Accepting the (current) limitations of machine learning (and some recent hiccups), it's still fascinating how AI capability is developing so fast despite the field being far from new. The 'AI beating humans' narrative (as Tom Chatfield talked about at last year's Google Firestarters on the theme of AI - describing it as the ‘usurpation narrative of human-machine interactions…a creation is pitted against its creators, aspiring ultimately to supplant them’) is not always that helpful. But still, the landmarks along this storyline (AI beating humans at Jeopardy and more recently the famous victory of Google Deepmind’s AlphaGo against GO world champion Lee Sedol) are quite compelling.
So now we have the AI system (developed by Carnegie Mellon University) that has just accumulated $1.7m worth of chips over twenty days playing against four of the world’s top professional poker players. What was particularly remarkable about AlphaGo's victory over Lee Sedol (which I remarked on at the time) was the way in which AlphaGo won, playing in highly unexpected and creative ways, with moves that left even its creators 'pretty shocked'. Wired described how:
‘…with its 19th move, AlphaGo made an even more surprising and forceful play, dropping a black piece into some empty space on the right-hand side of the board. Lee Sedol seemed just as surprised as anyone else. He promptly left the match table, taking an (allowed) break as his game clock continued to run.’
The commentator at the time noted how Sedol appeared to have trouble dealing with a highly unusual move that he had never seen before. It was a real moment in the development of AI/human relations.
And perhaps we've just passed another one since what’s again interesting about the poker example is that on its way to victory, the AI succeeded in out-bluffing its human opponents. Poker, the FT piece notes, tests different mental muscles since it involves 'strategising using imperfect information in a way that is more akin to the real world'. The AI started not so well, but then focused on fixing its own weaknesses and filling the holes in its strategy as it went. It was able to use an additional feedback loop to respond in real time to the human activity. The implications are that AI can take other imperfect information situations and derive a good strategy for that situation.
Another new milestone.
Image courtesy