Research map / Large language models / AI and society
Undermining Mental Proof: How AI Can Make Cooperation Harder by Making Thinking Easier
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Conceptual paper arguing that generative AI erodes "mental proof" (observable actions certifying unobservable mental facts) by lowering the cognitive cost of signaling and proof-of-knowledge behaviors, illustrated with apology and subculture examples.
Problem
Generative AI cheapens the mental effort behind social signals (apologies, applications, group membership), undermining how people credibly communicate private mental states in low-trust settings; existing work lacks an integrative framework explaining this across domains.
Approach
Introduces "mental proof" as a unifying concept and ties it to two mechanisms, costly signaling theory and proof-of-knowledge protocols from cryptography, arguing AI harms both by making the cognitive cost and the knowledge correlate of observable behavior disappear.
Key results
- Argues that AI collapses signaling power when it equalizes cognitive costs between high- and low-cost types, and only halves the difference when it halves both costs.
- Illustrates with sincere apology that people rate AI-heavy apologies as less sincere and less forgivable, while light tool use (spelling/grammar) carries no authenticity penalty (citing prior experiments).
- Table of examples (apology, subculture, cold e-mail, crisis hotline, technical interview) mapping how AI undermines signaling and proof-of-knowledge functions.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use heavy AI assistance to write an apology | Apology · Apology sincerity and forgiveness | perceived sincerity | Hurt |
| Use light AI assistance for spelling or grammar | Apology · Apology authenticity | perceived authenticity | No clear effect |
Surprising findings
- Socially costly signaling equilibria may be harmed by AI, and destroying such an equilibrium could in some cases improve outcomes.
Limitations
- No empirical experiments of its own; claims rest on literature and illustrative examples.
- Relies on assumptions about the cost structure of human mental activity that the paper itself notes are being disrupted.
- Mechanisms for replacing destroyed equilibria are left as an open question.
Experimental setup
Builds on: signaling theory interactive proof systems zero-knowledge proofs
Closest papers by meaning
- MetaMind: Modeling Human Social Thoughts with Metacognitive Multi-Agent Systems (NeurIPS 2025 · similarity 0.86)
- Theory of Mind and Persuasion Beyond Conversation: Assessing the Capacity of LLMs to Induce Belief States via Planning and Action (NeurIPS 2026 · similarity 0.86)
- Exploring Prosocial Irrationality for LLM Agents: A Social Cognition View (ICLR 2025 · similarity 0.85)
- Mental-Models for Multi-Agent Systems (NeurIPS 2026 · similarity 0.85)
- Many LLMs Are More Utilitarian Than One (NeurIPS 2025 · similarity 0.85)
- Thought Communication in Multiagent Collaboration (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.