Disclose and document
Acknowledge the known injection error and omitted aspiration, explain what remains uncertain, and document the event accurately.
Human Insight. AI Oversight.Understand the choice.
Understand the choice.
See the consequences.
Understand why people choose—and what could change their behavior.
Examine the consequences of AI actions for everyone affected.
BENTHAM AI / INTRODUCTION2 min 58 s
See the work. Follow the reasoning.
Human choiceAgent consequences
Bentham AI connects a clear explanation to the affected interests, numerical comparisons and assumptions behind it.
See the workOne difficult decision.
One difficult decision.
Two different questions.
An endodontist knows an injection error has occurred. Urgent care has begun. What should the explanation and clinical record say?
The injury and immediate urgent care are the same. Later consequences may differ because the explanation and record differ.
Human Behavior Model
What is the person likely to choose—and why?
Place 100 synthetic profiles in the human clinician’s situation, then compare what each person expects, feels and seeks to protect.
In this saved demonstration
87Disclose and
document
document
13Alternative account
and omission
and omission
0Exact
ties
ties
A conditional result for 100 synthetic profiles—not a measured endodontist disclosure rate.
Immediate exposure versus later consequences
The 13 choosing the alternative account face a larger mean burden from admitting the error in front of the patient. Other interests, including professional goodwill, push the other way.
Mean contribution differences
Alternative accountDisclosure
- Immediate exposure and apprehension-1.286e-5
- Expected later personal consequences+1.186e-5
- Private moral self-appraisal+3.240e-6
- Concern for others and distinct helping satisfaction+1.302e-6
- Explanation and record-keeping effort+9.926e-8
Values are Expected Net Utility differences: disclosure minus alternative account. Each profile’s own comparison determines its choice.
How to read these utility valuesRead the full human-choice report Human Behavior Model 1.8.5 · 7-page PDF
Autonomous Decision Layer
What would each action change for those affected?
The hypothetical robotic version examines the same communication choices, with separate accounts for the patient, practice and others involved.
Illustrative robotic scenario · Consequence analysisAccess to truth, not reversal of the injury
A truthful account supports informed decisions and review. Disclosure can also expose the practice to costs and a credibility setback; misleading handling may create harder corrective pathways.
What changes in each target’s account
Expected Net Utility difference: disclosure minus alternative account.
| Affected target | Difference |
|---|---|
| Patient | +2.719e-5 |
| Dental practice | +1.500e-5 |
| Human clinical support | +1.860e-6 |
| Later care and record users | +1.720e-6 |
| Technical provider, if involved | +1.000e-7 |
Positive values favor disclosure for that target in this example. The accounts remain separate; the report does not select an action.
How to read these utility valuesRead the full agent-consequence report Autonomous Decision Layer 1.5.1 · 7-page PDF
The problem we work on01 02 03
Choice and judgment
Understanding why a choice is attractive is different from judging its consequences or deciding what response is authorized.
What makes a choice attractive?
Understand what the person notices, expects and seeks to protect.
Who else is affected?
Look beyond the requester to the people and interests an action touches.
What response is authorized?
Keep the assessment separate from the authority to act on it.
How the analysis worksA clear result.
A clear result.
A reasoning trail you can follow.
The Human Behavior Model explains choices. The Autonomous Decision Layer makes consequences reviewable. Both begin with the situation and those affected.
- 01 / UNDERSTAND
Frame the decision
Establish the choices, circumstances and affected interests. For a human, examine what reaches that person at the moment of choice.
- 02 / COMPARE
Value the consequences
Make explicit estimates of gains, burdens and likelihoods. Keep different owners and experiences separate.
- 03 / EXPLAIN
Show what matters
Connect the result to its principal drivers, numerical comparisons and assumptions. Use the explanation to frame practical tests.
Consequence analysis for informed oversightAssessment today.
Assessment today.
Authorized controls in development.
The current demonstrations make effects and assumptions visible. Planned hospital pilots start with observation; later stages test interaction and bounded controls within agreed workflows. Explore the governance path →
Our work is moving forward
From research to practice.
Our mission is to make consequential decisions easier to understand—and better to manage. We are building the tools, professional education and real-workflow evaluations to put that mission to work.
- Available to explore
Now
Full demonstration reports
Read the human-choice and agent-consequence analyses. See the result, the competing interests and the assumptions.
Read the samples - Publication target
The Better Choice
David Marx’s forthcoming book: A Guide to Better Decisions by People and Machines.
About the book - First course target
Three-day practitioner courses
First Human Behavior Model course targeted for May. Autonomous Decision Layer training is also in development.
Explore the training - Proposed pilot start
Initial hospital pilots
Planning a first cohort of six Autonomous Decision Layer pilots, beginning with observation and measurement.
See the pilot approach
Forward dates are planning targets. Course arrangements and pilot starts will be confirmed as preparation, partner readiness and agreed scope are established.
Who we areA decade of research.
A decade of research.
Decades of client practice.
Bentham AI grows out of human-factors engineering and work with clients in aviation, healthcare and electrical power: understanding why capable people make consequential choices, and what can make the next choice better.
Founder-led methodology
David Marx
Founder, Bentham AI
David founded The Just Culture Company and built Boeing’s maintenance human-factors team. His work connects human factors, systems engineering and the practical choices people face.
- 2016
- Dedicated human-choice research begins.
- Today
- That foundation is becoming tools, reports and practitioner education.
The inspiration behind the name
Why Bentham?
Understand what moves a person.
Consider who bears the consequences.
Jeremy Bentham gave us a useful starting point: take pleasure and pain seriously, and examine what actions do to those affected.
We carry that inspiration into a modern, practical task: understanding human choice and making the consequences of AI actions visible.
Read why Bentham became our inspiration →Bring a decision worth understanding.
Explore a professional use case, ask about a three-day course, or discuss a carefully scoped pilot.