The Critical Thinking Premium

Edition 47 | Institutional Effectiveness Weekly

Two claims about artificial intelligence and human reasoning are circulating in the same news cycle, and they point in opposite directions. Employers are chasing critical thinking harder than at any point in recent memory. Gartner, the technology research and advisory firm that surveys IT and business leaders on emerging workforce trends, projects that concern over critical-thinking atrophy from generative AI use will push half of large organizations toward requiring “AI-free” skills assessments by the end of 2026 (Gartner, 2025). A recent survey of more than 500 C-suite executives found that three in four would choose a candidate strong in judgment and communication over one who has AI technical skills alone (Hyken, 2026). At the same time, new research on how people reason while using AI tools suggests that the skill employers want most may be eroding fastest in the populations that use AI the most.

That is the tension this edition examines. It is not a question of whether AI helps or harms critical thinking in general; the evidence supports both outcomes, depending on how the tool is used and what the task demands. It is a question of what happens when rising external demand for a skill meets a technology that can, under identifiable conditions, reduce the practice that builds it. Developing critical thinking has long been part of the educational mission that colleges and universities state for themselves, and institutional effectiveness offices exist in large part to produce evidence that the mission is being met. This turns a longstanding commitment into a live measurement question with new urgency, one that connects directly to the demonstration gap this newsletter has tracked since Edition 24.

The Mechanism Behind the Erosion Risk

The clearest evidence comes from two independent research efforts published within the past eighteen months. A 2025 study from Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers about 936 real instances of generative AI use at work. The finding that matters most for higher education: higher confidence in the AI tool was associated with less enacted critical thinking, while higher confidence in one’s own ability was associated with more critical thinking, though at greater perceived cognitive effort (Lee et al., 2025). Participants described a shift in the nature of their thinking work: from generating answers to verifying them, and from solving problems to integrating AI-produced responses. That shift is not inherently a loss. Verification is itself a critical-thinking skill. But it is a different skill than the one most assessment rubrics were designed to measure, and it disappears entirely when a user stops verifying and simply accepts the output.

A January 2026 Wharton School working paper gives that failure mode a name. Steven Shaw and Gideon Nave ran three experiments with more than 1,300 participants and nearly 10,000 individual trials, testing what happens when people have optional access to an AI tool while completing reasoning tasks. They call the pattern cognitive surrender: the adoption of an AI-generated answer with minimal scrutiny, in a way that overrides both intuition and deliberate reasoning rather than simply supplementing them (Shaw & Nave, 2026). Confidence rose by a similar margin whether the AI-supplied answer was correct or planted wrong, meaning participants felt no less certain when the tool had led them astray. Shaw and Nave frame this as evidence of a third mode of cognition, layered onto the fast-intuitive and slow-deliberate systems that behavioral science has documented for decades, one that can substitute for reasoning rather than assist it.

Neither study claims that AI use inevitably degrades critical thinking. Both identify the specific condition under which it does: routine, low-stakes tasks, completed under time pressure, by users who trust the tool more than they trust their own judgment. That is a description of a great deal of undergraduate coursework as currently assigned.

The Demand Side Has Never Been Louder

Set against that erosion risk is a labor market that is demanding the same skill more insistently than at any point in the last decade. A 2023 paper in the Journal of Intelligence argued that critical thinking functions as a job-proof skill precisely because it resists the kind of automation reshaping programming, legal research, and radiology, and that its value rises rather than falls as AI absorbs the routine cognitive tasks around it (Dumitru & Halpern, 2023). That argument, made before the current wave of generative AI adoption, reads as a forecast that has since been confirmed. The Gartner prediction cited above is not a warning that human judgment is disappearing. It is a warning that judgment has become difficult to detect inside AI-assisted work, since AI can now produce judgment-like output on request. The response employers are adopting is not to assume judgment is scarce; it is to test for it directly, under conditions where AI is not available to supply the answer.

The High Point University survey referenced above found that 90% of executives rate life skills, including critical thinking, communication, and adaptability, as the top predictor of professional success, ahead of technical AI proficiency (Hyken, 2026). Most executives in that survey do not treat this as a choice between AI skills and human judgment; 87% say the ideal candidate combines both. But the ranking is telling. When forced to choose, employers are choosing the skill that AI cannot supply on its own.

A New Dimension of the Demonstration Gap

Edition 24 of this newsletter examined why the liberal arts case for higher education, that broad-based education develops critical thinking, communication, and adaptability, is correct on the evidence and still insufficient on its own (Rudawsky, 2026b). The problem it identified was visibility: institutions develop these competencies but cannot make that development legible to employers, who have rationally stopped trusting the credential to certify it. What this edition adds is a second layer sitting on top of the first. It is no longer only a question of whether critical thinking, once developed, is visible to employers. It is also a question of whether the conditions under which students now complete coursework, with AI assistance available for nearly every assignment, still reliably produce the skill in the first place.

This is not a claim that institutions or faculty are doing something wrong by permitting AI use. Assignment design has always had to adapt to available tools, from calculators to spell-check to search engines, and it will adapt to this one. The point is narrower and more actionable: the research above describes specific, identifiable conditions under which critical-thinking practice is reduced, and those conditions (routine tasks, low stakes, time pressure, high trust in the tool) are common enough in current coursework that institutions should not assume the skill is developing simply because it always has. That assumption, reasonable a decade ago, is now a testable one.

The practical response starts in the classroom rather than the assessment office. Assignments that make reasoning visible, a short written account of where a student consulted AI and where they diverged from it, an annotated draft history showing how an analysis developed, or a brief oral defense of a submitted argument, give faculty a direct look at whether independent judgment occurred, not just whether the final product reads well. None of these require abandoning AI-integrated instruction. They require treating the reasoning process, and not only the finished artifact, as something worth assigning and grading. IE offices are positioned to help translate that classroom-level evidence into the kind of documented, program-level signal the demonstration gap requires, a role the next section develops further.

