Argomenti trattati
Gabriele Farina grew up amid vineyards in a hilly part of northern Italy, the son of parents without college degrees who nonetheless encouraged his curiosity by buying the technical books he wanted and supporting an education focused on science rather than classical studies. Early on he became captivated by the notion that human-designed algorithms could produce systems that surpass their makers at specific tasks, and by his mid-teens he was already exploring how machines might predict or decide more reliably than people. This fascination with the interplay between simple mathematical building blocks and emergent, powerful behavior set the stage for a career that bridges theory and real-world systems.
As a teenager, Farina wrote programs to analyze games he played with his younger sister, demonstrating that algorithms could find the optimal play and reveal losing lines long before human players recognized them; his sister, naturally, was less thrilled by such preordained outcomes. He went on to study automation and control engineering at Politecnico di Milano, but gradually shifted toward foundational questions about why methods work and how they can be extended. Encouraged by his advisor, Nicola Gatti, he moved from applied engineering toward computational game theory and then, one month after finishing his undergraduate studies, began a PhD in computer science at Carnegie Mellon University, where he earned distinctions including a Facebook Fellowship in Economics and Computation.
Research focus: theory that connects to practice
Today Farina serves as an assistant professor in MIT‘s Department of Electrical Engineering and Computer Science and leads work as a principal investigator at the Laboratory for Information and Decision Systems (LIDS). His research fuses game theory with machine learning, optimization, and statistics to create algorithms that enable reliable decision-making in environments with many strategic actors. While he is committed to rigorous mathematical foundations, he also emphasizes demonstrable impact: theoretical advances should translate into tools that scale and perform in concrete multi-agent settings. His blend of abstraction and application earned him the NSF CAREER Award in 2026, recognizing both the depth and the ambition of his program.
What equilibrium means in this work
At the mathematical core of Farina’s investigations is the concept of an equilibrium, a state where no participant can profitably change strategy given others’ choices. Calculating such points becomes computationally intractable as interactions, objectives, and actions proliferate. Farina’s goal is to use optimization and algorithm design to find these stable points efficiently, turning problems that might otherwise require infeasible amounts of computation into tractable ones. By clarifying the structural properties that make efficient computation possible, his team helps produce algorithms capable of reasoning about competing incentives in large systems.
Algorithms and scalability
Farina probes how classical mathematical languages from game theory can be retooled to handle modern scale and complexity. He studies how to model multi-agent interactions so that solution methods exploit regularities rather than suffer from combinatorial explosion. This includes designing algorithms that control complex dynamics and produce practical strategies in settings where naive computation would be impossible. The emphasis is not merely on proving existence of solutions but on furnishing constructive procedures that practitioners can deploy in AI systems that must act in strategic, uncertain environments.
From industry projects to imperfect information games
Between academia appointments, Farina spent a year as a research scientist at Meta’s Fundamental AI Research Labs, where he contributed to Cicero, an AI system that negotiated, formed alliances, and detected deception in a game-like environment. Cicero was explicitly engineered to decline agreements that were not in its interest and to infer when other agents might be bluffing, modeling incentives to decide whether proposed actions were credible. The system attracted attention as a step toward machines that can manage complex problems requiring compromise, with a 2026 article in the MIT Technology Review highlighting its role in advancing AI negotiation capabilities.
Revisiting board games: Stratego and bluffing at scale
Farina’s longstanding interest in games with hidden information—situations where some players possess knowledge unavailable to others—has shaped a major part of his research. He studies settings characterized by imperfect information, where the concealment and revelation of data are strategic assets; everyday examples include games like poker, where deception and timing preserve informational advantages. In a notable project, his team tackled Stratego, a game that demands complex risk calculation and misdirection. Using modern algorithms and cost-effective training that totaled less than $10,000 rather than the millions some prior efforts required, they produced a program that outperformed the game’s top human, recording 15 wins, four draws, and one loss.
Farina sees these game-focused successes as more than trophies: they are testbeds for methods applicable to negotiation, security, and multi-party decision-making in the real world. He argues that as machines become adept at strategic reasoning—even at bluffing—they will be integral to broader AI systems that must navigate competing objectives. His work continues to push both the mathematical underpinnings and the algorithmic machinery needed to make those systems reliable, interpretable, and practically useful.