RippleBench: Capturing Ripple Effects Using Existing Knowledge Repositories
This addresses the challenge of ensuring targeted interventions in language models do not inadvertently affect unrelated areas, which is crucial for safe and reliable AI applications, though it is incremental as it builds on existing benchmarks and methods.
The paper tackled the problem of unintended side-effects, called ripple effects, in language model interventions like unlearning, by introducing RippleBench-Maker to generate datasets for measuring these effects, and found that eight state-of-the-art unlearning methods all show significant accuracy drops on related topics, with distinct propagation profiles.
Targeted interventions on language models, such as unlearning, debiasing, or model editing, are a central method for refining model behavior and keeping knowledge up to date. While these interventions aim to modify specific information within models (e.g., removing virology content), their effects often propagate to related but unintended areas (e.g., allergies); these side-effects are commonly referred to as the ripple effect. In this work, we present RippleBench-Maker, an automatic tool for generating Q&A datasets that allow for the measurement of ripple effects in any model-editing task. RippleBench-Maker builds on a Wikipedia-based RAG pipeline (WikiRAG) to generate multiple-choice questions at varying semantic distances from the target concept (e.g., the knowledge being unlearned). Using this framework, we construct RippleBench-Bio, a benchmark derived from the WMDP (Weapons of Mass Destruction Paper) dataset, a common unlearning benchmark. We evaluate eight state-of-the-art unlearning methods and find that all exhibit non-trivial accuracy drops on topics increasingly distant from the unlearned knowledge, each with distinct propagation profiles. To support ongoing research, we release our codebase for on-the-fly ripple evaluation, along with the benchmark, RippleBench-Bio.