Linked by Eugenia Loli on Mon 28th May 2012 03:53 UTC
General Development FuriousFanBoys interviews Ben Goertzel regarding Artificial Intelligence. Ben started the OpenCog project (an open sourced AI non-profit), acts as an adviser to the Singularity University, and currently bounces back between Hong Kong and Maryland building in-game AI.
Thread beginning with comment 519699
To view parent comment, click here.
To read all comments associated with this story, please click here.
RE[5]: Memetics
by cfgr on Mon 28th May 2012 13:26 UTC in reply to "RE[4]: Memetics"
cfgr
Member since:
2009-07-18

AI has gone much further than that, and 20 years ago.

Computers are perfectly capable of learning by trial and error to beat humanity hands down. Gerald Tesauro developed a machine that can teach itself to play backgammon.

It does so by playing against itself and exploring policies, not moves, and by punishing/rewarding those policies for how well it is doing. It's called reinforcement learning. The code did not contain any strategies, only the game rules. After thousands of games against itself, it managed to beat average players, Tesauro included.

Then they let it learn from analysed situations from a database (i.e. like from a book) - again nothing coded. After this it could defeat the world's best players and in fact changed the way grandmasters play backgammon.

It's pretty awesome.

http://www.research.ibm.com/massive/tdl.html

Reply Parent Score: 2