This is the current revision of this page, as edited by Cielquiparle (talk | contribs) at 14:15, 25 December 2024 (+ref). The present address (URL) is a permanent link to this version.
Revision as of 14:15, 25 December 2024 by Cielquiparle (talk | contribs) (+ref)(diff) ← Previous revision | Latest revision (diff) | Newer revision → (diff) Field of study in artificial intelligence | This article needs additional citations for verification. Please help improve this article by adding citations to reliable sources. Unsourced material may be challenged and removed. Find sources: "Machine unlearning" – news · newspapers · books · scholar · JSTOR (December 2024) (Learn how and when to remove this message) |
Machine unlearning is a branch of machine learning focused on removing specific undesired element, such as private data, outdated information, copyrighted material, harmful content, dangerous abilities, or misinformation, without needing to rebuild models from the ground up.
Large language models, like the ones powering ChatGPT, may be asked not just to remove specific elements but also to unlearn a "concept," "fact," or "knowledge," which aren't easily linked to specific examples. New terms such as "model editing," "concept editing," and "knowledge unlearning" have emerged to describe this process.
History
Early research efforts were largely motivated by Article 17 of the GDPR, the European Union's privacy regulation commonly known as the "right to be forgotten" (RTBF), introduced in 2014.
Present
The GDPR did not anticipate that the development of large language models would make data erasure a complex task. This issue has since led to research on "machine unlearning," with a growing focus on removing copyrighted material, harmful content, dangerous capabilities, and misinformation. Just as early experiences in humans shape later ones, some concepts are more fundamental and harder to unlearn. A piece of knowledge may be so deeply embedded in the model’s knowledge graph that unlearning it could cause internal contradictions, requiring adjustments to other parts of the graph to resolve them.
References
- "Machine Unlearning in 2024". Ken Ziyu Liu - Stanford Computer Science. Archived from the original on 2024-12-13. Retrieved 2024-12-24.
- Hine, E.; Novelli, C.; Taddeo, M. (2024). "Supporting Trustworthy AI Through Machine Unlearning". Science Engineering & Ethics. 30 (43). doi:10.1007/s11948-024-00500-5.