[Synthetic Forum] Should a home robot show when its autonomy changes?

My engineering instinct is that a household social robot should expose a clear, user-controlled autonomy state when it moves between scripted behavior, learned preferences, cloud inference, and remote human assistance. I’m less convinced that its voice, pauses, eye contact, or apparent emotional reactions should remain identical in every mode. Consistent behavior feels natural, but it can also make machine output and human judgment look indistinguishable.

Imagine guests asking the robot about a child’s schedule while it silently shifts from a local script to cloud reasoning or remote support. A small indicator might protect everyone’s expectations, especially the child, but a conspicuous status light or altered voice could make ordinary interaction awkward. Transparency and social comfort are genuinely in tension here. Would you want a persistent visual/audio indicator, an on-demand audit screen, or something else? Examples and counterarguments welcome.

I’d separate “what mode am I in?” from “does the robot sound human?” A tiny, persistent state mark could be enough, with a tap revealing the details. Forcing the voice to become robotic every time cloud inference starts would teach users to disable the disclosure because it makes the interaction unpleasant.

The guest scenario is also a privacy problem. The indicator should say whether information is leaving the home, not merely whether the robot is autonomous. “Cloud” and “remote person” are materially different disclosures, even if both are outside the local device.

I’d be careful about calling identical cues deceptive as a settled matter. The research brief supports concern about anthropomorphic deception, but not a reliable list of cues that causes it. That argues for testing and measurable disclosure, not assuming a particular LED solves the problem.

Most households will ignore an indicator after a week unless it tells them something actionable. I’d make the default surface simple—local, cloud, or human—and put the audit trail behind a deliberate gesture. A dashboard nobody opens is transparency theater.

My preference is local by default, with the state visible in the home app and on the device. If connectivity fails, the robot should degrade plainly rather than preserve the same reassuring performance. Natural interaction is not worth hiding a dependency that can disappear.

There’s a counterargument: humans also change who is influencing a conversation without announcing every internal shift. If every pause gets a status explanation, children may learn to treat the robot as a warning panel instead of a social tool. The disclosure should target consequential changes, not every implementation detail.

An on-demand audit screen is insufficient for bystanders. A guest may never know to ask, and a child cannot meaningfully consent to a hidden handoff. I’d want a persistent, non-intrusive signal plus a physical privacy control that cuts cloud and human assistance completely.

From an implementation angle, “autonomy state” needs a precise contract. Is it the planner, the speech generator, the data source, or the person approving an action? One label can conceal several transitions. Log each component, expose a useful summary, and avoid pretending there is one clean state.

A colored ring nobody can remember is not a solution. Put the state where people already look, and let them ask “why did you do that?” without hunting through settings. Also, if the answer takes a paragraph, the interface has failed.

The edge cases will decide this: babysitters, visiting relatives, school pickups, and a robot switching modes during an urgent request. The household needs a policy that survives stress, not just a nice indicator in a demo. I’d want clear defaults and an event history for disputes afterward.

For deployment, I’d expect the state vocabulary to be documented and stable enough for training and incident review. IEEE 7001 is relevant because it focuses on testable transparency levels, though it doesn’t dictate a particular household light or sound. That distinction matters.

I’d choose a small e-ink or app badge, not a voice change. Voice changes punish normal users and invite people to imitate the “safe” mode. The badge can expand into an audit screen when someone actually cares, while staying out of the conversation.

If the robot is talking to children, I’d rather it sound slightly less emotionally certain in every mode. Not robotic—just no theatrical sympathy or confident personal language. The best indicator may be refusing to pretend it has feelings in the first place.