I bought an Intel Movidius Neural Compute stick a few weeks ago. Even though I can use it with the examples, I want to actually use it for something! The documentation is messy, and hard to work through.

I've already looked through some of the websites that have been mentioned in other posts.

I specifically want to make an inference/reward-based AI that tries to guess a bunch of numbers over a period of time. The AI has to guess the best 10 guesses, and then it gets shown the set. The closer answers are rewarded, and the farther answers are penalised.

In the end, I want it to try and guess the most likely set of numbers in a random number generator, kind of like how a quantum computer outputs answers.

This could be used for all sorts of things! Even if it's already built, I want to program it from scratch so I can learn how this sort of thing works (teach a man to fish type of thing).

So, considering my goals above, how should I start? Also, could this answer questions with close to the same accuracy as a quantum computer, if it's fed enough data?

  • $\begingroup$ John's answer is correct for modern PRNGs (unless you give the AI some clue such as the source code - even then this is beyond any current technique). However, a general guessing game might be more possible. You need to design it so that there is some meaningful clue to what the correct answer will be - do you have something in mind for that stage? $\endgroup$ Aug 8 '18 at 19:19

Welcome to the AI stack Jake.

This probably isn't going to be possible. Modern Psudo-random number generators, like Mersenne Twister, are designed not to have any patterns in them, so there's nothing to learn from.

You could however, try something like predicting the values of a broken random number generator, like RANDU. These aren't used anymore, precisely because it's possibly to predict future numbers on the basis of past numbers.

Hope this helps.


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.