Anthropomorphism raises capability expectations

What it is

A budgeting bot is drawn with eyes and says “let’s look at this month together.” The user then asks whether it can “understand how my marriage affects spending.” The eyes and the “let’s” lift the expectation from “log a transaction” to “get life.” Anthropomorphism raises capability expectations: people demand, by the standard they use for a subject that understands, a tool that in fact only logs.

What is lifted is what it can do, not how reliable it is or how little it forgets. That second class of expectation comes from another set of cues.

Why it happens

Nass and Reeves’s media equation: people involuntarily apply social heuristics to devices that have a face and talk. Among those heuristics: something that can converse probably also has sense. Waytz et al. treat anthropomorphism as mind perception — once it is taken as minded, capability bounds are estimated as a person’s, not as a feature list. The more the interface looks like a person, the further the estimate leaves the manual.

Once the expectation is up, task choice follows: requests that are more “gets people,” and have not been validated, get handed over. Failure is then experienced as lack of ability, not as scope. Scope never covered those requests.

Studying it

Same budgeting engine: one group sees a neutral tool look, another a face plus colloquial speech. Watch which tasks people are willing to submit (log one item / analyse motive / mediate a relationship) and their estimate of “what it understands.” Independent variables: face, colloquial speech, whether a feature list is still shown. Dependent variables: rate of tasks outside coverage, gap between estimated ability and true coverage.

If the feature list is still there and overreach does not fall, anthropomorphic cues have overridden the list. Report that gap.

Where it stops holding

In an explicit role-play product (a game character), raising capability expectation may be the fun itself; the risk has to be rewritten as whether the expectation is carried into a real tool after the scene ends. A professional look can also raise expectation, not only anthropomorphism; anthropomorphism is a strong path. Name, avatar, and tone lifting ability and reliability together is a finer cue bundle, not expanded here.

Applying it

• When anthropomorphism is wanted, pin capability with a feature list and refusal examples: “I can classify spending; I cannot analyse your relationships.” Pin it in the empty state, not after overreach.
• Being able to talk is not an opening of every “gets people” request. On a clear overreach, refuse on scope; do not continue with a more anthropomorphic “I understand you.”
• Check: show a new user the interface for thirty seconds and have them list five things they would hand it. Compare with your coverage. The extras are the capability expectations anthropomorphism has lifted.

Related

• Same group: L5.06.2 Disappointment after unmet expectation is larger than with a neutral voice · L5.06.3 First-person phrasing implies intent and responsibility
• Nearby: L5.12 Expectation Inflation from Anthropomorphism · L5.03 Trust Calibration · L1.02 Expressing Capability Boundaries
• Search terms: anthropomorphism · capability expectation · computers as social actors