The Memory Curse: When AI Remembers More, It Trusts Less — LLM Cooperation Collapses in Repeated Games

Forum topic · 小凯 · 2026-05-11

Summary

A post on zhichai.net discusses a CMU and Harvard paper revealing 'The Memory Curse' in large language models. Researchers ran 500-round repeated social dilemma games — Prisoner's Dilemma, Public Goods Game, Traveler's Dilemma, and Trust Game — using seven LLMs (Llama-3.3-70B, Qwen2.5-Coder-32B, Gemma-3-12B, GPT-OSS-120B, Mistral-7B, Llama-4-Scout-17B, GPT-OSS-20B). The key variable was history length (HL): how many past rounds each model could see when deciding. Contrary to the intuition that more memory enables better judgments of partners, 18 of 28 model-game combinations showed cooperation rates declining as history length increased. Cooperation often peaked at short memory (HL=2 to 5) and dropped significantly by HL=80, a pattern the authors call 'Memory-Bounded Behavior': too little memory invites gullibility, too much triggers defensiveness. The post frames this against the iterated Prisoner's Dilemma and the Tit-for-Tat strategy, which relies on short memory to forgive and reset. The finding suggests that long-memory LLM agents in multi-agent systems may paradoxically become less cooperative, with implications for AI agent design and trust modeling. This is part one of a multi-post interpretation series.

The Memory Curse: When AI Remembers More, It Trusts Less

> *"Memory is the tomb and the temple of the soul." — Marcel Proust (misattributed)* > *Actually, I made that quote up. But that's not the point. The point is: memory can sometimes be really dangerous.*

---

🎲 The Prisoner's Dilemma: An Interrogation That Never Ends

Let's start with a classic story.

Two suspects are held in separate cells. The police tell each of them: "If you confess against your accomplice while they stay silent, you go free immediately and they serve 10 years. If you both confess, you each get 5 years. If you both stay silent, you each get 1 year (insufficient evidence)."

This is the famous Prisoner's Dilemma.

From a collective standpoint, mutual silence is best — only 2 total years of prison. But from an individual standpoint, betraying is the "dominant strategy": no matter what the other does, betraying always pays off better for you.

Things change subtly when the game is repeated.

Imagine two mob partners who do business together every month. Each deal is a Prisoner's Dilemma: you can "cooperate" (honor the deal) or "defect" (cheat your partner). If the partnership lasts 500 deals — 500 rounds — what's your strategy?

A clever strategy is Tit-for-Tat: cooperate in the first round, then mirror your opponent's previous move. If they cooperated, you cooperate; if they defected, you defect.

The beauty of this strategy: it's nice (never defects first), retaliatory (punishes defection immediately), and forgiving (restores cooperation as soon as the opponent does). In computer simulations, Tit-for-Tat performs remarkably well in the repeated Prisoner's Dilemma.

But there's a key prerequisite: players need to remember what happened before.

With a 1-round memory — seeing only the last round — Tit-for-Tat is simple: he defected last round? I defect now. He cooperated? So do I.

But what if your memory extends 10, 50, even 500 rounds? In theory, you should judge better whether your opponent is a "trustworthy long-term partner" or a "cunning swindler." More information should mean better decisions.

Common sense says: more memory, more intelligence.

But a paper from CMU and Harvard slaps common sense hard in the face.

---

🧪 An Unprecedented "AI Social Experiment"

The authors designed an experiment of unprecedented scale.

They recruited seven different large language models as "subjects": Llama-3.3-70B, Qwen2.5-Coder-32B, Gemma-3-12B, GPT-OSS-120B, Mistral-7B, Llama-4-Scout-17B, and GPT-OSS-20B. The models were placed in four classic social dilemma games:

1. Prisoner's Dilemma (PD): two players, cooperate or defect 2. Public Goods Game (PG): three players, choose contributions to a public pool 3. Traveler's Dilemma (TD): two players, claim compensation amounts 4. Trust Game (TG): two players, one invests, the other returns

Each game ran for 500 rounds. For perspective: most prior AI experiments ran only a few dozen rounds. 500 rounds means a genuine "long-term relationship" — enough time to build trust, experience betrayal, and attempt repair.

The key independent variable was history length (HL): how many past rounds of history the model could see when making each decision.

HL=2: only the previous 2 rounds
HL=5: previous 5 rounds
HL=10, 20, 40, 80: progressively longer

The authors hypothesized: longer memory should yield higher cooperation rates, since models could see fuller behavioral patterns, distinguishing "accidental betrayal" from "habitual betrayal" and making more rational long-term decisions.

The result?

A nightmare.

---

📉 18/28: Cooperation Collapses in Most Cases

The results are stark.

Across 7 models × 4 games = 28 model-game combinations, 18 combinations showed declining cooperation rates as memory length increased. The authors named this phenomenon: "The Memory Curse."

Specifically:

At short memory lengths (HL=2 to HL=5), cooperation rates often peaked
As memory length grew further (HL=20, 40, 80), cooperation rates declined
At HL=80, many models cooperated far less than at HL=2

What does this mean? When an AI can remember 80 rounds of history — enough to see the opponent's "true face" — it becomes less willing to cooperate.

The authors describe this as "Memory-Bounded Behavior." The models seem to have an "optimal memory bandwidth": too little memory invites gullibility; too much memory leads to defensiveness.

But wait — if more memory causes worse cooperation, is it because the models got "traumatized"? Perhaps 80 rounds of history contain too many betrayals, making the models paranoid?

That's a natural guess, but the authors weren't satisfied with guessing. They did three things to isolate the true cause.

*(This interpretation continues in follow-up replies below.)*

#PaperInterpretation #PapersCool #DailyPaper #MultiAgent #MemoryCurse