← CHI 2026 map

CHI '26 · Honorable mention · full-paper review · confidence medium-high

Trust Formation in AI Delegation: The Interplay of Explainability and Anthropomorphism

Chenyang Li , Zhixuan Deng , Hao Ling , Xu Zhang

DOI PDF Program page Page JSON

This is a strong CHI paper because it turns a familiar design assumption into a testable interaction claim and backs it with two complementary studies. The main takeaway is practical and non-obvious: explainability does not automatically make anthropomorphic agents more trustworthy, and in some settings it can do the opposite.

Axes Lens

Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.

Review Summary

This paper is compelling because it addresses a real design belief that is often repeated but rarely tested carefully: that explainability and anthropomorphism should both help trust, and therefore should work even better together. The authors show that this intuition is too simple. In the online experiment, anthropomorphism actually reduced trust in an explainable agent, which is exactly the kind of counterintuitive result that can shift how CHI researchers and practitioners think about interface cues. The second study is especially valuable because it does not stop at a behavioral effect; it adds a preregistered lab experiment with eye-tracking and finds the opposite pattern under controlled conditions, where the combined design elicited the highest trust. That makes the paper more than a one-off anomaly: it argues that context and cognitive engagement shape whether social cues help or hurt. The mechanism claim is plausible and well aligned with the evidence quoted in the paper, but it should still be read as a bounded explanation rather than a universal law. The strongest contribution is therefore a design and theory insight: pairing explanation with human-like cues is not automatically additive, and designers may need to secure cognitive engagement before social cues can be beneficial. The main limitations are also clear from the paper’s own framing: the context manipulation is tied to online versus lab settings, Study 2 is smaller, and the task/modality choices are specific. Even so, the paper’s combination of scale, preregistration, and mechanism-oriented measurement makes it a solid honorable-mention-level contribution with good relevance to CHI’s broader trust and AI delegation agenda.

What Changed

Canon before

Prior CHI work often treats explainability and anthropomorphism as independently trust-enhancing cues, but the interaction between them is not well established.

Departure from common sense

The paper reports a counterintuitive result: adding anthropomorphism to an explainable agent can reduce trust rather than increase it, at least in the online condition. That directly departs from the common design intuition that more human-like cues plus more explanation should be jointly beneficial.

Actual novelty

The paper’s main novelty is not a new interface element but a tested interaction claim: it systematically examines how explainability and anthropomorphism combine, and it pairs a large online experiment with a preregistered eye-tracking lab study to explain why the effect changes across contexts.

Evidence

The evidence supports a two-part contribution: first, an online experiment with N=900 found an interference effect where anthropomorphism reduced trust in an explainable agent; second, a preregistered lab study with N=57 and eye-tracking found the combined design elicited the highest trust under controlled conditions. The paper also links the effect to cognitive engagement and attention to social cues.

Limits

Method limits

The studies rely on specific operationalizations of explainability, anthropomorphism, trust, and cognitive engagement. Study 2 is smaller and uses eye-tracking in a controlled lab setting, so the mechanism is suggestive rather than definitive across all AI delegation contexts.

Deployment limits

The findings are most directly relevant to interface design for AI delegation agents using static avatars and text explanations in consumer-like tasks. They may not transfer unchanged to other modalities, domains, or longer-term trust formation.

Boundary conditions

The paper itself indicates that context matters: the online setting produced interference, while the lab setting produced synergy. The effect appears contingent on engagement level, task framing, sample composition, and the specific cue combination used.

Manual check: Check the exact wording of the novelty claim against the introduction and related-work gap statement if a stricter canonization is needed.

Position in field

This sits at the intersection of XAI, anthropomorphism, and trust in AI delegation. Its value is in showing that these cues should not be assumed additive; instead, their interaction depends on cognitive engagement and setting.

Abstract

As AI agents act on behalf of users, designers increasingly combine explainability (XAI) and anthropomorphism to build trust. Yet, whether these cues create synergy or interference remains a critical, open question. Our online experiment (N=900) revealed a counterintuitive interference effect: anthropomorphism reduced trust in an explainable agent. A preregistered lab study with eye-tracking (N=57) reversed this finding: under controlled conditions, the combined design elicited the highest trust. Eye-tracking reveals the mechanism: XAI promotes deeper cognitive engagement (e.g., longer fixations), which primes users to allocate attention to social cues (e.g., avatars). Our findings show that trust depends on cognitive engagement moderating social cue processing, yielding a critical design insight: effectively pairing explanatory and anthropomorphic interfaces requires first securing the user's cognitive engagement to avoid undermining trust.