# How Interacting with AI Affects Confidence in Your Own Thinking: APA Study Findings
A study published in the journal APA Technology, Mind, and Behavior showed that using large language models doesn't make people dumber, but it can undermine confidence in their own reasoning—if you approach AI passively. The key factor isn't the use of AI itself, but the degree of active user involvement in generating and editing responses.
Passivity as the Main Risk
The experiment involved 1,923 workers from the USA and Canada. They were asked to complete ten simulated professional tasks using ChatGPT, Claude, and Gemini. The tasks covered typical office scenarios: creating plans with incomplete information, interpreting ambiguous data, arguing strategic decisions, and building multi-step sequences.
58% of participants agreed with the statement: "AI did most of the intellectual work for me." However, researchers found that reduced confidence in their own conclusions was directly linked not to using AI per se, but to how passively people accepted its suggestions. Those who took the model's responses almost unchanged more often reported a loss of sense of ownership and reduced trust in their own judgment.
In contrast, participants who edited, challenged, or rejected parts of the AI's responses maintained higher confidence levels and felt the result truly belonged to them. As neuroscientist Sarah Baldeo from Middlesex University notes: "The problem isn't using AI itself, but the degree of passive acceptance."
Task Type Shapes Behavior
Interestingly, the tendency toward passive delegation depends on the nature of the task. In open-ended, multi-step assignments—which make up a significant portion of professional work—participants more often handed full initiative to AI. However, in personal or introspective tasks (like self-assessing character or reflecting on personal experience), they were much more likely to critically evaluate and challenge the model's suggestions.
Additionally, professional experience played a protective role. Senior specialists argued with AI more often than juniors and reported higher confidence in the final result. This suggests that expert knowledge helps maintain critical distance even with active AI use.
Recommendations for Developers and Users
Based on the findings, Baldeo recommends that AI system developers implement mechanisms to encourage active engagement:
- Automatically suggest multiple alternative solutions.
- Prompt checks on key assumptions underlying the response.
- Explain the model's reasoning logic step by step.
- Provide hints encouraging users to expand or rework the output.
Similar approaches are already being tested in academia. For example, at the CHI 2026 conference, Mina Lee's team from the University of Chicago presented a study with 393 participants. Those who first partially worked on the task themselves and only then turned to the chatbot showed better critical thinking outcomes than those who started with AI right away. Though under tight deadlines, early AI use still offered a speed advantage.
Both studies agree on one point: AI enhances cognitive abilities only when people remain in the role of editor, not passive consumer.
Key Takeaways
- Using AI itself doesn't reduce intelligence, but it can undermine confidence in your own reasoning.
- The main risk factor is passive acceptance of responses without critical evaluation.
- Actively editing, challenging, and selectively adopting AI suggestions preserves a sense of ownership and confidence.
- Experienced professionals are less prone to "cognitive offloading" effects thanks to honed expert intuition.
- AI interface design should promote critical interaction, not simplify decisions to a single click.
— Editorial Team
No comments yet.