Computers beating humans in a game of wits is nothing new: chess grand master Garry Kasparov was first beaten by a computer in the 1990s. Google's AlphaGo programme triumphed over a human professional in 2015.
But both chess and go are closed-ended one-on-one games, where infinite processing power means a computer could look at all available moves and play a "perfect" game.
Poker - so the story goes - is different. While top players need an expert understanding of probability, pure processing power is not enough to beat the best. To excel at poker, a player needs to read the humans around them and pick up on cues so they know when to bluff and when to call.
This theory has been dealt a convincing blow by Pluribus, an algorithm created by Facebook and Carnegie Mellon university.
Pluribus won a big hand against 4 top poker players, despite having a weak set of cards. The algorithm had learnt that by playing aggressively and bluffing their way through, they could bully others into folding and win the hand.
As machine-learning development continues at a rapid pace, algorithms are beginning to expand into areas where the data is murkier and the outcomes less clear.
“People have this notion that [bluffing] is a very human ability—that it’s about looking into the other person’s eyes,” Dr. Brown said. “It’s really about math, and this is what’s going on here. We can create an AI algorithm that can bluff better than any human.”