26 Jun '18 09:54>7 edits
https://techxplore.com/news/2018-06-dota-bots-hard-taught-cooperative.html
Dota 2 is an extremely complex game with much more complex rules (in terms of the number of rules) than chess and, unlike chess, requires cooperative behavior between individual people i.e. teamwork.
This shows how AI can solve extremely complex problems and potentially better than humans.
Although this problem is just a game, it does make me wonder what the AI potential is for solving real-world complex problems of great practical significance. The one problem I have in mind in particular is the problem of figuring out all the relevant quantum physics behind high temperature superconductors and then figure out molecular designs for room temperature superconductors or, if the laws of physics are such that they indirectly forbid room temperature superconductors, at least give some kind of mathematical proof of that so we know not to waste any more time and research money trying to do that impossibility. Either result would be of great value (although, obviously, I would prefer that the former one will be realized). But this would require the AI first learning the complex rules of quantum physics but the details of the correct way to apply those rules to superconductors is currently ill-defined so it would have to somehow 'learn' that part and I guess that would make this problem a few orders of magnitude harder than merely playing a game of Dota 2 well.
Dota 2 is an extremely complex game with much more complex rules (in terms of the number of rules) than chess and, unlike chess, requires cooperative behavior between individual people i.e. teamwork.
This shows how AI can solve extremely complex problems and potentially better than humans.
Although this problem is just a game, it does make me wonder what the AI potential is for solving real-world complex problems of great practical significance. The one problem I have in mind in particular is the problem of figuring out all the relevant quantum physics behind high temperature superconductors and then figure out molecular designs for room temperature superconductors or, if the laws of physics are such that they indirectly forbid room temperature superconductors, at least give some kind of mathematical proof of that so we know not to waste any more time and research money trying to do that impossibility. Either result would be of great value (although, obviously, I would prefer that the former one will be realized). But this would require the AI first learning the complex rules of quantum physics but the details of the correct way to apply those rules to superconductors is currently ill-defined so it would have to somehow 'learn' that part and I guess that would make this problem a few orders of magnitude harder than merely playing a game of Dota 2 well.