Paper · AI-assisted planning
Designing AI-assisted walking challenges
This paper describes a limited use of AI: preparing a draft walking challenge for human review. It does not propose that AI should decide what is healthy, pressure people to participate, or control their money.
Chapter 1
Scope and current status
The useful near-term question is not whether an AI can become an autonomous sponsor. It is whether AI can help a program owner prepare a clearer, safer draft in less time.
A draft may include participants, a period, a step goal, reminders, evaluation measures, and, if relevant, a proposed budget or recipient. Each item remains a proposal until an authorized person checks and approves it.
Chapter 2
Roles for AI and people
| AI may | A person must |
|---|---|
| Draft several challenge options | Choose, revise, reject, or stop |
| Suggest participant communication | Check facts, tone, accessibility, and consent |
| Flag missing fields or unusual burden | Make the final safety and feasibility judgment |
| Summarize pilot observations | Interpret results and decide next steps |
AI does not make medical, employment, legal, or eligibility decisions. Participants need a human contact for questions and withdrawal.
Chapter 3
Money, data, and approval
If a proposal uses money or tokens, AI may suggest the amount and distribution conditions. The person or authorized sponsor checks the amount, recipient, conditions, network, fees, and legal and accounting treatment, then initiates every wallet or payment action themselves.
A smart contract does not remove the need for review. Code, administrator and upgrade permissions, data inputs, pause mechanisms, and external dependencies can all change the risk.
Collect only the data required for the stated purpose. Tell participants what is collected, who can see it, how long it is kept, and how to withdraw or request deletion.
Chapter 4
Evidence and limits
Research on commitment devices, loss framing, and social motivation can inform a design. Results depend on the population, amount, duration, comparison condition, and implementation. They do not guarantee that a particular ESPL program will improve adherence.
A change in steps does not establish an effect on health, medical costs, productivity, or social isolation. Those outcomes require separate measures and an appropriate evaluation design.
Chapter 5
What a pilot should verify
- Whether the draft saved useful time and what human corrections were required
- Whether participants understood the purpose, terms, data use, and way to stop
- Whether the proposal created pressure, exclusion, or avoidable risk
- Whether data was complete enough for the planned evaluation
- Whether access, change, approval, and audit records worked as intended
- Whether to continue, revise, or stop before broader use
Early pilots should keep the population, period, data, and financial exposure small. Expansion is a new decision, not an automatic next phase.
Related work
For a practical pilot outline or a partnership discussion, see the related pages below.