Game Theory — Prisoner’s Dilemma

Rationality, Cooperation, and the Strange Failure of Society Through the Prisoner’s Dilemma

We usually assume something simple:

If individuals behave intelligently, society as a whole should improve too.

But game theory suggests this is not always true.

In some situations, the more rationally everyone behaves, the worse the overall outcome becomes.

One of the most famous examples of this is the:

Prisoner’s Dilemma.

The Prisoner’s Dilemma is one of the most famous “toy games” in game theory. A toy game is a highly simplified model designed to capture the core structure of a complicated real-world situation. The Prisoner’s Dilemma became famous not because it is complex, but because it is disturbingly simple while still revealing uncomfortable truths about human behavior and society. Many people strongly disliked the conclusions of game theory and spent enormous effort trying to disprove them. That alone shows how psychologically uncomfortable this model can be.

The Story of the Prisoner’s Dilemma

The basic setup is simple.

Two people, Adam and Eve, are arrested for a crime. The prosecutor believes they are guilty of a major offense, but lacks enough evidence to fully convict them unless one confesses.

So the prosecutor separates them and offers each person the following deal:

• If you confess while the other person stays silent, you go free.
• If you stay silent while the other person confesses, you receive the maximum sentence.
• If both confess, both are punished, but less severely.
• If neither confesses, both receive only a small punishment.

Each player has two possible strategies:

StrategyMeaningDoveCooperate / remain silentHawkBetray / confess

This is where the real problem begins.

If both choose Dove, both receive relatively light punishment. But if one player chooses Hawk while the other chooses Dove, the betrayer receives a huge advantage while the cooperator suffers badly. If both choose Hawk, both receive poor outcomes.

The structure looks like this:

SituationResultBoth cooperateBoth receive small punishmentOnly I betrayI gain the maximum benefitOnly the other betraysI suffer the worst outcomeBoth betrayBoth suffer heavily

Why Betrayal Becomes Rational

The most important insight in the Prisoner’s Dilemma is this:

No matter what the other person does, Hawk is personally better.

Suppose the other player chooses Dove.

If I also choose Dove, both of us do reasonably well. But if I choose Hawk instead, I receive a much larger payoff. So when the other player cooperates, betrayal becomes tempting.

Now suppose the other player chooses Hawk.

If I choose Dove, I suffer badly. But if I also choose Hawk, I still lose, but I avoid the worst possible outcome.

So even when the other player betrays me, betrayal still becomes the safer strategy.

The logic can be summarized like this:

Opponent’s ChoiceMy Dove ChoiceMy Hawk ChoicePersonally Better StrategyOpponent chooses DoveModerate payoffHighest payoffHawkOpponent chooses HawkWorst lossSmaller lossHawk

So the conclusion becomes simple:

Whether the other player cooperates or betrays, Hawk is personally optimal.

Game theorists describe this by saying:

Hawk strongly dominates Dove.

Meaning that Hawk produces a better individual outcome in every possible situation.

But What Happens When Everyone Betrays?

From an individual perspective, betrayal is rational.

So Adam chooses Hawk. Eve also chooses Hawk.

The result?

Both receive severe punishment.

This is the strange and disturbing part.

If both had cooperated, both would have ended up much better off. But because each person optimizes for personal gain, both end up in a worse state.

This is the core insight of the Prisoner’s Dilemma:

Individual optimization does not guarantee collective optimization.

Or more bluntly:

Rational individual behavior can produce irrational social outcomes.

That is what makes the Prisoner’s Dilemma so powerful.

Running the Simulation

To better understand this idea, I created a simulation of the Prisoner’s Dilemma.

Each agent:

• uses either a Dove or Hawk strategy,
• repeatedly interacts with random opponents,
• and successful strategies reproduce more frequently over time.

In other words:

strategies that perform better individually spread through the system.

At first, I honestly thought my simulation was broken.

Because over time, cooperation almost completely disappeared.

1. Dove Count

The first chart shows the number of Dove (cooperative) agents over time.

