Facebook is no stranger to hokey intelligence . The social media troupe uses a form of A.I. , jazz as bass learning , to ramp up its facial recognition software , and has modernize a yet - to - be - released A.I. system that will describe photos to dim Facebook drug user . It makes a lot of good sense that the fellowship would strain to delay on the cutting boundary of both engineering and societal media by tinkering with A.I. systems . What ’s surprising , however , is that their latest A.I. experimentation is pitch not towards proceed up with the latest social media trends , but towards winning a 2500 - year - previous display panel game .
allot toWIRED , Facebook is trying to build up an A.I. system that will be able to puzzle the world ’s good Go players . Over the last few decades , computer have defeated the world ’s best human players at checkers , chess , Scrabble , and even Jeopardy . In 1996 , IBM ’s Deep Blue supercomputer beat world champion cheat player Gary Kasparov in one of the most famous chess games ever played . But no data processor has ever been created that can beat humans at the ancient Formosan board plot Go — and not for deficiency of trying .
Go is a misleadingly simple secret plan . player can place their pieces on any intersection of two lines on the 19 - by-19 grid , using cable of contiguous pieces to define territories , or to capture their opponent ’s pieces . Whoever capture the most infinite and the most pieces by the end of the plot is the winner . Unlike chess game , pieces do n’t move around the board in complex traffic pattern — in fact , they do n’t move at all unless captured . But the conceptual simplicity of the biz is precisely what makes Go so tough : since role player can place pieces anywhere on the grid , the range of different strategies and potential move is immense .
WIREDexplains that , traditionally , computers defeat world at board plot by “ analyzing the many potential outcomes of every potential move . ” Deep Blue beat Kasparov because it could mechanically analyse more moves than he could , in a much smaller amount of time . But the chain of mountains of possible relocation on a Go circuit board is not only too great for most reckoner to analyse while actively diddle , but the exact rules behind what make a dependable move can be unmanageable to articulate .
“ We ’re pretty indisputable the best [ human ] musician cease up look at visual patterns , front at the visuals of the board to help them translate what are good and speculative configurations in an intuitive way , ” Facebook CTO Mike “ Schrep ” Schroepfer toldWIRED .
So , Facebook is using deep eruditeness to develop a fresh approach to master Go . The troupe require to ramp up a arrangement that will incorporate the “ nonrational ” elements of Go scheme by seem at previous game and learning from them . Schroepfer explains , “ We ’ve take some of the BASIC of biz - take on A.I. and attached a ocular system to it , so that we ’re using the rule on the add-in — a optical rec[ognition ] organization — to tune the potential moves the system can make . ”
Unsurprisingly , Facebook is n’t seek to build a better Go information processing system just for fun . They go for that the advanced programme they ’re germinate for the board game will help them build up more practical software in the future tense : in the end , they need to build A.I. that can make predictions base on existing evidence , and better mime human hunch .
A " key problem in contrived intelligence is enter out what ’s choke to befall next , " Schroepfer toldWIRED . “ You do this all the time in orderliness to make your sidereal day go well . … What we ’ve got to do is learn computer scheme to understand the Earth in a similar way . ”
[ h / t : WIRED ]
