What GPS, Phone Numbers, and Term Papers Can Teach Us About the MIT “Cognitive Surrender” Report

Ask yourself how many phone numbers you actually know by heart. A decade and a half ago you knew a dozen, easily. Today most people can't recall their own spouse's number without opening an app, and nobody sounded an alarm about it. We just quietly outsourced the skill and moved on, because a contact list reliably does the job, and there was no visible cost to letting the muscle atrophy.

We did the same thing with navigation. GPS didn't make us worse drivers. It made us worse at building a mental map, and ask someone to get home without the app and you'll watch the hesitation set in. That's not a character flaw. That's what happens to any skill that depends on repeated, unaided practice once you remove the practice, and the research backs this up directly. Drivers with more lifetime GPS experience show measurably worse spatial memory when navigating without it, and a follow-up study found that increased GPS use over time predicted a steeper decline in hippocampal-dependent spatial memory, not the other way around.

An Old Mechanism, Applied to Something Far Less Contained

MIT's Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training just documented the same mechanism happening to something far less contained: thinking itself. Their report, covered by The New York Times on September 18, 2026, describes students reaching for a chatbot at the first sign of friction, bypassing what researchers call “productive struggle,” the effortful, uncomfortable part of learning that's actually where the learning happens.

The result is what the committee calls an “illusion of learning.” Homework scores look fine because the AI is doing the cognitive lifting, while retention and test performance quietly decline. Office hours sit empty. Study groups are vanishing, not because students don't care, but because the friction that used to route them toward each other and toward struggle has been engineered away.

Losing your sense of direction is a real loss, but it's a contained one. Critical thinking is the general-purpose infrastructure underneath almost everything else.

Losing your sense of direction is a real loss, but it's a contained one. It doesn't degrade your ability to write an essay or evaluate a claim. Critical thinking isn't a narrow skill sitting next to navigation and phone-number recall. It's the general-purpose infrastructure underneath almost everything else: judgment, analysis, decision-making under uncertainty.

Cognitive scientists call this broader pattern “cognitive offloading,” using an external tool to reduce the mental demands of a task. Offloading a phone number or a driving route costs us relatively little. Offloading reasoning itself is different in kind, not just degree, and if we let that atrophy the way we let phone-number recall atrophy, the cost doesn't stay contained to one skill. It compounds into every downstream decision that depends on it.

This Isn't an Argument Against AI. It's an Argument Against Bad Evaluation Design.

The MIT report is clear that AI's potential to augment work on campus is immense. The problem isn't the tool. Most institutions are still grading output, the essay, the homework set, the finished answer, instead of grading process: how the reasoning was built, whether it can be defended, where it would break under a follow-up question.

When you only assess the output, a tool that produces convincing output looks identical to a student who actually did the thinking. That's not a student problem. That's an institutional design failure, and institutions have been slow to admit it because the fix requires more work than banning a tool or writing a policy memo.

We Used to Build This Into the Grade on Purpose. We Just Stopped.

“Show your work” was never about torturing kids in math class. It was the entire point. A correct final answer with no steps told the teacher nothing. It could mean mastery, a lucky guess, or a calculator doing the thinking. The steps were the only evidence the reasoning actually happened, which is the whole reason partial credit existed: the process itself was data, separate from whether the final number was right.

Term papers worked the same way, and anyone who went through school before search engines remembers exactly what that process cost. You went to the library. You pulled sources by hand, cross-checked them, took notes, built an outline, wrote a draft, revised it, and produced a final copy: legible, cited, defensible, without the benefit of undo. The finished paper was the majority of the grade, but the teacher had usually seen the bibliography, the draft, the messy in-between steps, even if only informally. The process wasn't graded as heavily as the product, but it was visible, and that visibility is exactly what's missing now.

A polished essay today can arrive with zero trace of how it was built. No draft. No source trail. No evidence the struggle ever happened, because the tool that produced it doesn't leave footprints the way a library trip or a handwritten draft did.

That's the real shift. Today's students aren't less capable or less honest than a term-paper writer sweating over index cards a generation ago. The infrastructure that used to make the process visible by default has quietly disappeared. Nobody removed “show your work” as a requirement. The steps just stopped generating evidence on their own, because the tool does them invisibly. The MIT report isn't describing a new problem. It's describing an old, well-understood teaching principle running into a generation of tools that erase the very trail that principle depended on.

This lines up with decades of learning research, not just one report. Robert and Elizabeth Bjork's work on “desirable difficulties” shows that the conditions which feel hardest in the moment, retrieval, spacing, self-testing, are the ones that produce durable learning, while the easy, frictionless path produces only an “illusion of knowing.” Separately, the well-documented testing effect shows that the act of retrieving information from memory strengthens it far more than re-reading or re-exposure does. Read together, the pattern is consistent: performance that comes easily right now is a poor predictor of what will actually be retained later, and that is precisely the gap the MIT committee is describing when it distinguishes rising homework scores from falling retention.

The fix isn't really a fix. It's a restoration. Oral defense, drafts with visible revision history, in-class problem-solving: these aren't new pedagogical inventions. They're “show your work” and “bring your sources,” rebuilt for an environment where the default no longer produces that evidence for free.

A Readiness Gap, Not a Technology Debate

The split the Times reporting surfaces, administrators eager to embrace AI, faculty and students far more wary, isn't really a disagreement about technology. It's a readiness gap. It's what happens whenever an organization adopts a powerful capability faster than it redesigns the systems that were supposed to create trust in the outputs. Higher education is living through, in real time, the exact failure mode legacy organizations hit when they bolt AI onto old evaluation structures instead of rebuilding around what the technology actually changed.

You can't fix what you can't see, and right now, most institutions can't see the difference between a student who did the thinking and a tool that did it for them. We already ran this experiment once, quietly, with GPS and phone numbers, and shrugged because the stakes felt small. This time the stakes aren't small. The question isn't whether to use the tool. It's whether we redesign how we verify thinking before the gap between “looks like learning” and “is learning” gets too wide to see.

Sources

Arsenault, M., Goldstein, D., Blinder, A., & Mervosh, S. (2026, September 18). Universities sound warnings about A.I. even as they embrace it. The New York Times.

Massachusetts Institute of Technology, Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training. (2026). Committee report on generative AI in teaching, learning, and research training, as reported in The New York Times and The Philadelphia Inquirer, September 2026.

Dahmani, L., & Bohbot, V. D. (2020). Habitual use of GPS negatively impacts spatial memory during self-guided navigation. Scientific Reports, 10, 6310.

Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688.

Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19.

Bjork, R. A., & Bjork, E. L. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In M. A. Gernsbacher et al. (Eds.), Psychology and the Real World: Essays Illustrating Fundamental Contributions to Society. Worth Publishers.

Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.

Note: The MIT committee report is cited here as reported through New York Times and Philadelphia Inquirer coverage. For direct quotation or formal citation, consult the original committee report and the September 18, 2026 New York Times article directly.