When you last used AI, did it help you think, or did it think for you?
Generative AI refers to tools capable of producing, on request, a finished piece of reasoning (an essay, a differential diagnosis, a treatment plan) that emulates human thinking. In addition to generative AI, AI plays other roles in health professions education, image analysis, standardized patient simulation, and robotic-assisted training, but those roles are beyond the purview of this article. In what follows, I will focus solely on generative AI (abbreviated simply as AI) and its impact on cognition, particularly critical thinking. I will argue that AI can erode one’s ability to think critically, but when used deliberately, can strengthen it. AI is a disruptive force in higher education, requiring significant changes in our methods of teaching and learning assessment.
Anyone who has used AI to summarize an article, generate content, analyze data, or engage in other prompt-based tasks knows that AI makes mistakes. Anyone who uses AI also knows that its capabilities are remarkable, and that its reliability is on an upward trajectory. Thinking critically has always been important to leaders and learners, and as part of that, asking good questions is essential. The quality of what one generates from AI is proportional to the quality of one’s prompting. Good prompting requires good thinking. Complex projects can take hours, even days, of iterative prompting and conversation guided by critical thinking and reflection. It’s brainwork.
In contrast to iterative prompting, transactional prompting is a tempting shortcut: put in a prompt, whether well-constructed or not, get a result back, and you are done. Transactional prompting works well for simple tasks such as fact checking, copyediting a paper, organizing data, and calculating, but when used for complex problems–the kind that a student might encounter in a college assignment–it can produce incomplete, fragmented, and often erroneous results. Worse than a bad grade, using AI for a quick answer to a complex problem robs us of the learning that comes through critical thinking and reflection.
What is Critical Thinking?
While there are many definitions of critical thinking, the description by nursing educators Scheffer and Rubenfeld is applicable to any field and to a life informed by critical thinking. They describe critical thinking as resting on two coordinate components: habits of mind and cognitive skills. The habits of mind include confidence, contextual perspective, creativity, flexibility, inquisitiveness, intellectual integrity, intuition, open-mindedness, perseverance, and reflection. The cognitive skills include analyzing, applying standards, discriminating, information seeking, logical reasoning, predicting, and transforming knowledge.
I will assume that the importance of critical thinking is self-evident, though I recognize not everyone grants that assumption; I will not argue for it here. Whether we are speaking of habits of mind or cognitive skills, both are learned: sometimes deliberately, sometimes through trial and error, but always through metacognition, thinking about our own thinking. We become critical thinkers by consciously engaging with the world around us, wrestling with problems, testing assumptions, and revising judgments when the evidence no longer supports them. Anything that undercuts that cycle of conscious engagement threatens our capacity to think critically and, with it, our ability to problem-solve, make informed decisions, and act wisely.
Cognitive Offloading
The mechanism behind that erosion has a name in cognitive science: offloading. Risko and Gilbert, writing in Trends in Cognitive Sciences, define cognitive offloading as the use of an external tool to reduce the mental effort a task would otherwise require. Cognitive offloading is not new, nor is it inherently harmful–we offload phone numbers to contacts lists and directions to GPS. In her article in this issue of Noteworthy, Reuben describes how faculty can use AI to offload routine tasks, allowing them to focus on higher-priority activities. Jones explores how we can use agentic AI, AI acting autonomously as “agents,” to save time and accomplish multi-step goals.
In the context of learning, what is particularly interesting about cognitive offloading is the scope of what generative AI enables us to offload, for example, the work of critical thinking, problem-solving, evaluating evidence, and decision-making. Transactional prompting is perhaps the fastest, though not the only way to offload the habits of mind that a graduate needs for a career of lifelong learning. Kosmyna and colleagues at MIT found that students who wrote essays with the help of a large language model showed measurably weaker neural connectivity during the task than students who used a search engine or no tool at all. The study also found that this reduced engagement carried over into later tasks completed without AI. The researchers coined a term for it: cognitive debt, the condition in which habitual reliance on an external system replaces the effortful thinking of independent reasoning and diminishes critical thinking.
Though most research into the effect of AI on cognition comes from undergraduate students–not from students in medical, dental, nursing, or pharmacy programs–health professions educators should pay attention. The students who matriculate into medical schools, dental schools, and other graduate health professions programs will have spent four years as undergraduates in learning environments where the use of AI, both permitted and prohibited, is ubiquitous.
Cheating
Using AI to cheat is a form of detrimental cognitive offloading. In the largest study of AI use among undergraduates, involving over 95,000 students at 20 research-intensive public universities, Chirikov found that 26% of daily AI users reported using it to cheat. Princeton University recently made the news when faculty, with student support, voted to proctor all in-person exams, changing its 133-year-old honor system, which has relied on students to monitor themselves. Inside Higher Ed quotes Kim Lane Scheppele, a professor of sociology and international affairs at Princeton, as saying, “AI was the breaking point—where everyone thought that this introduced stealth cheating that was harder to detect without in-person supervision.” In 2025, The Daily Princetonian reported that nearly 28% of seniors admitted to using ChatGPT on an assignment, despite class policy prohibiting AI use. Cheating is not new, but AI makes it easier to do and harder to detect.
