Trust in AI is dynamically updated based on users' expectations
Abstract
Human reliance on artificial intelligence (AI) advice in decision-making varies, with both over- and under-reliance observed. Timing of AI advice has been proposed to address these biases. Additionally, trust in the AI also influences reliance. In a deepfake detection task, we investigated how AI advice affects human performance and decision-making. Using a large online participant pool, we compared task performance when AI advice was provided either concurrently with decisions or after an initial evaluation. We found that while AI advice improved deepfake detection, the timing of advice did not affect performance. Instead, we revealed, using computational modelling, that trust in the AI dynamically influenced participants' willingness to agree with its advice, based on expectations of its ability. These findings highlight the importance of developing appropriate trust in AI-powered decision support systems for effective human-AI collaboration, with applications in designing and testing decision-support tools for perceptual tasks.
Study specs
Researchers conducted an online study with participants performing a deepfake detection task, comparing performance across conditions where AI advice was provided either concurrently with decisions or after an initial evaluation. Computational modeling was used to analyze trust dynamics.
- Discipline
- Human-Computer Interaction
- Study Type
- Experimental Study
- Year
- 2025
- Human Data Platform
- Prolific
- Source
- View Source DOI Google Scholar
Measured Outcomes
Impact of AI advice and its timing on task performance, and the dynamic role of user trust in AI based on expectations of its ability.
Peer Review & Critical Discussion
Potential Selection Bias in 2023 Cohort
The participant pool shows a concerning overrepresentation of users from high-income demographics. Looking at Table 3, we can see that 78% of respondents had annual incomes above $75k, which significantly limits the generalizability of these findings to broader populations.
Non-naive Participants Issue
I've noticed a methodological concern regarding participant naivety. Given that Prolific users often complete multiple studies, there's a real risk that participants had prior exposure to similar experimental paradigms, which could confound the results.
RLHF Applicability to This Study Design
The implications for RLHF training pipelines are understated. If we accept the authors' conclusions about preference stability, this has direct consequences for how we should structure reward model training. The temporal decay effect described in Section 4.2 is particularly relevant.
Verify your expertise to join discussion
Create an account and verify your credentials to participate in peer discussions.