PaperlayerView exact revision on arXiv

Source-linked AI summary

Trust in AI and Its Role in the Acceptance of AI Technologies

Hyesun Choung, Prabu David, Arun Ross

arXiv:2203.12687v1cs.AIcs.HC

TL;DR

The paper examines how trust contributes to acceptance of AI technologies and addresses the need for a more sophisticated, multidimensional understanding of trust. Across two survey studies, it extends TAM analyses to AI voice assistants and broader consumer AI, finding that trust operates through perceived usefulness and attitudes, while distinguishing human-like and functionality trust.

  • Problem

    Research needed clearer, more consistent ways to assess social and psychological trust in AI and to explain its role in technology acceptance.

  • Method

    Two survey studies used TAM-based path analyses, with Study 2 applying exploratory factor analysis to trust measures in a representative U.S. sample.

  • Results

    Trust significantly contributed to AI acceptance indirectly through perceived usefulness and attitudes, with effects replicated across studies and trust separated into human-like and functionality dimensions.

  • Takeaways & Limitations

    Trust is a significant component of AI technology acceptance, and the paper provides a multidimensional trust measure for future trustworthy-AI research.

  • Takeaways & Limitations

    Because the path models are correlational, they cannot establish causal effects of trust on AI technology acceptance.

Abstract

from arXiv · show

As AI-enhanced technologies become common in a variety of domains, there is an increasing need to define and examine the trust that users have in such technologies. Given the progress in the development of AI, a correspondingly sophisticated understanding of trust in the technology is required. This paper addresses this need by explaining the role of trust on the intention to use AI technologies. Study 1 examined the role of trust in the use of AI voice assistants based on survey responses from college students. A path analysis confirmed that trust had a significant effect on the intention to use AI, which operated through perceived usefulness and participants' attitude toward voice assistants. In study 2, using data from a representative sample of the U.S. population, different dimensions of trust were examined using exploratory factor analysis, which yielded two dimensions: human-like trust and functionality trust. The results of the path analyses from Study 1 were replicated in Study 2, confirming the indirect effect of trust and the effects of perceived usefulness, ease of use, and attitude on intention to use. Further, both dimensions of trust shared a similar pattern of effects within the model, with functionality-related trust exhibiting a greater total impact on usage intention than human-like trust. Overall, the role of trust in the acceptance of AI technologies was significant across both studies. This research contributes to the advancement and application of the TAM in AI-related applications and offers a multidimensional measure of trust that can be utilized in the future study of trustworthy AI.

Trust and Use of AI Technologies: An Extended TAM

The paper extends the TAM by incorporating trust to explain AI technology acceptance, distinguishing human-like and functionality trust as potentially different dimensions.

  • The TAM explains technology adoption through perceived usefulness and perceived ease of use.
  • The study proposes that trust influences perceived usefulness and attitudes, with these factors mediating trust’s relationship with behavioral intention.
  • The study addresses limited and variable measurement of social and psychological trust in AI by examining multiple trust dimensions within TAM.
  • AI trust may differ from conventional technology trust because AI has autonomy and can replace human tasks and decisions.
  • The paper conceptualizes human-like trust as concern for AI’s social values and ethics, and functionality trust as concern for technological competence and expertise.

Participants and Procedure

The studies used online surveys and TAM-based measures, with path analysis testing hypothesized relationships among trust, usability, usefulness, attitudes, and usage intention.

  • Study 1: Study 1 collected responses from 312 undergraduate students at a large Midwest university, 96% of whom had used an AI voice assistant.
  • Study 1: Study 1 measured perceived ease of use, perceived usefulness, attitude, trust, and behavioral intention using adapted or newly created survey items.
  • Study 1: Study 1 tested hypothesized relationships with path analysis and evaluated indirect effects using bootstrapped 95% confidence intervals.

[Table 2 Near Here]

The path model supported TAM relationships and showed that trust contributed to usage intention indirectly through perceived usefulness and attitudes.

