人工智慧贏了德州撲克玩家

中央社，人腦的最後一役，人工智慧贏了德州撲克玩家，2017 年 02 月 02 日

路透社報導，卡內基美隆大學開發的人工智慧（AI）Libratus，與世界 4 名頂尖職業撲克玩家在費城進行長達 20 天的馬拉松德州撲克賽今天落幕，Libratus 贏得價值高達 170 萬美元的籌碼。

過去 20 年來，機器在無數比賽大勝人腦，包含象棋、西洋棋與以及最近 Google 人工智慧 AlphaGo 致勝的圍棋等，Libratus 的勝利之所以備受矚目，主要是德州撲克賽事中，電腦無法取得完整對戰資訊。 

就像真實的世界一樣，電腦沒辦法一次發現所有問題，且最困難的地方在於人類行為不可預測，讓德州撲克一直被視為人類最後一項豁免於被電腦擊敗的比賽。與博士生布朗（Noam Brown）共同創造 Libratus 的卡內基美隆大學電腦科學教授桑德霍姆（Tuomas Sandholm）表示：「能夠對非完整資訊進行策略推理的最好人工智慧現在已經超越最好的人腦。」 

這項勝利也掀起一股全球企業向 Libratus 尋求演算協助幫助問題解決的風潮。桑德霍姆說：「這項技術可應用在任何資訊不完整的情境下，包含商業協商、軍事策略、網路安全與醫療診斷等。」

Angela Moon, Machine beats humans for the first time in poker, Reuters, Feb 1, 2017

While machines have beaten humans over the last two decade in chess, checkers, and most recently in the ancient game of Go, Libratus' victory is significant because poker is an imperfect information game -- similar to the real world where not all problems are laid out and the difficulty in figuring out human behavior is one of the main reasons why it was considered immune to machines. 

"The best AI's ability to do strategic reasoning with imperfect information has now surpassed that of the best humans," said Tuomas Sandholm, professor of computer science at CMU who created Libratus with a Ph.D student Noam Brown said on Wednesday... 

One of the main reasons for Libratus' victory was the machine's ability outbluff humans.

"The computer can't win at poker if it can't bluff," said Frank Pfenning, head of the Computer Science Department at CMU. 

"Developing an AI that can do that successfully is a tremendous step forward scientifically and has numerous applications. Imagine that your smartphone will someday be able to negotiate the best price on a new car for you. That's just the beginning." 

Dong Kim, one of the four top poker players who participated in the tournament echoed the statement. The 28-year old, originally from Seattle, had also participated in a similar poker tournament with another AI machine built by CMU in 2015 named Claudico. 

"It was about half way through the challenge (with Libratus when) I knew we wouldn't come back," said Kim. 

"It had less bugs in the algorithm. We just ran over Claudico, bluffed it everywhere, but this time I felt like it was the other way around." 

In the battle against Claudico, the human players racked up more than $700,000 over 80,000 hands, winning almost every day of the tournament. 

In the same 2017 heads-up, no limit Texas Hold'em battle, the four human players only won five days out of 20 and split a $200,000 prize based on their performance.

在撲克牌遊戲中，有人輸有人贏，其和為 0，以經濟學賽局論 (game theory) 的術語 ，稱為零和遊戲 (Zero-Sum Game)；另外，你也不知道對手的牌，所以稱為不完整資訊。Tuomas Sandholm 教授的網頁有許多參考的文獻。