AlphaZero AI beats champion chess program after teaching itself in four hours

[Taomeow]

https://www.theguardian.com/technology/2017/dec/07/alphazero-google-deepmind-ai-beats-champion-program-teaching-itself-to-play-four-hours

 

The most interesting/disturbing part of the way AI learns/teaches itself how to solve problems is that it does not use human reasoning at all, and the methods whereby it arrives at its conclusions are, to humans, incomprehensible.  

 

(Edit: changed my mind about a tidbit that originally followed.)

 

Edited December 8, 2017 by Taomeow

[Wells]

On 8.12.2017 at 6:35 AM, Taomeow said:

The most interesting/disturbing part of the way AI learns/teaches itself how to solve problems is that it does not use human reasoning at all, and the methods whereby it arrives at its conclusions are, to humans, incomprehensible. 

 

Next task: Solve the problem of humankind...

Hello, Skynet.

[thelerner]

amazing..https://www.extremetech.com/extreme/260215-alphazero-new-chess-champion-harbinger-brave-new-world-ai- ..By simply playing against itself for a mere 4 hours, the equivalent of over 22 million training games, AlphaZero learned the relevant associations with the various chess moves and their outcomes. In doing so, it was learning much the way a human does, but because the computer can compress 100,000 hours of human chess play into a few minutes, it builds up a set of associations far more quickly than we ever could, and over a far wider range of move combinations.

 

And now its owned by Google.  It's been pointed out that its expertise is in limited areas.  Yet, the idea of a 100,000 hours of experimentation in a few minutes is amazing.  Seems like there are many scientific problems you could set it to and have it chug along practically randomly until it hits a workable solution.   Course, the computers we're using now are probably the super computer equivalents of the early 90's.  late 80's. 

 

Here's some perspective on the latest tech as far as chips.  https://www.popsci.com/intel-teraflop-chip#page-3

"..In fact, if you wind back the clock to 1976, the Cray-1 supercomputer could do just 160 megaflops (that’s a million FLOPS), meaning this singular new Intel chip is about 6,000 times as fast as that machine, says Steve Scott, a senior vice president and chief technology officer at Cray, Inc..." 

 

And thats a single Chip!  These days a kids Play Station 4 is processing over a teraflop at a time.