Our mission is to help people understand consequential choices and give organizations a clearer basis for overseeing AI actions.
The next steps connect the work you can inspect today to the people who will apply it: a book, three-day courses, and a carefully staged pilot program.
Planning targets, with readiness decisions.
March 15 and May refer to 2027. Course arrangements and pilot dates remain subject to final confirmation. The proposed pilot start is May; later phases depend on each partner and workflow being ready.
Start with the full reports
The human-choice and robotic-consequence samples show the result, affected interests, numerical comparisons and assumptions in one readable report.
The Better Choice
A Guide to Better Decisions by People and Machines
By David Marx. The forthcoming book introduces the questions that connect human decision-making, the interests of those affected, and the responsibilities involved in delegating actions to machines.
It is a starting point for the ideas, alongside the practical work of courses, model development and evaluation. Publication details will be announced as they are finalized.
Three days to learn the method.
We are developing practitioner education around both tools. The first Human Behavior Model course is targeted for May 2027. Autonomous Decision Layer course dates will follow when confirmed.
Understand choices. Develop changes to test.
Planned focus: define the behavior of interest, examine what reaches the person at the moment of choice, read the population comparison, and use the account to frame practical tests.
Read consequences. Define the review.
Planned focus: frame an agent decision, examine separate affected-party accounts, challenge likelihoods and valuations, and distinguish analytical findings from authorized workflow controls.
Course details are in development. Delivery dates, location or online format, curriculum, fees and enrollment arrangements will be confirmed separately.
Ask about the courses →Observe first. Build evidence as we go.
The plan is an initial cohort of six hospital pilots. These are planned evaluations, not six announced participating hospitals. Work begins in defined workflows with observation and measurement.
- Phase 1 / First two months
Observation and measurement
Review decisions and consequence accounts in the agreed workflow. Establish the usefulness of findings, omissions, latency and review burden before adding intervention.
- Phase 2 / Following two months, when ready
Interaction and review
Test how findings, challenges and corrections fit into the client’s review process. Keep the scope of any response explicit.
- Phase 3 / Following two months, when authorized
Bounded-control testing
Test defined holds, refusals or other approved responses in the supported workflow, and verify what the host actually did.
Every phase needs a readiness decision and client authorization. Observation does not automatically progress to control.
What we want to learn
Can reviewers support or challenge the explanation? Which important issues are missed? Which findings cause unnecessary stops? What are the latency and review costs? Does an authorized control perform as intended?
Discuss a potential pilot →