English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

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

Forum topic · 小凯 · 2026-05-11

Summary

A CMU and Harvard paper reveals a counterintuitive finding in multi-agent LLM behavior: giving large language models longer memory histories actually reduces cooperation. In a large-scale experiment, seven LLMs (including Llama-3.3-70B, Qwen2.5-Coder-32B, Gemma-3-12B, GPT-OSS-120B, Mistral-7B, and Llama-4-Scout-17B) played four classic social dilemma games—Prisoner's Dilemma, Public Goods Game, Traveler's Dilemma, and Trust Game—for 500 rounds each, with varying memory lengths (HL=2 to 80). In 18 of 28 model-game combinations, cooperation rates declined as memory length increased, peaking at short memory lengths (HL=2–5) and dropping significantly at HL=80. The authors call this the 'Memory Curse' and describe the pattern as 'Memory-Bounded Behavior': models seem to have an optimal memory bandwidth—too little memory leads to gullibility, too much triggers defensive defection. The finding challenges the intuition that more information always improves decision-making, with implications for designing multi-agent AI systems.

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

> *"Memory is the tomb and the palace 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 betray each other, 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 served. But from an individual standpoint, betrayal is the "dominant strategy": no matter what the other does, betraying always pays off more.

The situation changes subtly when the game is repeated.

Imagine two gang members doing business together every month. Each deal is a Prisoner's Dilemma: you can "cooperate" (stay honest) or "defect" (cheat them). If the business runs for 500 deals—500 rounds—what's your strategy?

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

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

But there's a critical premise: players need to remember what happened before.

With only 1 round of memory—seeing just the last round—Tit-for-Tat is simple: they defected last round? I defect this round.

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

Common sense says: more memory, more intelligence.

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

---

🧪 An Unprecedented "AI Social Experiment"

The authors designed an experiment of unprecedented scale.

They recruited seven 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. These 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, contribute to a common 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 context: 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 previous rounds the model could see when making each decision.

  • HL=2: only the previous 2 rounds
  • HL=5: the previous 5 rounds
  • HL=10, 20, 40, 80: progressively longer
  • The authors hypothesized: longer memory should mean higher cooperation rates, because 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 alarming.

    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 increased further (HL=20, 40, 80), cooperation rates began to fall
  • At HL=80, many models' cooperation rates were far below their HL=2 levels
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 use a vivid term for this: Memory-Bounded Behavior. The models seem to have an "optimal memory bandwidth": too little memory breeds gullibility; too much breeds defensiveness.

But wait—if more memory leads to worse cooperation, is it because the models got "frightened"? Maybe 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 analysis is published in multiple parts—see replies below for continuation.)*

Tags

#llm#multi-agent#memory-curse#prisoners-dilemma#game-theory#cooperation#paper-explainer

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177619870