New Lex Fridman Insight: Noam Brown: AI vs Humans in Poker and Games of Strategic Negotiation
Sent June 11, 2026
Key Insights
- Libratus AI defeated top poker players using Nash equilibrium strategies, winning $2 million over 120,000 hands.
- AI's success in poker shows the potential of search-based strategies over pre-computed ones, crucial for games with imperfect information.
- AI in poker benefits from chaotic strategies for viewer engagement, not just optimal play, highlighting the role of personality in competitive gaming.
- Cicero AI ranked second in Diplomacy, demonstrating AI's capability in complex, negotiation-heavy games.
- AI's limitation in modeling human irrationality affects performance in multi-player games like Diplomacy.
How the conversation moved
The episode begins with Lex Fridman introducing Noam Brown, focusing on the strategic complexities of poker, particularly No Limit Texas Hold'em. Brown explains the concept of Nash equilibrium, a foundational idea in game theory, which ensures that in any finite two-player zero-sum game, there exists an optimal strategy that prevents loss in expectation. This sets the stage for discussing how these concepts apply to AI in poker, a game known for its high variance and complex decision-making.
Brown elaborates on the development of the Libratus poker bot, which was designed to exploit Nash equilibrium strategies to defeat top human players. Libratus played 120,000 hands and won nearly $2 million, showcasing the effectiveness of these strategies. The conversation highlights the shift from pre-computed strategies to search-based approaches, which allow AI to dynamically adapt to the game's evolving state, a crucial advantage in games with imperfect information like poker.
Lex Fridman notes the tension between AI's optimal strategies and the entertainment value of chaotic, aggressive play styles. He suggests that while AI can achieve superhuman performance, it might benefit from incorporating elements that enhance viewer engagement. Brown acknowledges this point, explaining that AI's role in poker is not just about winning but also about creating an engaging experience, which is why professional players often earn more from sponsorships than from gameplay.
The discussion pivots to AI's application in the game of Diplomacy, where Brown highlights the challenges of modeling human-like negotiation and trust. AI's performance in Diplomacy, particularly the Cicero bot, demonstrates its capability in complex, negotiation-heavy games, ranking second among human players. However, the limitations in modeling human irrationality and suboptimal behavior are acknowledged, underscoring the need for AI to better understand human dynamics in cooperative settings.
Surprising moments
In-depth
AI in Poker
- Libratus AI used Nash equilibrium to defeat top poker players.
- AI strategies in poker are not just about winning but also viewer engagement.
- AI's chaotic strategies enhance the entertainment value of poker.
AI and Game Theory
- Search-based strategies are crucial for games with imperfect information.
- AI's success in Diplomacy highlights its negotiation capabilities.
- AI's limitations in modeling human irrationality affect performance.
Notable Quotes
In any finite two-player zero-sum game, there is an optimal strategy that if you play it, you are guaranteed to not lose an expectation no matter what your opponent does.
Still open
- Lex asked whether AI could incorporate chaotic strategies to enhance viewer engagement in poker, a point left open for future exploration.