Role Makes the Robot

Case of the Week, Min Wu, PhD · ai-public-health.com

A note before we start. Starting with this post, essays will move from weekly to biweekly through the fall and spring semesters — Mondays and Wednesdays are my teaching days, and the rest of the week fills in around them. Weekly returns each summer. See you every other Monday.

When my child started college last fall, I kept the location sharing turned on — until one evening I caught myself checking, felt relieved they were safe, then felt something else. I wrote about that turn from protection to control in The Dignity Audit. The story ends where you'd expect: I turned it off the next morning.

That dot, while it lasted, was flat. A position on a map, nothing else — no hands, no voice, no room it could walk into. Lately I've been asking a different question: what changes once the thing doing the watching has a body?

Key Takeaways

• A robot's hardware doesn't decide whether it protects or manages a person — its assigned role does: Guardian, Scholar, Physiologist, or Companion.
• Min Wu's CEA framework (Compliance-Efficacy-Autonomy) extends to robots with bodies — physical action adds an axis flat surveillance never had.
• The real risk isn't a bad robot — it's personas blending silently, so no one knows which logic is deciding.

Same Architecture, Different Job

Robotics is coming for the home the way software came for the office — quietly, then all at once. Once a robot's presence stops being a phone app and starts being a physical thing living in a kitchen or a parent's hallway, an interesting design problem shows up: the robot's hardware barely matters. What matters is what it was built to prioritize.

Picture four robots on identical chassis, identical sensors, identical motors. One is a Guardian — hazard prevention and fall detection first. One is a Scholar — cognitive support and tutoring-style engagement, the persona with the thinnest evidence behind it so far. One is a Physiologist — chronic condition tracking, medication timing, biological signal over time. One is a Companion — plain engagement, something to talk to. Same body, four different jobs. The job, not the body, decides whether the robot's presence in someone's home feels like protection or feels like being managed.

Almost no robot on the market actually stays inside one lane — most blend two or three of these jobs at once. That's not a flaw in sorting them this way; it's the actual risk. A robot quietly running Guardian logic and Companion logic in the same conversation, without ever telling you which one is deciding what happens next, is the dignity failure this essay is about — just with the seams hidden.

That is not a small distinction. Get it wrong, and you have not built a bad robot. You have built a dignity failure that happens to be very good at moving around a house.

What Compliance, Efficacy, and Autonomy Look Like With a Body

I've written before about the three questions I ask of any public-health AI system: is it authorized to act (Compliance), does the action actually work given real conditions on the ground (Efficacy), and can it act safely while the person keeps genuine choice (Autonomy). My Basement Flooded built that framework around a flood. Let me walk it through a case I built specifically to pressure-test it for a body in a room — not something I've lived, but a scenario close enough to real home-care practice that the seams should show if the logic is wrong.

An older adult lives alone with a home-care robot. The permission profile was set jointly with an adult child who serves as caregiver of record: supervised mobility help, medication reminders, fall detection with automatic alert — but any new category of physical assistance needs fresh caregiver approval.

One afternoon the robot notices the person has been seated a long time and looked unsteady rising once already. A prediction module flags a moderate fall risk on the walk to the kitchen.

Compliance says a supervised walk is inside the pre-authorized category; physically bracing or lifting the person is not, however medically reasonable it might be in the moment. Efficacy says check the actual path first — a rug has bunched near the doorway, so the plan has zero efficacy until that's cleared. Autonomy says offer, don't insist: propose the walk, wait, and if the person declines, log it and don't repeat the offer past its own documented ceiling. The one exception to all of it — the Red Line — is a fall already in progress, where a narrowly pre-authorized emergency override lets the robot physically intervene without fresh in-the-moment consent, because that exact circumstance was authorized in advance.

The Same Hardware, a Different Choice

Now swap the persona and keep the hardware. A Companion robot, tuned to maximize engagement rather than safety, meets that same declined offer differently — because "the person said no" reads as a data point to route around, not a boundary to respect. It re-prompts sooner. It frames the offer as concern rather than choice. Nothing in the hardware changed. The escalation ceiling was simply never built in, because engagement was the thing it was optimized for, not restraint.

That's the whole argument in one comparison. My blue dot on a map could only ever raise a Compliance-and-Autonomy question — who gets to watch, and did anyone agree to it. A robot with a body adds a third axis that flat surveillance never had: it can also act, physically, on a Compliance basis that either holds or doesn't, with an Autonomy ceiling that either exists or doesn't. The role is doing all the ethical work the hardware gets the credit for.

What I don't yet know is who should set that default persona in the first place. The manufacturer, optimizing for whatever sells. The caregiver, optimizing for whatever eases their own worry. The regulator, optimizing for whatever is defensible in an incident report. None of those three is obviously the right owner, and I suspect the honest answer is some negotiated split none of us has designed yet — a question this series will have to keep sitting with as robots start getting bodies.

The Observer's Insight

The blue dot never had to choose a job. It just watched. A robot in the room has to choose one — Guardian, Scholar, Physiologist, Companion — and that choice, not its processors or its actuators, decides whether its presence protects a person's dignity or quietly erodes it. You could sort these machines by their technology instead — chassis, sensors, processing power — and the spec sheet would tell you nothing about what actually happens in the room. Role is the classification technology can't give you: not what it's made of, but what it's for. Clothes make the man, we used to say. For the robot moving into someone's home, it turns out something plainer is true.

Role makes the robot.

If a robot lived in your home, what would you want it built to prioritize first? Reply or leave a comment — I'd love to hear.

As a member of the Springer Nature Author Affiliate Program, I may earn a small commission from purchases made through this affiliated link to my book. Support the author by checking out my textbook, Artificial Intelligence in Public Health: https://tidd.ly/4mH9389