The Conformity Connection

Edition 22 of this newsletter examined Jonathan Mastroianni’s research on the long-term rise of conformity across American institutions and what that trend means for intellectual risk-taking on campus (Rudawsky, 2026a). Cognitive surrender is a version of the same pattern at the level of an individual task: adopting the fluent, confident, immediately available answer rather than sitting with uncertainty long enough to reason through it. Productive deviance, the willingness to disagree with an available answer and work out a different one, was already identified as a competency at risk of underdevelopment in students who are rewarded for conformity. AI-assisted coursework adds a second pressure in the same direction, and a more constant one than social conformity alone. Where Edition 22 asked whether students are being taught to take intellectual risks, this edition asks whether they are getting the routine practice, assignment by assignment, that risk-taking and independent judgment require.

What IE Offices Can Measure and Protect

Three actions follow directly from the research above, and none require waiting for a campus-wide AI policy to be finalized.

● Audit assignment design in AI-permitted courses for evidence of enacted critical thinking, not just AI-tool competence. The Lee et al. finding that verification behavior is where critical thinking now concentrates suggests a specific question for program review: do current assignments require students to verify, challenge, or reason past an AI-generated starting point, or do they accept a polished AI-assisted output as the finished product?

● Use ASLO infrastructure to generate direct evidence rather than assume it. Edition 34 of this newsletter argued that most institutions already have the assessment infrastructure needed to document specific competencies at the student level; the gap has been in using that infrastructure for demonstration rather than compliance alone (Rudawsky, 2026c). The same infrastructure applies here. A rubric that already scores critical-thinking performance on a capstone or embedded assessment can, with modest adaptation, distinguish between work that shows independent reasoning and work that shows fluent AI integration without it. That distinction is now assessment-relevant in a way it was not five years ago.

● Consider where AI-free assessment has diagnostic value, not as a blanket policy but as a targeted measurement tool. Gartner’s prediction that employers will adopt AI-free skills assessments points toward a parallel opportunity in the classroom: a small number of proctored, unassisted assessments embedded at key points in a program can serve as a direct-evidence check on whether critical-thinking competencies are developing independent of AI availability, without requiring a wholesale retreat from AI-integrated instruction.

The Case for Treating This as an Opportunity

It would be easy to read the erosion research as one more reason for institutional anxiety. The more useful reading is the opposite. Employers are currently uncertain which candidates, and which institutions, can reliably demonstrate independent judgment in an AI-saturated environment. That uncertainty is itself a market opening. Institutions that can produce direct, current evidence, not an assumption inherited from a pre-AI curriculum, that their graduates reason well with and without AI assistance have a differentiating signal that few competitors can currently match. The demonstration gap has generally been framed as a deficit institutions need to close. In this specific case, closing it is also a recruiting advantage available to whichever institutions move first.

The underlying argument of this newsletter has been that higher education’s core challenge is proving what it produces, not producing it. This edition does not change that argument. It adds a reminder that production and demonstration are not always sequential. Where a new technology touches both the classroom and the hiring process at the same time, institutions that only work the demonstration side of the problem, and never verify that the underlying skill is still there to demonstrate, are solving half of a problem that has just become fully connected.

References

Dumitru, D., & Halpern, D. F. (2023). Critical thinking: Creating job-proof skills for the future of work. Journal of Intelligence, 11(10), Article 194. https://doi.org/10.3390/jintelligence11100194

Gartner. (2025, October 21). Gartner unveils top predictions for IT organizations and users in 2026 and beyond [Press release]. https://www.gartner.com/en/newsroom/press-releases/2025-10-21-gartner-unveils-top-predictions-for-it-organizations-and-users-in-2026-and-beyond

Hyken, S. (2026, July 26). Nine out of 10 leaders say life skills more important than AI/technical skills. Forbes. https://www.forbes.com/sites/shephyken/2026/07/26/nine-out-of-10-leaders-say-life-skills-more-important-than-aitechnical-skills/

Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (Article 1121, pp. 1–22). Association for Computing Machinery. https://doi.org/10.1145/3706598.3713778

Rudawsky, D. (2026a). The conformity challenge: Why higher education needs productive deviance. Institutional Effectiveness Weekly, Edition 22. https://donrudawsky.substack.com/p/the-conformity-challenge-why-higher

Rudawsky, D. (2026b). The demonstration gap: Why higher education struggles to show its value. Institutional Effectiveness Weekly, Edition 24. https://donrudawsky.substack.com/p/the-demonstration-gapwhy-higher-education

Rudawsky, D. (2026c). The infrastructure is already there: ASLO as a demonstration tool. Institutional Effectiveness Weekly, Edition 34. https://donrudawsky.substack.com/p/the-infrastructure-is-already-there

Shaw, S. D., & Nave, G. (2026). Thinking–fast, slow, and artificial: How AI is reshaping human reasoning and the rise of cognitive surrender (Wharton School Research Paper). SSRN. https://doi.org/10.31234/osf.io/yk25n_v1

Don Rudawsky, Ph.D. is the Founder of Dynalytic Solutions and former Vice President for Institutional Effectiveness at Nova Southeastern University. Through Institutional Effectiveness Weekly he provides evidence-based analysis for higher education leaders on analytics, strategy, and student outcomes. New editions publish Tuesdays at 9am.

Subscribe:

Connect on LinkedIn: https://www.linkedin.com/in/don-rudawsky/