsessing over intractability.

Regarding random choices, "monte carlo" has had major successes of late, but I suspect the success is related to processor speed, and may not be sufficient in greater complexity spaces, or problems where rationality is heavily bounded.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.