Description
Explains an individual AI prediction by fitting a simple interpretable model to the black-box model's behaviour around that specific case.
LIME perturbs the input around a case of interest, records how the model's predictions change, and fits a sparse linear model weighted by proximity, yielding the features that most influenced that prediction locally. It is model-agnostic and works for tabular data, text (which words drove a classification) and images. For UX research, LIME helps audit specific decisions users complain about and prototype explanation content. Its limits are well documented: explanations can be unstable across runs, depend on perturbation and kernel width choices, and can be manipulated. Rigour requires repeated runs to check stability, comparison with SHAP, and validation that explanations help real users.
Attributes
- Discipline
- Computational & Machine Learning
- Loss type
- Siloed
- Stage
- Design, Evaluation
- Data
- Analytical / expert
- Effort
- Days
Steps
- Select cases to explain, prioritising complaints, edge cases and high-stakes decisions.
- Configure LIME with an appropriate perturbation strategy and kernel width for the data type.
- Generate explanations limited to 5-10 features for readability.
- Rerun each explanation 10 or more times with different seeds and check that the top features are stable.
- Compare with SHAP attributions and flag disagreements.
- Use stable explanations to audit decisions and draft user-facing explanation prototypes for testing.
Sources
- The Alignment Problem (Brian Christian)
- Artificial Intelligence for Humans (Jeff Heaton)
- Methods Atlas (Vivek Gopinath)
- Ribeiro, Singh & Guestrin (2016) Why should I trust you? Explaining the predictions of any classifier
- The Alignment Problem (Christian)
- AI for Humans (Heaton)
- Methods Atlas (Gopinath)
- The Inquiry Series, Ch 58 — AI Explainability & Value Alignment Audit
Also known as
- Local Interpretable Model-agnostic Explanations
Categories
- Analytics & Behaviour
- Computational & Machine Learning