← Back to Main

Game Theory IRL

Published on February 5, 2026

Introduction

Game theory, the mathematical study of strategic decision-making among rational agents, has applications in various academic domains such as economics, politics, statistics, and biology. To provide brief context, I will define some terms as well as some types of games:

• Players: Decision-makers.
• Strategies: Possible actions a player can take.
• Payoffs: Outcomes or rewards resulting from combinations of strategies
• Cooperative games focus on group outcomes, while non-cooperative games analyze individual strategies.
• Simultaneous games involve players acting at the same time or without knowledge of others' actions (e.g., Rock-Paper-Scissors) while sequential games involve turn-based moves where subsequent players can observe previous actions (e.g., Chess).
• One-shot games are single, isolated interactions where players maximize immediate payoff without concern for future consequences. Repeated games involve playing the same game multiple times, allowing for reputation building, retribution, and cooperation.
• Zero-sum games means one person's gain must equal another's loss, (e.g., poker), while non-zero-sum games allow for mutual gain or mutual loss, (e.g., Battle of the Sexes) meaning the total outcome isn't fixed.
• Nash Equilibrium: A stable state where all participants are making their best possible decision based on the choices of their opponents. Think the sweet spot!

Now that you have enough game theory knowledge to pass the intro course, we can get into our main subjects.

Game Theory in CS

If you are involved in computer science fields, such as myself, you probably recognize game theory from models such as GANs. Generative Adversarial Networks (GAN) consist of two neural networks: a generator creating fake data, and a discriminator trying to spot the fakes. They are in a constant, simultaneous, zero-sum game against each other. Ideally, once training is complete, the models would reach a Nash Equilibrium where the generator produces data indistinguishable from real data, making the discriminator's accuracy 50% (random guessing). This adversarial process drives both networks to improve rapidly, leading to realistic AI-generated images. Another example is self-driving cars navigating traffic. Each car's AI needs to predict and react to the movements of other AI-driven or human-driven vehicles. This naturally brings many ethical conversations to mind, but that's for another time. By solving for Nash Equilibria in real-time, autonomous vehicles can predict traffic, negotiate merges, and execute complex maneuvers like yielding or overtaking. This is all nice and dandy, but in the real world, we can't model everything... or can we?

Game Theory in Real Life

You might be surprised how often you encounter game theory without even realizing it. Every time you make a decision that considers someone else's potential response, you're dabbling in game theory. Imagine you're on your morning commute. Do you take the main highway, which is usually faster but prone to heavy traffic, or the longer scenic route with less congestion? Your decision isn't just about your own preference; it depends on what other drivers choose to do. If everyone avoids the highway, it becomes the faster option.

However, when we defined game theory, we said the players or agents are rational. It's widely accepted that humans are irrational. The above examples assume players are rational, that they always act to maximize their own clear benefit. However, humans are messy and emotional. We don't just care about payoffs such as money or efficiency; we care about fairness, spite, love, revenge, and reputation. Now, we enter the realm of Behavioral Game Theory. Nowhere is this assumption of rationality tested more than in our social lives. In relationships, the "payoff" isn't just objective gain; it includes complex emotional states. Imagine a stranger is given $100 and has to offer you a portion of it. If you accept the offer, you both keep the agreed amounts. If you reject it, nobody gets anything. A purely "rational" person should accept even $1, because $1 is better than nothing. Yet, in real-life experiments, people routinely reject offers below 30% because it feels "unfair." They are willing to suffer a personal loss just to punish the other person for being greedy. In social conflicts, we often take actions that hurt ourselves just to hurt the other party more out of anger or spite. Similarly, in romantic or deep platonic relationships, standard game theory often breaks down.

We frequently perform acts of pure altruism, sacrificing our own immediate well-being for our partner's happiness without calculating an immediate return. Why? Because in the long "game" of a relationship, the payoff function changes. The other person's happiness becomes part of your own payoff. Furthermore, emotions like guilt, shame, and love act as powerful internal enforcers of cooperative behavior, preventing us from defecting on our partners even when we could get away with it.

My Unsolicited Two Cents

Game theory shouldn't be used to "win" a relationship; it should be used to design the environment so that "winning" becomes irrelevant. In traditional game theory, we analyze a game to find the strategy that maximizes our own payoff. In a relationship, if you find yourself calculating Nash Equilibria, you've already lost. If you're calculating Nash Equilibria about your loved ones, the "game" has shifted from Cooperative to Non-Cooperative. Most people view relationships through the lens of a one-shot game (like a single argument or a choice of where to eat). However, a healthy relationship is an iterated or repeated game with an unknown end date. The most successful social relationships are those where the players intentionally "break" their own rationality. Use game theory not as a playbook, but as a thermometer. If you and your partner are in a "Prisoner's Dilemma" (where both being selfish is the "logical" choice but leads to the worst outcome), game theory helps you see the trap. It helps you realize that if you punish your partner for being honest, you are effectively "incentivizing" them to lie in the next round of the game. It reveals when one person has more "information" than the other, creating an imbalance that leads to anxiety rather than trust.

If you optimize for your own payoff, you eventually turn your partner into an opponent. Once you treat a partner as a variable to be managed rather than a person to be known, the "emotional payoff" of the relationship drops to zero. Game theory requires a "Utility Function," a way to score a win. But how do you score the "utility" of a quiet morning or a shared laugh? If you can't measure it accurately, your model is built on sand. In a cold, mathematical world, vulnerability is a weakness that can be exploited. In a relationship, vulnerability is the only mechanism that builds the "Infinite Loop" of trust. You should use game theory to understand the mechanics of conflict, but you should never use it to govern the mechanics of connection. Think of game theory like a map of the ocean: it's great for seeing where the dangerous whirlpools are, but it's a terrible way to learn how to swim. To truly connect, you eventually have to put the map down and get wet.

With lots of love and Joy,

Liamo