The final bastion of human superiority in games is collapsing as I write this column. For the last week, four of the world’s leading poker professionals have been engaged in one-on-one combat against a software called Libratus. From the same Latin root as the weighing scales of the sun sign Libra, Libratus means balanced or poised. The scales of the match are tilted heavily in favour of the machine right now. With over half the hands completed, the humans are lagging by about $ 700,000, an almost impossible gap to overcome, and one that yawns wider at the end of most sessions.
It might come as no surprise that a program would best people at poker considering that humans have ceded superiority in games more associated with intellect like chess and Go. But poker is different in being a game of incomplete information, resembling real-life situations more than chess and Go, and making it difficult for machines to master. There isn’t necessarily a unitary calculable best move in a given situation in poker, since determining that would require knowing opponents’ hidden hands as well as their habits. Machines have for years been better than the best human players at a variant of poker called Limit hold’em. In that game, players can bet only a specific, not very large, amount at every step, reducing the number of combinations the machine has to calculate. The more popular variant of poker, No Limit hold’em, allows you to bet any amount at any point (though only what you have out on the table, fishing cash out of your wallet or throwing in car keys is forbidden). The variety of permitted bets pushes No Limit hold’em into the rarefied realm of games whose possible situations compare favourably with the number of atoms in the universe.
Libratus came fast out of the gate when the marathon event began on January 11, building up a substantial lead on the first day. The humans analysed its tendencies, discovered imbalances, and found counter-plays during their post-session discussion, and hit back in succeeding sessions, almost pulling even. But Libratus was studying them even as they were studying it, and lived up to its name by countering the counter-plays. At the moment, a sense of inevitability is setting in within the human camp. The machine learns faster than they do, it is a more intelligent poker player than they are.
Intelligence and consciousness
The academics at Carnegie Mellon University who wrote Libratus’s programme are doubtless celebrating, but the machine itself will feel no joy to balance the sting of defeat in the human camp. The recent rise of artificial intelligence after years of under-performance has inspired a number of movies in which machines become self-aware and attempt to take over the world. I believe that scenario stems from a fundamental misunderstanding in which intelligence is confused with consciousness.
Since humans are the most intelligent living creatures, and also the most self-aware ones, the two qualities are intimately linked in our minds. It is only a small leap to the notion that consciousness emerges at a certain neuronal density and machines will replicate it once we endow them with sufficient processing power. Since no machine has yet exhibited the self-awareness of even a jellyfish, such a line of thought is almost certainly flawed. More likely, machines will keep getting more intelligent without transiting into self-awareness.
Jobs destroyed
This means that Terminator-style doomsday scenarios are unlikely. However, the kind of intelligence machines now demonstrate has begun destroying technical and white collar jobs, and the trend is going to get very bad very soon. I wrote about the danger it poses to software professionals in India after Google’s AlphaGo beat Lee Sedol last year. For 10 years after its introduction, Google Translate was an extremely popular but laughably bad language converter. After a decade at more or less the same rudimentary level, Translate suddenly became proficient in a number of languages in November 2016, a consequence of switching overnight to a new AI-based system. It looks like human translators are going to be out of work soon, as are taxi drivers. AI has bolstered the capability of self-driving cars to the point that these autonomous machines outdo humans in most situations.
India will probably be the toughest nut to crack in this respect, but it doesn’t say much for us if the only thing keeping cab riders in their jobs is the state of our roads and the chaos on our streets.
A lot of journalism these days is content aggregation, and that’s another area where robots are catching up fast. Just last week, the China Daily News carried a short piece written by a machine. The bot took all of one second to compose 300 characters.
Optimists argue that AI is like previous technological advances. Since the harnessing of iron, new technology has always destroyed jobs and always created more new ones than it destroyed. But this time round, I’m not sure if the new jobs will be as well-paid or as numerous as the obsolete ones. It’s a game of incomplete information, but I’m willing to bet that this time is different.
Corrections and clarifications: An earlier version of this article erroneously stated that the bot composed 300 “words” in one second instead of 300 “characters”, as mentioned in the China Daily News story linked above.