  • All four proposed predictors—perceived ease of use, trust, perceived usefulness, and attitude—were included in the voice-assistant usage-intention model.
  • A good-fitting path model linked perceived ease of use, trust, perceived usefulness, attitude, and usage intention.
  • Perceived ease of use positively predicted perceived usefulness, trust perception, attitude, and usage intention, supporting H2.
  • Attitude positively predicted usage intention (β = .62, p < .001), supporting H1.
  • Trust increased perceived usefulness, which increased attitudes and usage intention through significant mediated paths.
  • Perceived ease of use had the largest total effect on usage intention (β = .63, p < .001).

Discussion

The discussion emphasizes trust’s indirect role in AI acceptance, the distinction between human-like and functionality trust, and practical implications for usable, useful, and trustworthy AI.

  • Study 1: Study 1 empirically tested the TAM for acceptance of AI voice assistants.
  • Study 1: The validated model explained 52% of the variance.
  • Study 1: Perceived usefulness had a greater direct effect on usage intention than perceived ease of use, while ease of use had the larger total effect.
  • Practical implications: The findings support designing AI that is trustworthy, useful, and easy to use.
  • Study 1: Ease of use contributed to trust, while trust predicted positive attitudes and perceived usefulness associated with greater usage intention.
  • Study 2: Study 2 replicated Study 1 findings with a representative national sample and treated trust as multidimensional.
  • Study 2: The two trust dimensions were human-like trust and functionality trust, reflecting social-ethical values and technological competence, respectively.

[Table 5 near here]

Study 2 tested two trust dimensions in path models of smart-technology acceptance. Both dimensions related to perceived usefulness, attitude, and usage intention, while functionality-related trust had the greater total impact.

  • Two separate path models examined human-like trust and functionality-related trust in AI smart technologies.
  • Attitude strongly predicted intention to use smart technologies (β = .65, p < .001) in both models.
  • Perceived ease of use predicted perceived usefulness, both trust dimensions, attitude, and usage intention in the proposed models.
  • Perceived usefulness predicted attitude (β = .54, p < .001; β = .53, p < .001) and usage intention (β = .22, p < .001) in both models.
  • Both trust dimensions had significant indirect effects on usage intention through perceived usefulness and positive attitudes.
  • Functionality-related trust had a greater total impact on usage intention than human-like trust (β = .42, p < .001 versus β = .36, p < .001).

Discussion

The discussion argues that trust is central to AI acceptance within TAM and should be understood through both functional and human-like dimensions. It also emphasizes ease of use and users’ direct experience as trust-building factors.

  • Study 2 supported TAM’s applicability to consumer smart technologies and replicated the significant predictors identified in Study 1.
  • Both human-like and functionality trust predicted perceived usefulness and positive attitude, which in turn predicted greater usage intention.
  • Functionality-related trust had a greater total impact than human-like trust, although both dimensions contributed to understanding trust in AI.
  • The findings characterize AI as both a functional and social technology, requiring both dimensions of trust in trustworthy-AI design.
  • Perceived ease of use and usefulness were essential determinants of AI acceptance, with ease of use consistently having the greater impact.
  • Users’ direct experience helps shape judgments about usefulness and ease of use, so trustworthy AI should integrate experience beyond system design.
  • The authors recommend including trust as an integral component of predictive models of AI acceptance.

Conclusion

The paper frames trust as increasingly important for understanding AI acceptance as AI development and everyday use expand. It examines trust dimensions within the established TAM framework.

  • AI’s growing influence and black-box mechanisms increase the need for a sophisticated understanding of trust in AI.
  • The study examined different trust dimensions within the Technology Acceptance Model, a framework widely used to study technology acceptance.
  • The paper identifies its key contributions as extending TAM with trust and examining trust across AI technology applications and populations.

Theoretical Contributions

The paper extends TAM by treating trust as multidimensional and develops concepts intended to support research and practice on trustworthy AI. It also identifies boundaries for future work.

