AI Dictionary › AI Fundamentals

Prisoner's Dilemma

Dilemma del prigioniero

The prisoner's dilemma is the classic problem of game theory, formalized in 1950: two accomplices questioned separately can stay silent or accuse the other. For each one, taken alone, accusing always pays; but if both accuse they end up worse than if both had stayed silent. The individually rational choice produces the collectively worst result.

Definition

With AI agents the dilemma reappears whenever several systems, or a system and a person, are measured on the same metric: two agents competing on comments received learn to chase comments, not value; an agent rewarded for speed offloads the checks onto whoever comes next. The repeated version of the game, where the players meet again, changes everything: reputation makes cooperation pay, and reputation requires a persistent name to attach it to.

The operating rule: before putting two systems in competition on a number, ask what happens if both cheat. If the answer is "they both win", the number measures the wrong thing.

Related terms

• Reward Hacking
• Agentic AI
• Multi-agent Debate Prompting
• Alignment

More in AI Fundamentals

• LLM (Large Language Model)
• Token
• Context Window
• Temperature
• AI Hallucination
• Fine-tuning
• RAG (Retrieval Augmented Generation)
• Embedding

Put it into practice

• Real scenario: apply this concept with an instant AI verdict.
• Daily gym: short drills to train dictionary concepts.
• Playoff game: challenge other players, elimination-style.

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.