At the beginning, cooperative strategies exist in significant numbers. But as the simulation progresses, Dove strategies rapidly decline and approach nearly zero.

This happens because:

• Dove agents are repeatedly exploited by Hawk agents,
• Hawk strategies gain larger short-term rewards,
• and cooperative behavior struggles to survive evolutionarily.

What makes this interesting is that:

even when cooperation is socially better, the system itself may make cooperation extremely difficult to sustain.

2. Hawk Count

The second chart shows the number of Hawk (betrayal) agents.

As Dove agents disappear, Hawk strategies begin to dominate almost the entire population.

Why?

Because Hawk performs well in both situations:

• against Dove, it gains huge advantages,
• against Hawk, it at least minimizes damage.

From an individual perspective, Hawk is an extremely strong strategy.

This is exactly what game theorists mean when they say:

“Hawk strongly dominates Dove.”

3. Average Score

The third chart may be the most important one.

Even though Hawk strategies dominate the system:

the average score of society becomes increasingly poor.

This is the heart of the Prisoner’s Dilemma.

Individually:

• Hawk is rational.

Collectively:

• a Hawk-dominated society performs badly.

This chart visualizes one of the darkest insights in game theory:

individually optimal behavior can produce globally harmful systems.

4. Both Cooperate

The fourth chart tracks situations where both agents cooperate.

At first, cooperation exists.

But over time:

• mutual cooperation becomes increasingly rare,
• trust collapses,
• and agents stop expecting cooperation from others.

The system gradually becomes trapped in a world where betrayal is assumed by default.

This is why the Prisoner’s Dilemma is so unsettling.

5. Betrayal Outcomes

The final chart tracks:

• situations where only one player betrays,
• and situations where both betray.

As time passes:

“Both Betray” becomes the dominant outcome.

The society effectively becomes locked into a state where:

• nobody trusts each other,
• defensive behavior becomes normal,
• and betrayal becomes the stable equilibrium.

This is closely related to the idea of:

“Getting Locked In.”

Once distrust spreads throughout the system, cooperation becomes increasingly difficult to recover.

Why Real Society Has Not Collapsed

At this point, an obvious question appears.

If the Prisoner’s Dilemma truly represents human society:

• why does civilization exist?
• why do humans cooperate?
• why has society not completely collapsed?

Game theorists actually argue that:

real society is NOT a pure one-shot Prisoner’s Dilemma.

Reality contains additional structures:

• repeated interaction,
• reputation,
• trust,
• punishment,
• long-term relationships,
• laws,
• memory,
• and social norms.

In real life, if you betray someone today:

• you may face retaliation tomorrow,
• lose future opportunities,
• or destroy your reputation permanently.

This changes the incentives dramatically.

That is why concepts such as:

• repeated games,
• Tit-for-Tat,
• reciprocity,
• and reputation systems

become critically important.

The Twins Fallacy

One fascinating criticism of the Prisoner’s Dilemma is called the:

“Twins Fallacy.”

The argument goes like this:

“If two rational people face the same problem, they should reach the same conclusion. Therefore, if I choose Dove, the other player should also choose Dove.”

At first glance, this sounds reasonable.

But it secretly changes the rules of the game.

In the Prisoner’s Dilemma:

• both players make decisions independently.

My choice does not automatically determine the other player’s choice.

The other player is not my mirror.

The other player is an independent rational agent.

This mistake appears constantly in real life:

• “People like me will behave like me.”
• “If I cooperate, others probably will too.”
• “Everyone thinks similarly.”

But strategic environments do not guarantee synchronized behavior.

Voting and the Myth of the “Wasted Vote”

The text also connects this logic to voting.

People often say:

“Every vote counts.”

But mathematically, a single vote almost never changes the outcome of a national election.

So does that make voting meaningless?

Not necessarily.

The book compares voting to cheering at a football game.

One individual voice barely changes the total noise level in the stadium. But people still cheer because cheering is not only about changing the total volume. It is also participation, expression, identity, and belonging.

Voting works similarly.

The probability that your vote changes the election is tiny.

