The Lineup

Opening Scene

Sometimes the full, precise timeline is more detail than a case actually needs. What a jury really needs is a clear, simplified reenactment of what happened — accurate enough to be trustworthy, simple enough to actually follow. That’s a different, complementary approach to the same problem SHAP tackles: explaining one specific decision, but through a simplified local stand-in rather than a precise, game-theoretic accounting.

That’s what LIME does.

In Plain English

LIME (Local Interpretable Model-agnostic Explanations) explains one prediction by training a simple, interpretable model — usually a small linear model — that closely approximates the complex model’s behavior in the immediate neighborhood of that one prediction. It doesn’t try to explain the whole model everywhere; it builds a locally faithful, easy-to-read stand-in for just this one decision.

The Old Way

Before LIME formalized this, people used similar local-approximation instincts in less rigorous ways:

A teacher simplifying a complex topic into an analogy for one specific student’s question, accurate enough for that context without being universally precise.
A lawyer explaining a complex ruling through a simplified hypothetical tailored to the specific facts of one case.
An engineer using a simplified local model of a complex system to reason about behavior in one specific operating condition, without needing the full, complex model to be understandable everywhere at once.

The core idea — simplify locally rather than globally — long predates any formal machine learning technique.

What’s Changing (and Why AI Is the Reason)

LIME formalized “simplify locally” into a model-agnostic technique that works on any black-box model, whether it’s a deep neural network, a large ensemble, or something else entirely.
Because LIME doesn’t need to know anything about a model’s internals, it became one of the first widely usable explanation tools for genuinely opaque models, ahead of many method-specific alternatives.
LIME and SHAP are now often used together in practice, since a locally faithful simple model (LIME) and a precise, game-theoretically grounded attribution (SHAP) can each catch things the other might miss.

The Metaphor, Fully Extended

The Investigation	LIME Concept
A simplified reenactment of one specific incident	A simple local model approximating one prediction
The reenactment being accurate for this case, not every case	LIME’s local, not global, fidelity
Working from the outside without needing to know a suspect’s inner thoughts	LIME’s model-agnostic approach
Using a reenactment alongside a full forensic timeline	Using LIME alongside SHAP for a fuller picture

For Beginners: What to Actually Do

Try LIME on a model you already have SHAP explanations for, and compare — noticing where they agree builds confidence in both.
Remember LIME’s explanation is only locally faithful; don’t extrapolate it to describe how the model behaves everywhere.
Use LIME specifically when you need a model-agnostic technique and don’t know or control the model’s internal type.

For Practitioners and Leaders: The Deeper Layer

Consider LIME and SHAP complementary tools in the same investigative toolkit, not competing choices — use both when a decision genuinely matters.
When onboarding a new team onto explainability tooling, LIME’s simple linear stand-in models are often the easiest entry point before more mathematically involved techniques.
Be explicit with stakeholders that a LIME explanation describes local behavior around one decision, not a universal account of the model.

Quick Recap

LIME explains one prediction by fitting a simple, interpretable model in the local neighborhood of that prediction.
It’s model-agnostic, working on any black-box model regardless of internal structure.
LIME and SHAP are complementary local-explanation techniques, often used together.
A LIME explanation is only locally faithful and shouldn’t be extrapolated globally.

Where This Fits in the Series

Articles 5 and 6 covered two complementary techniques for explaining single decisions. Article 7 looks at a different, forward-looking kind of explanation: what would have had to change for the outcome to be different.