It’s board meeting day, and I get to present our financial update. The trouble is, I’ve spent the past week in back-to-back, nose-to-the-grindstone activity, with barely a moment to analyse our recent performance. And yet I’m not worried. One hour ago I fed our numbers into a business intelligence AI and generated a brilliant set of insights that would once have taken me half a day or more to produce. It made sense of information I’d usually skip over, cross-referenced data from previous years, and benchmarked us against similar companies. A quick read-through and I’m ready to present.
My team are impressed. Our advisors nod sagely. Win-win-win, right? Maybe. But something else is also true, and it’s the thing that will catch up with thousands of leaders like me one, five, or eight years from now. I have skipped a process. My apparent financial insight and analytical fluency are borrowed from AI, not earned by effort. And this cognitive borrowing is accruing interest in a currency none of us have yet learned how to account for.
In a previous article, A Recipe for Self-Deception, I argued that letting AI displace human dialogue is a recipe for confident self-deception. But there is a second-order problem underneath it. Even when leaders use AI critically, we are taking on a debt that doesn’t show up on any financial statement: cognitive debt. The interest on it compounds, and unlike financial debt, you cannot simply refinance it.
What is cognitive debt?
My five-year-old is learning the piano. For him, nothing is more exciting than the ‘demo’ button on our electric piano, where he can pretend to play a perfect Mozart sonata as it pumps out of the speakers. All the expertise, with precisely no effort. We know there’s a big gap between him and the real skill, though. He needs to strengthen his tiny fingers, learn his scales, read music and practise endless nursery rhymes before he ever gets near Mozart.
That is how mastery is learned, over time. It’s hard, sometimes boring, and along the way his brain and body build the neurological and muscle memory, piece by piece. If he ever reaches that sonata, he’ll be able to play it for life. He’ll have earned it. That is cognitive growth. Cognitive debt is the opposite: it is produced when we press the ‘demo’ button too often, using AI to replace our learning, reflection and embodied understanding. It accrues whenever we produce things, or information, without having put in the effort to earn it.
What cognitive debt does to our performance
Margaret-Anne Storey, who teaches software engineering at the University of Victoria, watched this play out in real time with her entrepreneurship class. Student teams sprinting with AI-assisted code initially produced incredible outputs at record speed. But they hit a wall in week seven. They could no longer make even simple changes without breaking something unexpected.
Their first instinct was to blame technical debt: messy code and hurried implementation causing errors. But the problem was deeper. No one on the team could explain why certain design decisions had been made, or how the parts of the system were meant to work together. The shared understanding of what success took had fragmented or disappeared. Everyone was getting faster while skipping the experience that turns into replicable wisdom. They had accumulated cognitive debt faster than technical debt, and it paralysed them. What is true for a student team is also true for a department, a leadership team, an industry, and ultimately the human race.
Cognitive debt is accrued at three levels
The individual: a quietly dimming brain
A four-month MIT Media Lab study tracked brain activity in three groups writing essays: one using ChatGPT, one using a search engine, and one going human-only. For the tech users, a dimming pattern was consistent and worsened over time. Brain rhythms tied to deep memory formation declined across all four months. Synchronisation between executive reasoning and memory retrieval areas weakened. By the final session, 78% of the ChatGPT group couldn’t quote anything from essays they had written minutes earlier. In the no-AI group, that figure was just 11%.
A separate study showed AI-using students outperforming a control group by 22 points on immediate recall. Three weeks later that gap had collapsed to 6 points and was no longer statistically significant. For higher-order tasks like synthesising, evaluating and applying information, the no-AI group led every time. The takeaway is tough to chew on: reliance on AI is already dimming our individual brain power.
The organisation: a quietly thinning culture
Etienne Wenger’s work on communities of practice describes how true organisational knowledge doesn’t live in documents or processes. It lives in the relationships, conversations and shared repertoires of professional communities. Newcomers enter through what Wenger calls legitimate peripheral participation: they take on the small jobs, sit in the rooms, watch their seniors work, and gradually become central. That on-ramp is exactly what AI substitution removes most efficiently.
When the junior analyst no longer has to do the painful pivot table, the messy customer call, the late-night reconciliation, the work still gets done. But the apprenticeship through which the analyst would have become someone who understands the business is quietly gone. They will be promoted on their AI-assisted outputs, and in seven years they will walk into the leadership meeting with thin recall and limited embedded experience to fall back on.
The wider commons: a quietly emptying granary
Imagine you visit your doctor with a suspicious-looking mole. For now, she looks at it and makes a judgment that helps you. Two months later you return and she can verify whether her original diagnosis was right. Perhaps a colleague has a look too. Over time, each doctor-patient interaction creates a body of shared, three-dimensional experience: a small but important deposit into our collective bank of human knowledge that tells us what to do if.
In an AI-dominated world, we get our mole checked on a phone app. Perhaps it’s more accurate than a hard-to-reach doctor, and everyone saves time. But the doctor sees fewer moles, and piece by piece the skill and discernment fade. We stop paying into the shared bank. Then diseases mutate, a new type of mole emerges, and there are no longer experts who can tell what is or isn’t healthy. We have drained the account.
This may sound like a distant world, but it is already here. Code generated by models trained partly on AI-generated code shows subtle, systematic degradation. AI-generated text used to train subsequent models produces measurable declines in diversity, accuracy and coherence over iterations. It becomes a photocopy of a photocopy of a photocopy. Economist Daron Acemoglu’s conclusion is uncomfortable: past a certain social optimum, more capable AI makes people worse off.
Why cognitive debt is hard to see
The sneakiest property of cognitive debt is that it feels nothing like debt while it is accumulating. Financial debt arrives with a statement. Technical debt eventually announces itself through a failing build. Cognitive debt does the opposite: as it accrues, teams report feeling more capable, not less. We are getting more done. The outputs look better. Confidence is up. The board meeting goes well.
The deficit only becomes visible years later, when we can neither navigate nor think without AI. When the AI is wrong and no one in the room knows enough to notice, let alone fix it. When there’s a power cut and suddenly nobody knows anything. The trap is elegant and terrible: every individual decision to lean on AI is rational. But little by little, what will we lose?
Now what?
Cognitive debt is a real threat to our individual intelligence, our teams and our socially built knowledge. We have to start taking it seriously as we make decisions about our organisations and our educational systems. But I don’t want to paint too dark a picture. Plenty can be done to swerve the impacts of cognitive debt, but only if we are first aware of it.
We’ll look at what can be done in the next article. In the meantime, I’d love to hear your thoughts. When do you notice cognitive debt in yourself, and in your system? And what consequences do you predict?
Sarah Lloyd-Hughes
The UK’s leading inspiring speaking expert & best-selling author. Sarah Lloyd-Hughes is a multiple-award winning public speaking coach, founder of Ginger and author of “How to be Brilliant at Public Speaking” (Pearson).
