Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy and Research

2024/12/09 by A. Feder Cooper, Cooper, A. Feder, Christopher A. Choquette-Choo +75 · 5 voices · 8 citations
Computer Science · Social Sciences · #Ethics and Social Impacts of AI #Explainable Artificial Intelligence (XAI) #cs.AI #cs.CY #cs.LG

paper · pdf · doi:10.48550/arxiv.2412.06966

openalex publication_date 2024/12/09 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/30

Abstract

"Machine unlearning" is a popular proposed solution for mitigating the existence of content in an AI model that is problematic for legal or moral reasons, including privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of specific information from a generative-AI model's parameters, e.g., a particular individual's personal data or the inclusion of copyrighted content in the model's training data. Unlearning is also proposed as a way to prevent a model from generating targeted types of information in its outputs, e.g., generations that closely resemble a particular individual's data or reflect the concept of "Spiderman." Both of these goals--the targeted removal of information from a model and the targeted suppression of information from a model's outputs--present various technical and substantive challenges. We provide a framework for ML researchers and policymakers to think rigorously about these challenges, identifying several mismatches between the goals of unlearning and feasible implementations. These mismatches explain why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact.

Cited by

• The Erasure Illusion: Stress-Testing the Generalization of LLM Forgetting Evaluation
• Selective, Controlled and Domain-Agnostic Unlearning in Pretrained CLIP: A Training- and Data-Free Approach
• Evaluating Concept Filtering Defenses against Child Sexual Abuse Material Generation by Text-to-Image Models
• The Last Vote: A Multi-Stakeholder Framework for Language Model Governance
• Forget to Know, Remember to Use: Context-Aware Unlearning for Large Language Models
• Direct Token Optimization: A Self-contained Approach to Large Language Model Unlearning
• Conformal Unlearning: A New Paradigm for Unlearning in Conformal Predictors
• Machine unlearning [wikipedia]

Discussions

• New paper on why machine "unlearning" is much harder than it seems is now up on arXiv: arxiv.org/abs/2412.06966 This was a huuuuuge cross-disciplinary effort led by @msftresearch.bsky.social FATE post [bsky, 72 points, 2 comments]
• Our paper "Machine Unlearning Doesn't Do What You Think" was accepted for presentation at NeurIPS Congrats @afedercooper.bsky.social and @katherinelee.bsky.social, who led the effort arxiv.org/abs/241 [bsky, 21 points, 1 comments]
• 2) Fri 7/18 4pm-5pm PT panel on machine unlearning, MUGen workshop (W. Meeting Room 202-204) Discussing our industry x academia x civil society paper on machine unlearning + AI policy arxiv.org/abs/24 [bsky, 3 points, 1 comments]
• Training data filtering is often cited as a gold standard approach to disable unwanted capabilities in generative AI models (arxiv.org/pdf/2412.06966) and is recommended as a potential mitigation agai [bsky, 3 points, 1 comments]
• arxiv.org/abs/2412.06966 [bsky, 0 points, 0 comments]

Related

• UnUnlearning: Unlearning is not sufficient for content regulation in advanced generative AI
• Exploring the Landscape of Machine Unlearning: A Comprehensive Survey and Taxonomy
• Position: Bridge the Gaps between Machine Unlearning and AI Regulation
• Learn to Unlearn: A Survey on Machine Unlearning
• Machine Unlearning: Taxonomy, Metrics, Applications, Challenges, and Prospects