Tremble, humans! After AlphaGo, there is the Cold Poker Master

I'm afraid everyone started paying attention to AI after the hot article about AlphaGo defeating Lee Sedol. However, now we not only have AlphaGo, but also Cold Poker Master, an AI robot specializing in Texas Hold'em.

[[188363]]

First, let's understand the difference between Texas Hold'em and chess. Go is a "perfect information game" in which all situations on the scene are clearly laid out on the board in black and white, and both sides have equal information. However, Texas Hold'em hides a lot of information and is an "imperfect information game." Players don't know what cards their opponents have, what the results of the five public cards will be, or what their opponents guess at their hands. Therefore, although the number of decision points in a game of Texas Hold'em is 10 to the 160th power, which is less than Go, Texas Hold'em cannot be played by searching alone. Texas Hold'em has thus jumped out of the competition in search volume and put forward another test for AI: dealing with hidden information.

The basic principles of AlphaGo are: Policy Network, which predicts and samples the next move given the current situation; Fast rollout, which increases the speed while sacrificing the quality of moves; Value Network, which estimates the winning rate of both sides given the current situation; Monte Carlo Tree Search (MCTS), which connects the above three parts into a complete system.

The algorithm of Lengpu Master is mainly based on modules such as Nash equilibrium, Counterfactual Regret Minimization (CFR), end-game solver and self-reinforcement learning.

However, for users, which one is more useful, AlphaGo or Cold Poker Master, other than allowing us to read the news?

In real life, the things we encounter are more like playing Texas Hold'em: the unknown cards of the other party in business negotiations, the unpredictable actions of competitors in house auctions, and some hidden inside information in stock trading. Sandholm believes that Texas Hold'em AI has opened the door to solving random events and hidden information through artificial intelligence. Such AI is expected to leave the virtual chess and card world and become a good helper for humans in negotiations, games and investments in real life.