Interpretability

Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary

The degree to which a model's inner workings are understandable by design, so a person can trace how inputs become outputs without post-hoc explanation tools.

What is Interpretability?

Interpretability is the degree to which a person can understand how a model turns inputs into outputs by inspecting the model itself, because its structure is simple enough to read directly.

Decision trees, linear and logistic regression, and small rule sets are interpretable in this sense: their logic is the model. The property matters most where decisions must be checked and contested, in credit, medicine, and criminal justice.

How Interpretability Works

An interpretable model earns the label through structure. A linear model states each feature’s effect as a weight you can read off: one more year of credit history adds this many points to the score. A decision tree is a chain of if-then splits you can trace by hand.

Keeping models this readable imposes discipline on the builders: few features, no tangled interactions, relationships a person can hold in their head.

Interpretability comes in two scopes. Global interpretability means you understand the model’s entire logic, everything it will do on any input. Local interpretability means you understand one decision: why this applicant, this diagnosis, this flag.

Interpretability vs Explainability

An interpretable model is understandable by design; explainability reconstructs an understanding after the fact, using post-hoc tools such as SHAP or LIME on a model that is not readable itself.

The distinction has teeth in high-stakes settings. Cynthia Rudin’s widely cited 2019 argument holds that when decisions seriously affect people, institutions should build interpretable models rather than explain black boxes, because a post-hoc explanation can be a plausible story that fails to match what the model computed.

InterpretabilityExplainability
Where understanding comes fromThe model’s own structure, by designReconstructed after the fact, from outside the model
Works onSimple models: linear, trees, small rule setsAny model, including black boxes
Typical toolsNone needed; the logic is the modelSHAP, LIME, counterfactuals
Main riskSimplicity can cost accuracy on complex tasksExplanation may not match what the model computed

Example of Interpretability

A bank screens loan applications with a decision tree. The first split asks whether the debt-to-income ratio exceeds 40 percent; the second, whether credit history is shorter than two years; the third, whether income clears a minimum.

An applicant is declined, and the loan officer traces the exact path: debt-to-income was fine, but the credit history was fourteen months, which routed the application to a leaf labeled decline. Nothing was approximated or reconstructed. The officer read the decision straight out of the model and can tell the applicant precisely what to change.

Related AI terms: Explainability · Transparency · Black-box Model · Decision Tree

Did you like the Interpretability gist?

Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.

Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About

Read the Governor's Letter

Stay ahead with Governor's Letter, the newsletter delivering expert insights, AI updates, and curated knowledge directly to your inbox.

By subscribing to the Governor's Letter, you consent to receive emails from AI Guv.
We respect your privacy - read our Privacy Policy to learn how we protect your information.

Browse All AI Terms A–Z

Every term in the dictionary, in alphabetical order. Jump to a letter or scroll the full list.

A

B

C

D

E

F

G

H

I

J

K

L

M

N

O

P

Q

R

S

T

U

V

W

X

Y

Z