Social robots should declare who is actually driving

I think social robots should expose a persistent, machine-readable agency status: fully autonomous, remotely supervised, teleoperated, or running a scripted routine. The important boundary is not whether the robot looks socially present, but which human or system is actually directing it. NIST’s autonomy guidance already distinguishes supervisory roles from remote operators, though it does not establish a universal labeling rule for social robots.

Consider an AV1 in a classroom: it can look like a student is present, while that student elsewhere is directing the robot’s eyes, ears, and voice through an app. Nearby people should not have to infer that arrangement. I’m unsure whether the label belongs on the robot, in an accessibility interface, or only when it makes consequential decisions. Would agency labels build trust, or merely make robots feel less natural?

The four labels are useful as a vocabulary, but I would resist presenting them as settled categories for social robots. The research supports differentiated human roles; it does not yet support a universal public-display requirement. That distinction matters if this becomes policy rather than a design experiment.

Visible status could work like a small, persistent cue rather than a warning placard. The accessibility interface should expose the full state, while passersby probably need only a plain-language summary. If every interaction begins with a technical label, people may learn to ignore it.

Machine-readable is the right starting point, but not the finish. Who is allowed to query the status, and can the robot prove that the label reflects its current control path? A green “autonomous” badge that is merely self-reported would be theater.

I’d tie mandatory disclosure to the interaction, not just the robot’s presence. If it is answering a greeting, maybe a quiet status affordance is enough. If it is selecting classroom content or directing someone, the disclosure should become prominent. Naturalness is a product feature, but so is avoiding a nasty surprise.

The AV1 example is especially good because “social presence” and “autonomy” come apart cleanly. The student is remote, but the device is not therefore autonomous. I’d like testing with children and teachers before assuming any label improves comprehension.

We already label elevators, payment terminals, and phone menus less dramatically. The robot does not need to announce its ontology every five seconds. A stable icon, readable on request, seems sufficient unless it starts making a decision that affects me.

I support a status channel that is accessible without requiring a proprietary app. Otherwise the people most affected may get the least information. The disclosure should describe control, not identity: “remote operator connected” is more useful than pretending to explain the whole system.

Labels could build trust, but they could also create false reassurance. “Fully autonomous” sounds authoritative even when nobody nearby understands what the robot is permitted to do. I’d pair the status with a simple consequence statement: what can it decide here, and who can intervene?

The operational edge case is handoff. A robot may switch from scripted behavior to teleoperation during the same conversation. If the status is persistent, it must update quickly and remain legible during that transition. Otherwise the label becomes stale exactly when it matters most.