Passing the Buck to AI: How Individuals' Decision-Making Patterns Affect Reliance on AI

Katelyn Xiaoying Mei, Rock Yuren Pang, Alex Lyford, Lucy Lu Wang, Katharina Reinecke

TL;DR

This paper investigates how individual psychological traits—specifically decision-making "coping patterns"—influence how we interact with AI. The researchers found that "buckpassers" (people who tend to defer decisions to others) are more likely to seek AI help and report higher reliance on it, yet they spend significantly less time scrutinizing AI explanations, making them uniquely vulnerable to AI misinformation.

What Problem Does This Paper Address?

As conversational AI tools like ChatGPT become go-to resources for everything from medical advice to financial planning, a critical question emerges: Why do some people follow AI blindly while others remain skeptical?

Most existing research focuses on the AI system itself—how its accuracy or transparency affects user trust. However, this paper shifts the lens toward the user. The authors address a gap in understanding how internal psychological patterns—specifically how we handle the stress of making a choice—dictate our "reliance" on AI. They specifically look at whether AI is being used as a tool for informed decision-making or simply as a way to escape the burden of choosing.

Methods

The researchers conducted an online experiment with 810 participants via the LabintheWild platform.

  1. Psychological Profiling: Participants first completed the Melbourne Decision-Making Questionnaire (MDMQ), a validated psychological tool that measures four coping patterns:
  2. Vigilance: Thorough information gathering and evaluation.
  3. Buckpassing: Deferring decisions to others to avoid responsibility.
  4. Hypervigilance: Rushed, anxious decision-making.
  5. Procrastination: Putting off the decision (not focused on in this study's hypotheses).
  6. The Task: Participants evaluated 10 nutrition statements (e.g., "The digestion process begins in the mouth") to determine if they were facts or myths.
  7. The AI Intervention: Crucially, AI help was optional. Participants could choose to see the "AI Decision" and then further choose to see an "AI Explanation."
  8. Analysis: The team used mixed-effects logistic and linear regression models to correlate MDMQ scores with "clicking" behavior (seeking help), "reading" behavior (time spent on explanations), and self-reported reliance.

Datasets

  • Nutrition Statements: A set of 30 popular food facts and myths curated from prior research (Florença et al., 2021).
  • AI Responses: The researchers used ChatGPT (GPT-4) to generate decisions and explanations. Notably, they intentionally included a mix of accurate and inaccurate AI suggestions to observe how users reacted to "hallucinations."
  • New Contribution: The authors have made their experimental dataset—including participant demographics, MDMQ scores, and interaction logs—publicly available on GitHub.

Key Insights and Ideas

  • The Buckpassing Paradox: People with high "buckpassing" tendencies are the most likely to seek out AI suggestions and report that they "rely" on the AI. However, they spend the least amount of time reading the AI's explanations. For them, the AI isn't a teacher; it’s a "get out of jail free" card for decision-making stress.
  • Vigilance as a Filter: Conversely, "vigilant" individuals don't necessarily seek AI help more often, but when they do, they spend significantly more time scrutinizing the explanations. They use the AI as a data point to be cross-referenced, not a final authority.
  • Vulnerability to Misinformation: Because buckpassers seek the answer but skip the reasoning, they are significantly more susceptible to AI hallucinations. The study showed that when the AI was wrong, participants' accuracy dropped significantly if they chose to see the AI's decision.
  • Demographic Trends: The study found that younger participants (under 18) and those with lower education levels tended to score higher in buckpassing, suggesting these groups might need more robust "cognitive forcing functions" (design features that encourage critical thinking) when using AI.

Key Results

  • Seeking Help: A higher buckpassing score was a significant predictor of the likelihood to seek an AI decision ($p < 0.05$).
  • Scrutiny Time: For every one-point increase in a user's vigilance score, they spent an average of 0.53 seconds more reading AI explanations. Conversely, a one-point increase in buckpassing led to 0.4 seconds less time spent reading ($p < 0.05$).
  • Self-Reported Reliance: High buckpassers were 9% more likely to report high reliance on the AI for each point increase in their score.
  • Agreement Rate: Participants agreed with the AI's decision approximately 83% of the time when they chose to see it, even though the AI was intentionally wrong in 4 out of 10 cases.

Limitations and Future Work

  • Low-Stakes Domain: The study focused on nutrition. The authors acknowledge that decision-making patterns might shift in high-stakes environments like legal or medical fields where the consequences of "passing the buck" are more severe.
  • Static Interface: The experiment used a simplified, button-based interface rather than a dynamic, back-and-forth chat interface, which might change how users process information.
  • Cultural Context: While the sample was global, it was primarily English-speaking and Western-leaning. Prior research suggests buckpassing tendencies vary significantly across cultures (e.g., East Asian vs. Western).
  • Future Direction: The authors suggest designing "adaptive" AI interfaces that detect a user's decision-making style and provide more (or different) "cognitive nudges" to those prone to blind reliance.

Summary updated

BibTeX
@article{mei2026passing,
  author = {Katelyn Xiaoying Mei and Rock Yuren Pang and Alex Lyford and Lucy Lu Wang and Katharina Reinecke},
  title = {{Passing the Buck to AI: How Individuals' Decision-Making Patterns Affect Reliance on AI}},
  journal = {ACM Trans. Comput. Hum. Interact.},
  year = {2026},
  doi = {10.1145/3786326},
}
Figure