Lee Sedol may be a great go player, but I don't think he's a very good software tester. Specifically, I criticize his last move -- resigning! I would advise playing it out well beyond where most humans would agree to stop play. Probably won't work, but you don't find bugs in software by just assuming it must be fine. You find bugs by pushing it. I think there is a consensus that no previous go program has managed to "solve" the problem of exactly when the game is over and even in positions where go masters resign, a player like me often cannot even tell who is ahead (!) and so even positions masters consider easy, are by no means easy. And it isn't so easy to write software like this without bugs. In the Kasparov-DeepBlue match #2, which K lost 3.5 to 2.5, Kasparov resigned one game in a position where he could have forced a draw. (Which would have changed match outcome to 3-3.) Even I, who was 800 elo dumber than Kasparov (if I'm generous to myself) saw the rough outlines of that draw line and would have gone for it instead of resigning. I quite likely would have failed (since it required some precise play, which I might not have been able to pull off) but it would have been better than resigning, and Kasparov should have been strong enough to find the right moves in that line. So why did he resign? He "explained" that the computer had outplayed him all game, was clearly super grandmaster, and therefore it was inconceivable the computer had missed that comparatively obvious draw line, so it wasn't even worth trying. Say WHAT? The reality, which Kasparov did not appreciate, is perpetual check draws were a fairly well known tough issue for computers. Hard for them to see. And in fact, the computer had not seen it. And what humans find are "easy" and "hard" and "obvious" positions to handle, are not necessarily the same as what computers find easy and hard and obvious to handle. And computers have in the past experienced hugely embarrassing bugs in matches with humans and with other computers, and alphago has been little tested by anybody external to google. Another game, Kasparov decided to play some opening he was not familiar with, and blundered in a book position into a book trap loss. Apparently K figured the computer knew his favorite openings since they'd input all or most of K's games and had the computer pre-analyse them. Also dumb choice. The truth was, computers at that time were known not to be as strong as correspondence chess masters, i.e. DeepBlue lost a correspondence match with Mike Valvo... the longer the time to think, the worse chess computers did versus humans; but in fast games the computers are supreme. Because the humans have (or anyhow certainly at that time had) effectively a smaller growth factor C in their exponential: TimeUsed=C^DepthSearched. So in my opinion, if K had not employed stupid anti-computer strategy, the truth was he probably still would have been a bit stronger than DeepBlue even in the 2nd match. It also is known that in at least one move of one of the K-D matches, D basically crashed due to a bug, and played a "safety move" due to some bug-recovery fallback routine, which luckily for it happened to be an acceptable move. -- Warren D. Smith http://RangeVoting.org <-- add your endorsement (by clicking "endorse" as 1st step)