Earlier this month, Google-DeepMind's AlphaGo AI program took a 2-0 lead in a 5 game match of GO, with Lee Sedol. GO is a game more complex Chess so this win was monumental, in fact, by the end of the month AlphaGo won all games but one. link
Arguments both for and against AI rile up any time a similar match is set up between human and AI. Matrix-like questions about enslavement emerge out of nowhere for some while others dream of some Jetson-like utopia. After the emotions cooled a bit, another argument emerged: Is AI being oversold?
Jaron Lanier, a computer scientist, makes us remember that Sedol wasn't just playing against one AI program. In fact, he was playing against the "aggregate of 30 million moves made during previous games by human players". Lanier warns against a "premature mystery reduction" in which society, or at least some scientists, pretend that something works when it really doesn't.
Currently, most AI software today is programmed via massive datasets that represent all kinds of human activity. Lanier continues, "We want to pretend that there's this AI behind the curtain that's freestanding. But, actually, there's millions of people there, too.”
This picture is licensed under Wikimedia Commons.