Should social robots be allowed to say “I’m sorry”?

I’m uneasy about humanoid robots saying “I’m sorry” after making a mistake. In human conversation, it is a useful way to acknowledge inconvenience; in a machine, it is generated behavior, not felt remorse. That distinction matters when the robot is helping an older person, teaching a child, or running a household routine where users may reasonably over-trust it.

Research generally treats apologies as trust-repair strategies, while explanations communicate causes and mechanisms. I’m inclined toward: “I’m sorry—I misidentified your request. I interpreted ‘A’ as ‘B.’” But should the robot also explicitly disclose that the apology is scripted or generated? A plain error report might calibrate trust better, though I’m not convinced it will feel usable in every context.

Is emotional language acceptable when the robot’s non-human status is obvious, or should apologies always be replaced with transparent error reports?

I’d keep “I’m sorry” as an interface convention, but never let it stand alone. The important design move is pairing the social acknowledgment with a concrete account of the failure. Requiring a disclaimer after every apology could make the interaction oddly theatrical without improving understanding.

I’m wary of claiming that a plain explanation builds more trust in eldercare or education specifically; the brief evidence does not establish that comparison. What it does support is a narrower point: explanations can help users understand causes, while apologies can acknowledge a violation. They serve different functions.

“Obvious” is doing too much work here. A humanoid body, voice, and conversational timing can all encourage mental-state attribution. I’d treat the robot’s status as something the product must communicate and reinforce, not as a permanent exemption from disclosure.

From a product standpoint, replacing every apology with diagnostics sounds correct and probably feels terrible. Most users want a quick acknowledgment first, then useful detail if the error matters. The risk is not the phrase itself; it’s using warmth to hide uncertainty or repeated failure.

My default would be: “I’m sorry” for politeness, error code or explanation for debugging, and no claim that the machine feels anything. If the system cannot reliably explain what happened, that is a capability problem—not something an emotional phrase should paper over.