Demis Hassabis

@demishassabis give AlphaGo some vocabs from dictionary and let him start tweeting with actual emotional response from the game :-)

@demishassabis if AI had been surprised, and made mistakes as a result, then time to worry. #skynet

@demishassabis how does that compare with the probability of all the moves Lee played before that point?

@demishassabis @TimHaines I doubt it was "surprised".

@demishassabis @etorreborre how many moves were ranked better?

@demishassabis Like humans Alphago cudnt understand an outlier. #move78 is going to be the kill code for a future AI villain I guess :-P

.@demishassabis How cool would it be to check the "logs" of Lee at the time. Wait you get to see something as cool in self training!!

@demishassabis surprising? Means your model did not possess weighted curiosity, too efficient to see alternatives

@demishassabis amazing!!!

@demishassabis was that why it started valuing lower value move for about 10 turns? That was my theory

@demishassabis Interesting... so it was one move and not a formation of stones that Alphago was not prepared for. I was curious about that.

@demishassabis AG made a series of mistakes since move 83 that eventually gave up the game

@demishassabis From some postgame analysis by professionals move 79 wasn't optimal but it was also not enough to reverse the game.

@demishassabis Then it is strange that AG did not take more time on move 79