So I’ve been reading about how in chess there are an extremely high combination of moves that can be made and trying to write out every combination is impossible. In an eli5 style, how are algos like alphago able to overcome this fundamental issue? How do they go about solving this problem?
In: 0
Alpha Go is an AI. It doesn’t work on decision trees in the same way as most of the other software you might use. It’s been trained through repetition and practice and has a brain that’s closer to a living creature’s brain than a computer algorithm. As such it’s more likely to see patterns and react to them rather than brute force processing of what might happen if a specific move were taken.
Humans don’t have decision trees (mostly…). We learn what works, what doesn’t, and make choices that tended to kick the ass of computers doing it the hard way. Alpha Go takes the human approach instead, building a brain and teaching it how to play the game by making it actually play the game.
The downside is the need for training. Alpha Go studied Go for who know how many years (equivalent to many human lifetimes), playing against variants of itself until the devs were satisfied with the wiring of its brain and estimating it to be good enough to beat a human grandmaster level player. And… *it did*.
Having watched both Alpha Go vs Lee Se-dol, and Alpha Star (Starcraft 2 AI) vs some humans, I have to say, both seem to fall back on some pretty absurd behaviours when losing. I can’t help but wonder if Alpha Go is actually incapable of resigning and its (official, by resignation) loss against Lee was actually a human at Google deciding to put an end to its misery. Decision tree programs usually have a good idea when they’re beaten and resign gracefully whereas Alpha Go throws a small tantrum or something… It’s not a decision tree, it has something sorta like a mind.
Latest Answers