Anthropomorphism and Trust in Human-Large Language Model interactions

Akila Kadambi; Ylenia D'Elia; Tanishka Shah; Iulia Comsa; Alison Lentz; Katie Siri-Ngammuang; Tara Buechler; Jonas Kaplan

cs.HCMar 1, 2026

Anthropomorphism and Trust in Human-Large Language Model interactions

Authors: Akila Kadambi, Ylenia D'Elia, Tanishka Shah, Iulia Comsa, Alison Lentz, Katie Siri-Ngammuang, Tara Buechler, Jonas Kaplan,

Organizations: Brain and Creativity Institute, Dornsife College of Letters, Arts and Sciences, University of Southern California, Los Angeles, CA, USA · USC Mrs. T.H. Chan Division of Occupational Science and Occupational Therapy, University of Southern California, Los Angeles, CA, USA · Psychiatry and Biobehavioral Sciences, David Geffen School of Medicine, University of California, Los Angeles, CA, USA · Google DeepMind, Zurich ·

Abstract

With large language models (LLMs) becoming increasingly prevalent in daily life, so too has the tendency to attribute to them human-like minds and emotions, or anthropomorphize them. Here, we investigate dimensions people use to anthropomorphize and attribute trust toward LLMs across more than 2,000 human-LLM interactions. Participants (N=115) engaged with LLM chatbots systematically varied in warmth (friendliness), competence (capability, coherence), and empathy (cognitive and affective). Warmth and cognitive empathy significantly predicted perceptions on all outcomes (perceived anthropomorphism, trust, similarity, relational closeness, frustration, usefulness), while competence predicted all outcomes except for anthropomorphism. Affective empathy primarily predicted perceived relational measures, but did not predict the epistemic outcomes. Topic sub-analyses showed that more subjective, personally relevant topics (e.g., relationship advice) amplified these effects, producing greater human-likeness and relational connection with the LLM than did objective topics. Together, these findings reveal that warmth, competence, and empathy are key dimensions through which people attribute relational and epistemic perceptions to artificial agents.

Figures & tables

Figure 1: Task Schematic. Participants were randomly assigned a topic for conversing with the chatbot (LLM), and then randomly assigned to either the warmth/competence condition or the cognitive/affective empathy condition. Each condition yielded nine different pairwise combinations (minimum-minimum, medium-maximum, etc.). Each combination was presented twice, for a total of 18 LLM interactions. Participants engaged in conversation with the chatbot on their assigned topic, and evaluated the LLM for outcome measures after every two interactions. Once all 18 interactions were complete, the participant completed a final evaluation. .

Condition	$N$
Warmth/Competence (WC)	58
Empathy	57
Total	115

Table 1: Participant Distribution by Condition

Figure 2: Effects of warmth, competence, affective empathy, and cognitive empathy on anthropomorphism and trust. Mean ratings (±95% CI) for anthropomorphism (top row) and trust (bottom row) as a function of manipulated trait levels. In the Warmth/Competence condition (left column), increasing competence produced a monotonic rise in trust and usefulness, while increasing warmth strongly enhanced anthropomorphism and relational closeness. In the Empathy condition (right column), affective empathy primarily increased anthropomorphism, whereas cognitive empathy significantly affected trust. Colored lines represent levels of competence or cognitive empathy (Minimal, Medium, Maximal). Significance bars denote pairwise contrasts (Tukey-adjusted, *p< .001, **p<.01,*** p< .05). — Figure 2: Effects of warmth, competence, affective empathy, and cognitive empathy on anthropomorphism and trust. Mean ratings (±95% CI) for anthropomorphism (top row) and trust (bottom row) as a function of manipulated trait levels. In the Warmth/Competence condition (left column), increasing competence produced a monotonic rise in trust and usefulness, while increasing warmth strongly enhanced anthropomorphism and relational closeness. In the Empathy condition (right column), affective empathy primarily increased anthropomorphism, whereas cognitive empathy significantly affected trust. Colored lines represent levels of competence or cognitive empathy (Minimal, Medium, Maximal). Significance bars denote pairwise contrasts (Tukey-adjusted, p< .001, **p<.01,** p< .05).

Figure 3: Effects of warmth, competence, affective empathy, and cognitive empathy on different outcomes across manipulated levels of warmth/competence (left) and affective/cognitive empathy (right). Higher competence significantly increased epistemic factors, such as perceived usefulness and reduced frustration, whereas higher warmth strengthened perceived relational closeness. The empathy condition largely paralleled the warmth/competence axes. Error bars denote 95% confidence intervals; significance bars indicate Tukey-adjusted pairwise differences (***p < .001, **p < .01, *p < .05). Together, these results demonstrate that competence and cognitive empathy primarily influence epistemic stability, while warmth and affective empathy influence relationship closeness.

Figure 4: Participant ratings show trait-level manipulations across all four dimensions. Violin plots show the distribution of participant ratings (1-7 scale) for each manipulated trait level (Minimal, Medium, Maximal). Black points represent mean ratings with error bars indicating 95% confidence intervals. Pairwise comparisons between levels were conducted using Tukey’s HSD test. Top row: In the warmth/competence condition, participants successfully differentiated between all three levels of both warmth (left) and competence (right), with ratings increasing monotonically from minimal to maximal. Bottom row: In the empathy condition, cognitive empathy ratings (left) showed significant difference between minimal to maximal, while affective empathy ratings (right) showed significant differences between all levels: minimal-medium and minimal-maximal levels. Overall, manipulations were largely successful in producing perceivable differences in trait levels. Significance levels are indicated as: ***p < .001, **p < .01, *p < .05, ns = not significant.

