What’s it like to play poker against an AI bot? Really weird and really difficult, it turns out.
That was the vibe when we checked in on the 20-day “Brains vs. AI: Re-Match” in Pittsburgh, which pits four professional poker players against a Carnegie Mellon University-designed bot named Libratus.
Computer scientists at the university had tried this stunt with a bot named Claudico in 2015, but the humans picked up on its weird machine weaknesses and beat it soundly. So the scientists went back to the lab and rewrote its algorithm in a few key areas to make it unbeatable.
Their work paid off.
On Jan. 30, Libratus made history by trouncing the humans in a marathon 120,000 hands, coming out on top by $1,766,250 in chips — and pretty much killing it every day before that. The humans ended their inglorious run by splitting shares of a $200,000 prize.
The bot’s win has numerous implications for any field where decision-making is based on incomplete, or hidden, information. Or straight-up misinformation. Fields where people are forced to think like poker players, basically. Business negotiation, military strategy, cybersecurity, and medical treatment are all areas the bot’s designers say could benefit from the algorithms that govern its automated decision-making.
Get ready, humanity — the bots have entered the building.