The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers

• Authors: Hao-Ping (Hank) Lee, Advait Sarkar, Lev Tankelevitch, Ian Drosos, Sean Rintel, Richard Banks, Nicholas Wilson
• Source: Excerpts from “Critical Thinking Survey”

Summary

This paper investigates the impact of Generative AI (GenAI) tools on critical thinking skills and practices among knowledge workers. Through a survey of 319 participants who shared 936 real-world examples of using GenAI in their work, the study explores when and how critical thinking is enacted and how GenAI affects the effort involved. Key findings indicate that higher confidence in GenAI is associated with less critical thinking, while higher self-confidence in one’s ability to perform the task is linked to more critical thinking. Qualitatively, GenAI shifts the focus of critical thinking towards information verification, response integration, and task stewardship. The research highlights design challenges and opportunities for developing GenAI tools that better support critical thinking in knowledge workflows.

Main Themes and Important Ideas/Facts

1. The Ambiguous Impact of Technology on Cognition:

• The introduction of new technologies has historically raised concerns about their potential negative effects on cognitive abilities. GenAI is the latest in this lineage, following writing, printing, and calculators.
• The “irony of automation” [7] is highlighted: by automating routine tasks, users may lose opportunities to practice their judgment and cognitive skills, leading to atrophy.
• This research focuses specifically on the impact of GenAI on critical thinking, an area that has not been directly explored in prior studies on GenAI’s effects on memory and creativity.
• The study defines GenAI tools as “any ‘end user tool […] whose technical implementation includes a generative model based on deep learning’” [1].

2. Defining and Measuring Critical Thinking:

• The study acknowledges the multiple frameworks for defining critical thinking but adopts Bloom et al.’s Taxonomy of the Cognitive Domain due to its strong research support, wide adoption in education, and relative simplicity for survey instrumentation.
• Bloom’s Taxonomy includes six cognitive activities associated with critical thinking: Knowledge, Comprehension, Application, Analysis, Synthesis, and Evaluation.
• Critical thinking was measured through a one-item five-point scale assessment for each of these six cognitive activities for each GenAI usage example provided by participants.

3. Survey Design and Methodology:

• The survey collected 936 first-hand examples of GenAI use in work tasks from 319 knowledge workers. Participants were asked to describe their tasks, the GenAI tool used, and how they used it.
• Task-Related Factors: The survey considered task type (Creation, Information, Advice, based on Brachman et al. [13]‘s taxonomy) and three aspects of user confidence: confidence in self, confidence in GenAI, and confidence in evaluating AI output.
• User Factors: The survey measured participants’ tendency to reflect on work (using Kember et al.’s Reflective Thinking Inventory) and their trust in GenAI (adapted from the Propensity to Trust Technology scale [56]). Demographic data (gender, age, occupation) was also collected.
• Participants reported their perceived enaction of critical thinking (yes/no with free-text justification) and the perceived change in effort for each of the six cognitive activities of Bloom’s Taxonomy when using GenAI compared to not using it.
• Quantitative analysis involved random-intercepts logistic and linear regression models to account for repeated measures. Qualitative analysis focused on the free-text responses to understand the “when and why” behind the quantitative findings.

4. Key Quantitative Findings:

• Confidence Effects:Higher confidence in GenAI significantly negatively correlated with the perceived enaction of critical thinking (𝛽=-0.69, < 0.001).
• Higher confidence in one’s own ability to do the task without GenAI positively correlated with the perceived enaction of critical thinking (𝛽=0.26, = 0.026).
• Higher confidence in evaluating AI responses also positively correlated with the perceived enaction of critical thinking (𝛽=0.31, = 0.046).
• User Factors: A higher overall tendency to reflect on work had a positive effect on the perceived enaction of critical thinking (𝛽=0.52, < 0.001). Overall trust in GenAI did not show a significant correlation.
• Task Type: No main effect of task type (Creation, Advice, Information) was found on perceived critical thinking.
• Effort: Participants perceived a decrease in effort for most cognitive activities (Knowledge, Comprehension, Application, Analysis, Synthesis) when using GenAI. Evaluation was the only activity where some participants reported increased effort.

5. Key Qualitative Findings: Shifts in Critical Thinking Practices:

The study identified three main phases in knowledge workers’ GenAI tool workflow where critical thinking is enacted:

