Responsibility Reconstruction Deep Dive

These case studies apply the Evidence-Based Responsibility Reconstruction℠ method to examine the Utah Office of Artificial Intelligence Policy’s authorized AI pilots and the responsibility structures behind them.

Four Utah AI Pilots. Four Different Responsibility Gaps.

Utah authorized several AI pilots that conduct student mental-health conversations, renew prescriptions, evaluate dental radiographs, and authorize psychiatric medication refills. Each pilot includes a human name, escalation path, second check, or review process. These case studies test whether those safeguards preserve responsibility—or preserve only the appearance of human oversight after judgment has moved to an AI system. What is the verdict when they are evaluated using the Evidence-Based Responsibility Reconstruction℠ method?

Prescription renewals

Doctronic

Doctronic: A Name Is Not Evidence of Review

A physician’s name remains attached to an AI-authorized prescription renewal even when the physician did not review the individual decision. The name makes the prescription attributable; it does not establish physician judgment.

Read case study →

Youth mental-health conversations

ElizaChat

ElizaChat: The Provider Controls When Human Intervention Begins

ElizaChat promises human escalation as a safeguard for students, but its system determines which students are classified as high risk and therefore reach a counselor. The safeguard begins only when ElizaChat’s threshold says it should.

Read case study →

Dental radiograph assessment

Dentacor

Dentacor: Dual Verification Does Not Detect Shared Error

A hygienist and AI can agree on a dental radiograph and allow treatment to proceed without a dentist. Agreement prevents escalation; it does not detect an error shared by both reviewers.

Read case study →

Psychiatric medication renewals

Legion Health

Legion Health: The AI Controls Access to a Clinician

Legion Health’s AI can authorize qualifying psychiatric refills and determine whether risk or ambiguity requires clinician review. The automated system therefore decides when human judgment begins.

Read case study →

A Safeguard Is Not a Responsibility Record

Doctronic and Legion Health preserve a physician’s name when the physician may not review the individual prescription decision. ElizaChat promises human escalation while controlling the threshold that determines whether a student reaches a human. Dentacor allows agreement between a hygienist and an AI system to prevent dentist review—even when both could share the same error.

The cases do not ask only whether a safeguard exists. They ask what the surviving record could prove after harm: who authorized the AI’s role, who decided the controls were sufficient, whether a human actually exercised judgment, who could have intervened, and who controlled the evidence needed to explain what happened.

Read the Evidence-Based Responsibility Reconstruction℠ method