AI is forcing significant changes in teaching, learning, and assessment. If students are not required to exercise critical thinking, many will not. Again, that is nothing new. But now, with AI, a student who wants the grade and nothing more now has a way to get it without doing the reasoning that the assignment was intended to require. An essay or a take-home case write-up was, for decades, a reasonably good way to promote and assess critical thinking. Writing a good one without doing the requisite thinking takes real effort or a willing accomplice. Not only does AI reveal design flaws in assignments, it can also remove the need for skills which the assignments aim to develop.
Assessment in the AI Era
AI is compelling academic institutions to react when they would prefer to be proactive, particularly in assessing learning. Redesigned assessment methodologies are necessary to protect critical thinking from AI misuse. Some assessments remain effective, for example: the oral case presentation, Objective Structured Clinical Examinations (OSCEs), or any real-time exchange in which the learner is required to explain or reason aloud. In these types of learning assessments, AI can help a student prepare. It cannot think critically for students in the room, the clinic, the hospital, in real time, in response to a question. These assessments have always been the strongest way to assess the depth of critical thinking beyond its appearance, and AI has not changed that.
Gerlich, in the journal Data, had participants write essays under three conditions: no AI at all, ChatGPT with no guidance, and ChatGPT used under structured prompts designed to encourage deeper thinking, prompts that led participants to reflect independently, conduct their own research, and construct and revise an argument. He found that unguided use of AI fostered cognitive offloading but did not improve the quality of reasoning. In contrast, structured prompting significantly reduced offloading and enhanced both critical thinking and reflective engagement. What separated the two conditions was not the presence of AI, but when and how AI entered a person’s thinking. Both groups had the same tool in hand, but only one group used reasoned guidance before engaging AI. The order in which a person’s reasoning and AI assistance occurs seems to matter more than whether AI is used at all. Rather than grading solely on the final result, that design principle can help promote and assess critical thinking by requiring students to reason before using AI and then making AI a tool for their continued thinking rather than a substitute for it.
No Simple Answers
Cheating with AI makes for sensational news, but AI is widely used and encouraged in many higher education courses. AI is arguably the best tool to facilitate learning since the printing press. We must accept the reality that AI is here to stay, that it is improving quickly at producing reliable results, and that both students and faculty members will use it. Given the phenomenal power and potential of AI, as well as its dangers, academic institutions and their faculty have a responsibility to guide students in using AI critically, ethically, and strategically.
Amidst the boos when commencement speakers mentioned AI in their addresses to the classes of 2026, Nvidia CEO Jensen Huang conveyed an insightful and balanced message to Carnegie Mellon graduates who worry that AI will take their jobs (Pangambam S). His message applies not only to graduates but also to those who teach the next generation of workers and leaders across all fields. He said, “A new era of science and discovery is beginning. AI will accelerate the expansion of human knowledge and help solve problems once beyond our reach.” Later, he admonished, “AI is not likely to replace you. But someone using AI better than you might.” To rephrase his admonition for academic leaders, educators, and students: the transactional-prompt hack is unlikely to replace you. But someone who thinks critically and uses AI as a tool to advance their knowledge, understanding, and judgment will almost certainly replace you.
Disclosure: Norms for disclosing AI use in academic and professional writing are still forming, and what counts as adequate disclosure will vary by context. What follows is not offered as a model disclosure statement, but simply as an account of how I used AI in writing this article.
The article grew out of reflections on my use of AI, jotted down in handwritten notes over several days. I used AI to transcribe and organize the notes, which served as the initial structure for the article and subsequent research. Most of the works cited were identified through a Google search rather than generative AI. I engaged AI as an iterative sounding board to critique various drafts of the article. I consulted AI to find or to confirm the best word or phrase to express an idea. AI was used to format references from downloaded articles. Finally, AI served as a copyeditor to check spelling and grammar.
References
Igor Chirikov, Ivan Smirnov, and René F. Kizilcec, “Generative AI Use and Misuse Call for Assessment Reform in Higher Education,” Science, Vol. 392 (2026).
The Daily Princetonian, “Senior Survey 2025: Academics.”
Michael Gerlich, “From Offloading to Engagement: An Experimental Study on Structured Prompting and Critical Reasoning with Generative AI,” Data, Vol. 10, No. 11, Article 172 (2025).
Nataliya Kosmyna, Eugene Hauptmann, et al., “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task,” arXiv preprint (2025).
Pangambam S., “Jensen Huang’s 2026 CMU Commencement Speech (Transcript),” The Singju Post, May 12, 2026.
Evan F. Risko and Sam J. Gilbert, “Cognitive Offloading,” Trends in Cognitive Sciences, Vol. 20, No. 9 (2016).
Barbara K. Scheffer and M. Gaie Rubenfeld, “A Consensus Statement on Critical Thinking in Nursing,” Journal of Nursing Education, Vol. 39, No. 8 (2000).
Emma Whitford, “Princeton Introduces Proctoring, Changing Century-Old Honor Code,” Inside Higher Ed, May 15, 2026.
Author:
N. Karl Haden, PhD
President, AAL
