Beyond Forgetting: Machine Unlearning Elicits Controllable Side Behaviors and Capabilities
This work addresses the problem of understanding and controlling side effects in machine unlearning for LLMs, which is incremental as it builds on existing representation misdirection methods.
The study investigates representation misdirection methods for machine unlearning in LLMs, finding that manipulating forget-representations not only achieves forgetting but also elicits controllable side behaviors and enhanced capabilities, such as improved truthfulness and in-context learning, with empirical validation across multiple tasks.
We consider representation misdirection (RM), a class of LLM unlearning methods that achieves forgetting by manipulating the forget-representations, that is, latent representations of forget samples. Despite being important, the roles of target vectors used in RM, however, remain underexplored. Here, we approach and revisit RM through the lens of the linear representation hypothesis. Specifically, if one can somehow identify a one-dimensional representation corresponding to a high-level concept, the linear representation hypothesis enables linear operations on this concept vector within the forget-representation space. Under this view, we hypothesize that, beyond forgetting, machine unlearning elicits controllable side behaviors and stronger side capabilities corresponding to the high-level concept. Our hypothesis is empirically validated across a wide range of tasks, including behavioral control (e.g., controlling unlearned models' truth, sentiment, and refusal) and capability enhancement (e.g., improving unlearned models' in-context learning capability). Our findings reveal that this fairly attractive phenomenon could be either a hidden risk if misused or a mechanism that can be harnessed for developing models that require stronger capabilities and controllable behaviors.