How game theory and AI are reshaping strategic decision-making

Gabriele Farina’s fascination with machines outperforming humans began in his teenage years, long before he became an assistant professor at MIT. Growing up in Italy’s wine country, Farina was captivated by the idea that mathematics and algorithms could create systems capable of making better predictions or decisions than their creators. His early experiments with board games and his sister’s skepticism about his analytical prowess laid the groundwork for a career at the intersection of game theory, machine learning, and real-world decision-making.

Today, Farina leads research at MIT’s Laboratory for Information and Decision Systems (LIDS), where he explores how game theory can be combined with optimization and statistics to solve problems where multiple agents with competing objectives must find stable solutions. His work doesn’t just aim for theoretical breakthroughs—it seeks practical applications in AI systems that navigate real-world complexity.

From board games to AI breakthroughs

Farina’s journey into game theory began during his undergraduate studies at Politecnico di Milano, where his advisor, Nicola Gatti, introduced him to computational game theory. Unlike many engineering students who apply existing techniques, Farina was drawn to the foundational questions: How do we extend the theory itself? This curiosity led him to pursue a PhD in computer science at Carnegie Mellon University, where he earned distinctions for his research and later received a Facebook Fellowship in Economics and Computation.

After completing his doctorate, Farina spent a year as a research scientist at Meta, contributing to Cicero, an AI designed to excel in the board game Diplomacy. Unlike traditional board games with complete information, Diplomacy requires players to form alliances, negotiate, and detect bluffs—skills that mirror real-world strategic interactions. Cicero’s ability to outperform humans was rooted in its understanding of incentives: it refused alliances that weren’t in its interest and recognized when opponents were bluffing. A 2022 MIT Technology Review article highlighted Cicero as a step toward AI systems that can solve complex problems requiring compromise and negotiation.

Simplifying complexity: The equilibrium challenge

Farina’s work focuses on a core question in game theory: finding equilibrium—a state where no player has an incentive to change their strategy. In large-scale systems, calculating equilibrium can become computationally intractable, taking billions of years to solve. His research aims to bridge this gap by developing algorithms that efficiently identify stable points in complex, multi-agent scenarios.

A key area of interest is imperfect information—scenarios where some agents possess private knowledge that others lack. In such cases, information itself becomes a strategic asset. Farina points to poker as a prime example: players bluff to obscure their hands, making it harder for opponents to deduce their true intentions. He notes that machines now excel at this kind of deception, a skill that translates to real-world applications like negotiation, cybersecurity, and financial markets.

Stratego: A game that broke AI’s winning streak

One of Farina’s most notable achievements involves Stratego, a military strategy game with deep layers of misdirection and risk calculation. Unlike chess or Go, where perfect information leads to clear outcomes, Stratego’s fog of war—where players don’t see their opponent’s pieces—makes it uniquely challenging. For years, researchers spent millions of dollars attempting to build AI systems capable of defeating top human players, only to fall short.

Farina and his team took a different approach. By leveraging new algorithms and training techniques that cost less than $10,000—far below previous budgets—they developed an AI that not only matched but surpassed the best human player in history. The results were staggering: 15 wins, four draws, and just one loss. Farina describes the achievement as both thrilling and cost-effective, emphasizing that these techniques could be scaled to solve other large-scale strategic problems.

The future: AI’s role in strategic decision-making

Farina’s research aligns with a broader trend in AI: the push to develop systems that can reason strategically in environments with incomplete information. His work suggests that the next frontier for AI isn’t just beating humans at games—it’s solving real-world problems where uncertainty, negotiation, and imperfect information play critical roles. From poker to international diplomacy, the applications are vast.

Looking ahead, Farina is optimistic about the integration of these algorithms into mainstream AI pipelines. “We’re seeing steady progress in building systems that can make sound decisions despite vast action spaces or hidden information,” he says. “The real challenge now is ensuring these tools are accessible, interpretable, and aligned with human values as they become more powerful.”

For Farina, the journey from a small-town winemaker’s son to a leader in AI research is a testament to the power of curiosity and foundational theory. His work reminds us that sometimes, the most groundbreaking solutions come from revisiting age-old questions—like why do people bluff, and how can machines learn to do it better?