• Goal and Query Formation:Critical thinking is applied to form goals before using GenAI, analyzing needs and how the tool can help.
• Example: “analyze what my goal was and how I was going to achieve it… I had to first learn what was I going to use in order to make progress.” (P140)
• Critical thinking is used to form queries (prompts) to elicit the desired responses.
• Example: “[I] was reflective when it came to giving the correct prompts, in order to get the correct result a correct description needs to be given.” (P97)
• Inspect Response:Critical thinking involves ensuring quality through objective criteria, such as compliance with the prompt or functional correctness.
• Example: “I had to make sure each piece of text generated met the requirements of the client based on criteria [in the prompt] like colour palette, and people in photos -male/female, skin tone, etc.” (P278)
• Quality is also assessed using subjective standards, including real-world feasibility, internal logic, and relevance.
• Example: “really think about whether the answer the GenAI tool gave me would be easily transferrable to real life situations in social care… not every company has the budget and necessary equipment to provide this most of the times.” (P297)
• A significant aspect of critical thinking becomes verifying information by cross-referencing external sources (reputable websites, official documents, other GenAI tools, human experts).
• Example: “the AI may suggest repertoire [for the concert I direct], but it sometimes is very American-centric. I often have to use my judgment to come up with a repertoire that fits my reality.” (P133)
• Integrate Response:Critical thinking is used to integrate partial responses from GenAI into the user’s work.
• Users modify the style of the GenAI output to be appropriate for the task and to sound less “AI-generated” and more personal.
• Example: “often the AI writes awful stuff like ‘our groundbreaking and fundamental analysis shows…’ that sounds too emphatic and does not fit the scientific style.” (P210)
• Example: “I did make sure it [email composed by ChatGPT] read properly and made sense and did sound like an email that I had composed myself and that a colleague would send.” (P254)

6. Why Critical Thinking Effort Changes with GenAI:

• Decreased Effort:GenAI automates lower-level cognitive activities (e.g., recalling facts, summarizing), leading to perceived less effort for Knowledge, Comprehension, Application, Analysis, and Synthesis.
• GenAI provides personalized feedback loops, reducing the need for external (human) review.
• For tasks where GenAI is perceived as highly competent (e.g., grammar in a non-native language), its outputs are assumed to have fewer errors, decreasing the perceived effort for Evaluation.
• Increased Effort:Evaluation can require more effort when users need to critically assess the accuracy, bias, or limitations of GenAI outputs, especially when cross-referencing with external sources.
• The need to refine prompts and guide GenAI towards the desired outcome can also add effort.

7. Factors Inhibiting Critical Thinking with GenAI:

• Awareness Barriers: Users may not perceive the need for critical thinking when the GenAI tool use is secondary to their goals or when the task is considered trivial or insignificant.
• Motivation Barriers: Trust and over-reliance on GenAI can discourage critical reflection. Users may assume AI is competent, especially for simple tasks, sometimes overestimating its capabilities. Self-doubt in one’s own ability can also lead to accepting AI outputs by default.
• Example: “With straightforward factual information, ChatGPT usually gives good answers.” (P289)
• Example: “It’s a simple task [make a passage professional] and I knew ChatGPT could do it without difficulty, so I just never thought about it, as critical thinking didn’t feel relevant.” (P275)
• The study echoes the “Ironies of Automation” [7] and introduces the “Ironies of Generative AI” [122], highlighting the risk of cognitive skill deterioration if critical thinking is only applied in high-stakes situations without regular practice in common or low-stakes scenarios.
• Perceived flaws in GenAI outputs (e.g., being generic, shallow, inaccurate for niche tasks) can, however, motivate critical revision.
• Example: “the output is way too cookie cutter, full of cliché [text] and boring. I have to edit it a lot to get something out of it that I could ever give to my bosses.” (P92)

Implications and Future Directions

• The findings underscore the importance of fostering a reflective approach to GenAI use to mitigate the risks of over-reliance and cognitive atrophy.
• There is a need for interventions and design strategies that encourage knowledge workers to engage in critical thinking when using GenAI tools.
• Future research should explore longitudinal studies to track the long-term impact of GenAI usage on critical thinking skills.
• Developers of GenAI tools can leverage telemetry and user feedback to understand how to better support critical thinking within their applications.

Quotes

• higher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking.

• GenAI shifts the nature of critical thinking toward information verification, response integration, and task stewardship.

• Used improperly, technolo-gies can and do result in the deterioration of cognitive faculties that ought to be preserved.

• a key irony of automation is that by mechanising routine tasks and leaving exception-handling to the human user, you deprive the user of the routine opportunities to practice their judgement and strengthen their cognitive musculature, leaving them atrophied and unpre-pared when the exceptions do arise.

• knowledge workers’ confidence in AI doing the tasks indeed negatively correlates with their enaction of critical thinking (𝛽=-0.69, < 0.001).

• knowledge workers’ overall tendency to reflect on their work had a positive effect on perceived enaction of critical thinking (𝛽=0.52, < 0.001).

• critical thinking is perceived to be less effort because GenAI tools provide personalised feedback loops for tasks (40/319) that users otherwise do not have access to.

• Complementing our quantitative findings, knowledge workers’ trust and reliance on GenAI (83/319) doing the task can discour-age them from critically reflecting on their use of the tools.

• Without regular practice in common and/or low-stakes scenarios, cognitive abilities can deteriorate over time [5], and thus create risks if high-stakes scenarios are the only opportunities available for exercising such abilities. This phenomenon is well-documented, as in Bainbridge’s ‘Ironies of Automation’ [7], and has been recently revisited in the context of GenAI by Simkute et al. [122] as the ‘Ironies of Generative AI’.