Artificial Intelligence as Evidence

Paul W. Grimm, Maura R. Grossman, and Gordon V. Cormack

Northwestern Journal of Technology and Intellectual Property 19, 9 (2021)

A technically informed, step-by-step framework for deciding when AI-based outputs should be admitted as evidence.

PDF Publisher page Google Scholar BibTeX Citation

Overview

AI systems increasingly generate classifications, predictions, scores, and other outputs that may be offered in civil or criminal proceedings. This article explains how such systems work and translates their technical properties into familiar evidentiary questions.

The central distinction is between validity—whether a system accurately measures, classifies, or predicts what it claims to—and reliability—whether it produces consistent, accurate results under substantially similar conditions. The authors then examine bias, function creep, transparency, explainability, and the sufficiency of pre-deployment testing.

Why this paper matters. The article gives judges and lawyers a practical bridge between machine-learning concepts and the law of evidence. Rather than treating “AI” as a single technology, it directs attention to the particular system, task, data, testing, and use at issue.

Key contributions

Explains AI for a legal audience without assuming technical expertise.
Separates validity from reliability and shows why both matter.
Addresses bias, opacity, explainability, function creep, and objective testing.
Provides a structured analysis for offering, challenging, and admitting AI evidence.

Suggested citation

Paul W. Grimm, Maura R. Grossman & Gordon V. Cormack, Artificial Intelligence as Evidence, 19 Northwestern Journal of Technology and Intellectual Property 9 (2021).