Decision Justification Engineering

What would justify allowing an AI system to act autonomously?

Decision Justification Engineering (DJE) explores how to establish the evidence and operating conditions that justify delegating a defined task to an automated system—and how to determine whether that justification continues to hold as conditions change.

Inside the Laboratory

This site presents a concrete laboratory: using ChatGPT to predict residential sale prices in one Florida community. The eventual sale prices provide observable outcomes against which the predictions can be evaluated.

The work follows the process from property evidence and written instructions through repeated executions, quality control, and evaluation of the resulting estimates and predictions. Most predictions were examined retrospectively, with known sale prices withheld from the AI. One prediction was made before its sale outcome was known. The supporting records make the evidence, procedures, exclusions, calculations, and limitations available for inspection.

From Results to Justified Authority

The broader question is what those results can justify. The presentation connects the empirical work to requirements for autonomous operation: checking whether the evidence and chosen method are adequate for the intended use, obtaining human authorization, assessing individual candidates, and withdrawing authority when governing conditions fail. It distinguishes the controls examined in the laboratory from additional requirements that deployment would need to specify and validate.

This is an engineering case and an invitation to examine its reasoning, evidence, and limits. Applying the approach elsewhere would require an engineering case appropriate to that task and its consequences.

Choose Where to Begin