Page 1 of 1

Almost Feels Like First Contact

PostPosted: Sun Dec 10, 2017 5:16 am
by RinkyDink
An alien intelligence masters chess in four hours.


Re: Almost Feels Like First Contact

PostPosted: Sun Dec 17, 2017 8:14 pm
by RinkyDink
More machine learning achievements.

https://www.nytimes.com/2017/12/14/scie ... ystem.html

Re: Almost Feels Like First Contact

PostPosted: Fri Dec 29, 2017 7:22 am
by CFrantz
That's really nice.

Re: Almost Feels Like First Contact

PostPosted: Sat Dec 30, 2017 3:49 pm
by pmcaero
no it doesn't. Chess is still very limited in scope and number of choices. Machine Learning applied to real world data still has a ways to go before it appears remotely intelligent. It's useful for some limited subsets - the so-called specialized AIs - but will not pass for human for decades to come.

Re: Almost Feels Like First Contact

PostPosted: Sat Dec 30, 2017 4:51 pm
by RinkyDink
pmcaero wrote:no it doesn't. Chess is still very limited in scope and number of choices. Machine Learning applied to real world data still has a ways to go before it appears remotely intelligent. It's useful for some limited subsets - the so-called specialized AIs - but will not pass for human for decades to come.

I think it depends on your definition of "intelligence". Some people believe that a vending machine possesses a limited form of intelligence. I think in the realm of chess Alpha Zero shows the highest level of intelligence--no human can defeat it, no computer program, as far as I know, can defeat it. The question I have is whether Alpha Zero can learn to consistently defeat a human playing against it with the aid of a powerful chess engine. I'd love to see Magnus Carlsen play 100 games against it while using Stockfish 8 to aid him in choosing his moves. Would Alpha Zero be able to learn to defeat a combined human/machine intelligence? I think we'll probably get an answer to that question in the near future. As far as complexity goes, Alpha Zero breezed through learning chess, but an even more impressive achievement was its learning of Go. There are way more possibilities involved in a move in Go than there are in chess.

Re: Almost Feels Like First Contact

PostPosted: Sun Dec 31, 2017 2:13 am
by pmcaero
I'm more impressed by this deep learning feat:

Re: Almost Feels Like First Contact

PostPosted: Sun Dec 31, 2017 6:01 pm
by RinkyDink
pmcaero wrote:I'm more impressed by this deep learning feat:

That's cool too. It makes you wonder what the program has learned about how arm movements affect a human-like structure's overall movement, particularly in connection to speed. I'd be interested to see what happens when they simulate a surfing break, give the program a virtual board, and ask it to learn to popup. Now that's a real test :D

Re: Almost Feels Like First Contact

PostPosted: Sun Dec 31, 2017 6:19 pm
by RinkyDink
I do worry, however, about what will happen when this technology becomes widely available. I figure it would be a hacker's dream come true to get their hands on it. It appears to belong to Google, at the moment, and that really makes the Pentagon seem irrelevant.