  • The core theoretical contribution extends classical TAM by incorporating trust-specific aspects into AI acceptance models.
  • A multidimensional trust approach captures variability in AI trust perceptions and can guide research linking ethical principles to trust dimensions.
  • The framework can inform value-oriented, ethical, and human-centered AI design, governance, and procedures for encouraging appropriate trust.
  • The findings suggest AI technologies should be easy to use, useful, and trusted, with trustworthy values treated as a critical design concern.
  • The study focuses on lower-risk consumer products, leaving high-stakes applications such as self-driving cars and healthcare for future research.
  • Trust is dynamic across technology stages, and initial and ongoing trust may play different roles.
  • The path models are correlational, so they cannot guarantee causal effects of trust on AI acceptance.
  • Online-panel respondents may not represent the general population because computer and internet access may skew the sample toward higher technology acceptance.

Basis of trust in technology

The trust measures distinguish human-oriented and functionality-oriented bases of trust in AI, alongside technology-acceptance constructs. These constructs show positive associations, and the Study 1 path model reports significant relationships among them.

  • Study 1 path model: In Study 1, perceived ease of use significantly predicted usefulness, trust, attitude, and behavior intention.The standardized paths were β = .36, .39, .26, and .19, respectively, with p-values from <.001 to .012.
  • Study 1 path model: In Study 1, trust significantly predicted usefulness and attitude, while attitude strongly predicted behavior intention.The reported standardized paths were trust → usefulness β = .39, trust → attitude β = .23, and attitude → behavior intention β = .62.
  • Trust dimensions: Trust in AI is organized around human-oriented and functionality-oriented dimensions.The functionality dimension includes competence, reliability, dependability, and task-related features, while human-oriented items include benevolence and integrity.
  • Measurement: Study 2 used principal components exploratory factor analysis to examine the trust structure.The reported factor analysis used a sample of N = 640 and an oblique promax rotation.
  • Associations: Study 2 correlations were positive across ease of use, both trust dimensions, usefulness, attitude, and behavior intention.Behavior intention correlated .57 with human trust and .65 with technology trust; attitude correlated .68 and .69 with these dimensions, respectively.

Standardized Regression Path Analysis with Human-like Trust in AI (Study 2)

The Study 2 path analysis evaluates how technology trust relates to usefulness, attitude, and behavior intention. All reported paths are statistically significant, with attitude showing the strongest direct association with behavior intention.

  • Direct paths: Attitude significantly predicted behavior intention with β = .65, the strongest reported direct path.The path was reported with p < .001.
  • Direct paths: Technology trust significantly predicted perceived usefulness, attitude, and behavior intention with β = .38, .31, and .14, respectively.Each reported path had p < .001.
  • Direct paths: Perceived usefulness significantly predicted attitude and behavior intention with β = .53 and .22, respectively.Both paths had p < .001.
  • Indirect effects: Technology trust showed indirect effects through usefulness and attitude, including a total sequential path effect of .07 through usefulness and attitude to behavior intention.The reported 95% confidence interval for this sequential effect was .05 to .10.
  • Indirect effects: The indirect effect of technology trust through attitude to behavior intention was .20, with a 95% confidence interval of .13 to .29.The indirect effect through usefulness to behavior intention was .08, with a 95% confidence interval of .04 to .14.

Appendix

The appendix reports the survey constructs, item wording, response scales, and reliability information used across the two studies. It distinguishes human-like and functionality trust in smart technologies and adapts technology descriptions by study.

  • Measures: Survey items used five-point agreement scales ranging from 1 (strongly disagree) to 5 (strongly agree).
  • Measures: Perceived usefulness items assessed whether AI technologies improve performance, productivity, effectiveness, task completion, and speed.
  • Measures: Attitude items assessed whether using AI technologies feels positive, pleasant, smart, or a good idea.
  • Measures: Behavioral-intention items measured intentions to use or continue using AI virtual assistants or smart technologies in the future.
  • Trust measures: Study 2 measured human-like trust through benevolence and integrity items, and functionality trust through competence, reliability, and dependability items.
  • Study adaptations: The same item wordings were used across studies for several constructs, while Study 1 referred to AI virtual assistants and Study 2 to AI smart technologies.
Loading 2203.12687v1…