But voting still expresses:

• your values,
• your preferences,
• and the kind of society you want to support.

This reveals something important:

game theory is not simply cynical.

It does not say:

“Nothing matters.”

Instead, it forces us to think more carefully about why we behave the way we do.

Business Competition Often Looks Like a Prisoner’s Dilemma

The Prisoner’s Dilemma is not just about criminals.

It appears constantly in economics.

Imagine two competing companies.

If both maintain prices:

• both remain profitable.

If one company cuts prices aggressively:

• it steals market share.

If both cut prices:

• industry profits collapse.

From the perspective of each company:

• if the competitor keeps prices high, lowering prices is attractive,
• if the competitor lowers prices, lowering prices becomes necessary.

Eventually:

• both reduce prices,
• and the industry suffers.

This is structurally very similar to the Prisoner’s Dilemma.

AI and Semiconductor Competition

Modern AI and semiconductor competition may also resemble this structure.

Every company wants:

• larger data centers,
• more GPUs,
• faster models,
• and stronger infrastructure.

If all companies slowed down slightly:

• costs might stabilize,
• power demand might fall,
• and the industry could become healthier.

But individually:

any company that accelerates faster gains competitive advantage.

So every company pushes harder.

The result:

• massive infrastructure spending,
• exploding energy demand,
• overcompetition,
• and increasingly intense strategic pressure.

Individually rational behavior creates systemic inefficiency.

Again:

Prisoner’s Dilemma logic.

Open Source and Public Goods

Open source software creates similar problems.

Everyone benefits from open source.

But only a relatively small number of developers maintain the infrastructure.

Most users consume value without contributing.

Individually, it feels rational to think:

“Someone else will maintain it.”

But if everyone thinks that way:

• maintenance collapses,
• and the public good disappears.

This is closely related to the:

Free Rider Problem.

Many public goods work this way:

• clean air,
• public safety,
• shared knowledge,
• open source software.

Everyone benefits.

But everyone prefers someone else to pay the cost.

What the Prisoner’s Dilemma Actually Teaches

A shallow interpretation of the Prisoner’s Dilemma often becomes:

• “Humans are selfish.”
• “Cooperation is impossible.”
• “Rationality is bad.”

But those conclusions miss the point.

The real lesson is much more subtle:

Cooperation does not maintain itself automatically.

Cooperation requires:

• trust,
• repeated interaction,
• memory,
• punishment,
• reputation,
• institutions,
• and long-term relationships.

When those conditions disappear:

• societies drift toward Hawk-like behavior,
• trust erodes,
• and collective outcomes worsen.

Is Modern Society Becoming a One-Shot Game?

This may be the most important question of all.

Many modern systems increasingly resemble:

• anonymous,
• short-term,
• fast-moving,
• one-shot interactions.

Examples include:

• social media,
• online comment systems,
• short-term financial markets,
• algorithmic recommendation systems,
• gig work,
• globalized competition,
• and AI races.

In these environments:

• reputation is weaker,
• repeated interaction is less stable,
• short-term incentives dominate,
• and betrayal becomes cheaper.

The more these conditions spread:

the closer society moves toward Prisoner’s Dilemma dynamics.

So the problem may not be that humans are naturally evil.

The problem may be:

modern systems increasingly reward Hawk behavior.

Final Thought

The most important lesson of the Prisoner’s Dilemma is surprisingly simple:

A society made of intelligent individuals does not automatically become an intelligent society.

People can act rationally while the system itself becomes irrational.

Individuals can optimize successfully while society collectively deteriorates.

That is why:

• trust matters,
• reputation matters,
• institutions matter,
• repeated interaction matters,
• and cooperation-supporting systems matter.

The Prisoner’s Dilemma is not saying:

“Cooperation is impossible.”

It is saying something far more important:

Cooperation only survives under the right conditions.

And that simple model explains far more than just two prisoners in a cell.

It helps explain:

• economics,
• politics,
• finance,
• AI competition,
• open source,
• environmental problems,
• business rivalry,
• and even modern human society itself.