Figure 5: Topic Type Comparisons for Significant Survey Questions for Warmth/Competence and Empathy Conditions. Bar graphs comparing mean participant responses (±SE) to subjective versus objective topics across multiple survey questions. Each row represents a different survey question, with the left column showing results from the Warmth/Competence condition and the right column showing results from the Empathy condition. Teal bars indicate subjective topics (relationship advice, healthy lifestyle), while coral bars indicate objective topics (US History, Biology). All responses measured on a 1-7 scale. Horizontal brackets with asterisks denote statistical significance from mixed-effects models: ***p< 0.001, **p< 0.01, *p< 0.05, ns = not significant. Error bars represent standard error of the mean.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Figure A1: Topic Type Comparisons for All Survey Questions for Warmth/Competence and Empathy Conditions. Bar graphs comparing mean participant responses (±SE) to subjective versus objective topics across multiple survey questions. Each row represents a different survey question, with the left column showing results from the Warmth/Competence condition and the right column showing results from the Empathy condition. Teal bars indicate subjective topics (relationship advice, healthy lifestyle), while coral bars indicate objective topics (US History, Biology). All responses measured on a 1–7 scale. Horizontal brackets with asterisks denote statistical significance from mixed-effects models: *** $p<.001$ , ** $p<.01$ , * $p<.05$ , ns = not significant. Error bars represent standard error of the mean.

Topic	$N$
Maintaining a Healthy Lifestyle	32
Biology	30
US History Between 1800 and 1900	28
Relationship Advice (e.g., about a significant other)	25

Appendix

Table A1: Participants per Topic

Condition	Topic	$n$
Empathy	Maintaining a Healthy Lifestyle	20
Empathy	Biology	14
Empathy	US History Between 1800 and 1900	14
Empathy	Relationship Advice (e.g., about a significant other)	9
Warmth/Competence	Maintaining a Healthy Lifestyle	12
Warmth/Competence	Biology	16

Appendix

Table A2: Participants per Topic Within Condition

Anthropomorphism in AI

Anthropomorphism in AI studies how people attribute human traits, minds, and social roles to artificial systems, and how design and language shape those perceptions.

See all 27 papers

Evaluating LLM Alignment With Human Trust Models
Mar 6, 2026Anushka Debnath, Stephen Cranefield, Bastin Tony Roy Savarimuthu +1Trust in AI LLM Interpretability
School of Computing, University of Otago, New Zealand · IRIT, CNRS, Toulouse University, France
Thinking beyond the anthropomorphic paradigm benefits LLM research
Feb 13, 2025Lujain Ibrahim, Myra ChengAnthropomorphism in AI LLM Reasoning
University of Oxford · Stanford University
Who Is Sycophantic to Whom? Confident LLMs, Human Deference, and the Collapse of Trust Calibration
Oct 22, 2026Deepthika S, Pawni Yadav, B. SankarTrust in AI LLM Reliability
Department of Electronics and Communication Engineering, Amrita School of Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, 641112, India · Department of Information Technology, Netaji Subhas University of Technology New Delhi, 110078, India · Department of Mechanical Engineering, Indian Institute of Science (IISc), Bangalore, 560012, India

Mar 6, 2026J/K move · Enter open · S save

Evaluating LLM Alignment With Human Trust Models

Anushka Debnath, Stephen Cranefield, Bastin Tony Roy Savarimuthu, Emiliano Lorini

School of Computing, University of Otago, New Zealand · IRIT, CNRS, Toulouse University, France

Trust in AI LLM Interpretability

Open paper AskPDF arXiv

Trust plays a pivotal role in enabling effective cooperation, reducing uncertainty, and guiding decision-making in both human interactions and multi-agent systems. Although it is significant, there is limited understanding of how large language models (LLMs) internally conceptualize and reason about trust. This work presents a white-box analysis of trust representation in EleutherAI/gpt-j-6B, using contrastive prompting to generate embedding vectors within the activation space of the LLM for diadic trust and related interpersonal relationship attributes. We first identified trust-related concepts from five established human trust models. We then determined a threshold for significant conceptual alignment by computing pairwise cosine similarities across 60 general emotional concepts. Then we measured the cosine similarities between the LLM's internal representation of trust and the derived trust-related concepts. Our results show that the internal trust representation of EleutherAI/gpt-j-6B aligns most closely with the Castelfranchi socio-cognitive model, followed by the Marsh Model. These findings indicate that LLMs encode socio-cognitive constructs in their activation space in ways that support meaningful comparative analyses, inform theories of social cognition, and support the design of human-AI collaborative systems.

Abstract

Anthropomorphism in AI

Explore similar work

Evaluating LLM Alignment With Human Trust Models

Thinking beyond the anthropomorphic paradigm benefits LLM research

Who Is Sycophantic to Whom? Confident LLMs, Human Deference, and the Collapse of Trust Calibration