This essay is part of the Minds in the Machine Age series. Read the overview and full reading order on the Minds in the Machine Age page.
There is a moment that most of us now recognise — you are mid-sentence, reaching for a word or a fact you know you know, and instead of pausing to let it surface, your hand moves to your phone. The search result arrives in seconds. The mental friction dissolves. And somewhere, quietly, the memory never quite forms.
This is cognitive offloading — the practice of using external tools to handle mental work our brains would otherwise do. Humans have always done it: we write lists, draw maps, annotate books. But something is shifting. As AI tools become fluent conversation partners capable of summarising, solving, and explaining almost anything on demand, we are outsourcing not just the storage of information but the process of working through it. That process, it turns out, is where most of the learning actually happens.
The Effort Is the Point
In 1978, Norman Slamecka and Peter Graf published a study that would become foundational in learning science. Participants who generated answers from partial cues — even simple word-completion tasks — remembered them far better than those who simply read the answer. This became known as the generation effect: the act of producing information, rather than receiving it, encodes it more deeply in memory.
It is an inconvenient finding in the age of AI autocomplete.
Robert Bjork, a cognitive psychologist at UCLA, has spent decades building on this insight. His concept of desirable difficulties — challenges that feel frustrating in the moment but strengthen long-term retention — runs directly counter to the experience AI tutors are designed to deliver. We want learning to feel smooth. But smoothness, Bjork argues, is often a signal that very little is being retained.
When an AI tool writes the first draft, summarises the paper, solves the equation, or explains the concept before we have had a chance to wrestle with it, it short-circuits exactly the kind of productive struggle that consolidates understanding. The answer arrives. The learning does not.
What Metacognition Has to Do With It
The deeper concern is not just memory. It is metacognition — our ability to monitor and regulate our own thinking. Knowing what you know, knowing where you are confused, knowing when an explanation has actually landed: these are not byproducts of learning. They are the engine of it.
A 2011 study by Betsy Sparrow, Jenny Liu, and Daniel Wegner — published in Science — found that people who expected to have access to information online were less likely to remember it, and more likely to remember where they could find it. The researchers called this the Google effect: we adapt our memory strategy to the tools available. We store access paths rather than content.
AI accelerates this tendency dramatically. When a student can generate a polished essay on any topic in seconds, the motivation to internalise the underlying knowledge — to truly understand the argument structure, the evidence, the counter-positions — weakens. Why carry something in your head when the machine carries it for you?
The problem is that metacognitive awareness requires having actually grappled with material. You cannot know you don’t understand something if you never encountered the confusion in the first place.
Not the Tool — How the Tool Is Used
None of this is an argument against AI in education. It is an argument for clarity about what AI can and cannot do for a learner.
Used well, AI tools can function as what Lev Vygotsky called a zone of proximal development scaffold — giving just enough support to help a learner reach the next level of understanding, then stepping back. An AI that asks questions rather than answers them. A tool that offers a framework and invites the student to fill it. Adaptive systems that withhold the solution until the learner has attempted it.
Used poorly — which is to say, used as a replacement for thinking rather than a support for it — AI risks producing what Annie Murphy Paul, in her book The Extended Mind, describes as cognitive outsourcing at scale: a generation that becomes highly competent at directing tools without developing the underlying knowledge structures those tools depend on.
The practical difference between these two modes is not always obvious. A student using AI to check their reasoning after they have attempted something is learning. A student using AI to avoid the attempt entirely is not. The behaviour looks similar from the outside. The cognitive outcome is entirely different.
Educators are beginning to grapple with this distinction — not by banning AI tools, but by redesigning the conditions around them. Some researchers advocate for interleaved practice structures: AI-assisted study alternating with periods of unassisted retrieval. Others emphasise process documentation — asking students to show not just the answer but the reasoning, the wrong turns, the revision.
What We Carry
There is something worth sitting with in all of this. The history of human cognition is a history of tools that extended our minds — writing, printing, calculating machines, search engines. Each technology changed what we needed to know and how we knew it. Each brought legitimate gains and legitimate losses.
AI is not different in kind. It is different in scale, speed, and the intimacy with which it integrates into thinking itself. The question is not whether to use these tools, but whether we are being intentional about what we are keeping. What knowledge we are choosing to carry in our own bodies and minds, rather than delegating to a system that — however capable — cannot carry meaning for us.
The word you were reaching for at the start of this essay? You probably know it. Give it a moment.
References
- Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592–604. https://doi.org/10.1037/0278-7393.4.6.592
- Bjork, R. A. (1994). Memory and metamemory considerations in the training of human beings. In J. Metcalfe & A. Shimamura (Eds.), Metacognition: Knowing about knowing (pp. 185–205). MIT Press.
- Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google effects on memory: Cognitive consequences of having information at our fingertips. Science, 333(6043), 776–778. https://doi.org/10.1126/science.1207745
- Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes. Harvard University Press.
- Paul, A. M. (2021). The extended mind: The power of thinking outside the brain. Houghton Mifflin Harcourt.
→ Next: The Quiet Bias: How AI Systems Reflect Our Blind Spots
Related reading from Between Brain & Binary: Looking Ahead: The Future of Mind and Machine
Continue in this series: The Quiet Bias: How AI Systems Reflect Our Blind Spots. Or return to the Minds in the Machine Age overview.
Comments
Post a Comment