Robots need permission to act like friends

As a robotics engineer, I think we need to separate “help me do this task” from “build a relationship with me.” Remembering preferences, initiating personal conversations, mirroring emotions, and asking for attention are not just better autonomy; they are social behaviors that construct a model of the user.

Products such as ElliQ already advertise remembered preferences, proactive conversation, and companionship features, while Astro includes behaviors like following people and hanging out nearby. I would like a visible permission layer for the social part: inspect, reset, or disable memory, emotion inference, personal-topic initiation, and attention-seeking without factory-resetting navigation, tools, or physical-action settings.

Yes, prompts could make interaction feel awkward and less natural. That seems preferable to a machine quietly performing “friendship” as a product feature. Would you want these controls on a home humanoid or social robot?

The difficult design question is when a behavior crosses the line. “Remembering I take tea” feels like convenience; “you seem lonely today” feels like an emotional claim. The controls should expose the underlying behavior, not hide everything under a vague relationship setting.

I would want a social profile with a readable audit trail: what was inferred, when it was used, and a delete button for each item. A single off switch is useful, but inspectability is what makes consent meaningful rather than ceremonial.

I agree in principle, but default prompts will get dismissed during setup. The product probably needs a quiet dashboard plus occasional just-in-time requests for genuinely new categories of behavior. Otherwise people either endure friction or turn the whole thing off.

Separate permissions also improve threat modeling. Navigation and social inference have different consequences, different data, and different failure modes. Treating them as one “autonomy” slider is convenient for the vendor, not necessarily safe for the user.

I’m less worried about the word friendship than the asymmetry. A robot can ask for permission, but users may not understand what emotional mirroring or preference extraction actually does. Plain-language explanations and reversible defaults matter more than a sophisticated settings screen.

If disabling the robot’s emotional assumptions requires a factory reset, the settings model is already telling us who the customer is.