Capability mismatch causes rapid disappointment
Aliases: expectation violation · capability gap · trust disconfirmation
What it is
An expectation–capability mismatch occurs when appearance, demonstration, or language promises a level of competence that the robot cannot reliably deliver. Disappointment is more than an increased failure count: disconfirmation of the user's model can lower judgements of other capabilities, honesty, and dependability.
Why it happens
Early cues create a capability prior, and each success or failure updates it. An inflated prior makes a salient first failure a large prediction error. Users may attribute the discrepancy not only to one weak function but to exaggeration, uncontrollability, or deception. Negative evidence involving safety, privacy, or relational promises is especially resistant to later minor successes, producing a sharper trust collapse than a conservative initial promise.
Studying it
Studies can manipulate expectations through form, description, or demonstration before exposing participants to the same objective success rate. Repeated measures can track capability estimates, calibrated trust, redelegation, and abandonment. Actual performance must be separated from framing, and the severity and diagnosability of failures reported. A post-use satisfaction score alone cannot distinguish disconfirmation from an intrinsically difficult task.
Where it stops holding
Not every failure produces durable disappointment. Accurate prior boundaries, understandable causes, and effective recovery can yield more precise rather than uniformly lower trust. Exploratory and entertainment tasks tolerate occasional failure better than clinical, care, or safety-critical tasks, where one consequential violation may end use. Novelty decay can also reduce enthusiasm but is not a capability mismatch.
Applying it
- Use one capability inventory across marketing, packaging, onboarding demonstrations, and physical styling; remove implications that cannot be reproduced in the deployment environment.
- Show common boundaries and recovery before an ideal path. Ask users to predict performance on edge cases to reveal their model.
- Record pre-failure expectation, subsequent redelegation, and reasons for discontinuation. If one failure depresses trust across unrelated capabilities, repair the promise and explanation rather than merely polishing success feedback.