It is easy to forget that the public launch of ChatGPT is only a few years behind us. In that short span, generative AI has become ubiquitous, and it has already changed how many of us work and think. The open question — for each of us individually, and for us collectively — is whether it makes us sharper or duller.
This piece is about the practical side of that question. An earlier piece in this series argued that the human abilities AI cannot reach are becoming the real differentiator. Here I want to be concrete about how I try to protect and expand my own cognitive capacity in daily practice — not as a prescription, but as one working approach.
What is AI doing to how we think?
Open LinkedIn on any given morning and the shift is hard to miss: a stream of generic AI-generated posts and comments, guides on running a business with a team of AI agents, and experts promising the one prompt that will change everything.
Alongside that, the evidence on the other side is accumulating — and it is more concrete than it was even a year ago. The recurring theme in this research is cognitive offloading: the more we hand a mental task to a tool, the less our own faculties do the work, and abilities we do not exercise tend to weaken.
A large MIT Media Lab study used EEG to watch what happens in the brain when people write with ChatGPT versus with a search engine versus with nothing at all. The pattern was clear: the more external help people used, the weaker and less connected their brain activity was during the task — and those who wrote with AI showed the lowest engagement, reported the least ownership of their own essays, and often could not quote a sentence they had produced minutes earlier. The researchers call the effect "cognitive debt."
A study of over 600 people by Michael Gerlich, published in the journal Societies, found a strong negative relationship between frequent AI use and critical-thinking scores, with cognitive offloading as the mechanism driving the decline — an effect most pronounced in younger, heavier users.
Research from Microsoft and Carnegie Mellon with knowledge workers found that the more people trusted an AI's output, the less critical thinking they applied — their effort shifting from doing the work to simply checking and stitching together what the machine produced.
These are early findings, not final verdicts. But the direction is consistent enough to take seriously: used carelessly, AI does not just save effort — it can quietly erode the faculties we stop using. Alongside them sit reports of AI-generated "workslop" dragging down real productivity, and accounts of AI-fuelled episodes serious enough that clinicians have started to note them.
The honest summary is that we have all been enrolled in one of the largest social experiments in human history, with very few guardrails in place. That is not a reason for panic. It is a reason to be deliberate.
How can we use AI responsibly?
The genuinely useful question is whether we are using AI to become more capable — to do things we could not do before, to hand off tedious work so we have more room for creative and consequential thinking — or whether we are letting it make us lazier, sloppier, and more impatient, slowly outsourcing the very faculties that make us discerning.
Both outcomes are on the table. At the collective level, the jury is still out on which will dominate. At the individual level, the choice is ours to make — and choosing not to choose is itself a choice, one made on our behalf by whatever the default happens to be.
I do not need more evidence to confirm what I already sense: that AI use is a slippery slope, with real potential to erode cognitive ability if we are not careful. So I have set guardrails for myself.
What do I refuse to use AI for?
My rule of thumb is simple. If I need a particular ability to think, express myself, and hold my own in a real conversation about something complex, I do not outsource that ability to AI. These are core competences related to sense-making, self-expression, and relating to other people, and I want them to stay mine.
In practice, that means I avoid using AI for:
Initial ideation — I capture my own ideas first, before I let any model near them.
Structuring my thinking — whether it is a post or a workshop, I build the flow myself.
Interpreting complex situations or forming an opinion — sense-making is the foundation of sovereignty, and outsourcing it is, in a real sense, giving away freedom.
Formulating full sentences for an original thought-leadership piece — articulating our thoughts is bound up with discovering and shaping who we are. Curating AI output is not the same as using my own voice.
Personal emails and direct messages — authentic communication is central to building real relationships.
Replacing ongoing learning — I want to hold new knowledge in a way that lets me actually recall and use it.
A stand-in therapist — these systems tend to be sycophantic and to reinforce whatever perspective you seed them with, which can feed distortion rather than clarity.
Deep metaphysical questions — a language model is not a holder of wisdom.
What do I actually use AI for?
The list of what I do use it for is just as deliberate. I use AI:
To complement and widen my perspective — surfacing additional viewpoints and arguments, but only after I have articulated my own take first.
To expand my vocabulary — exploring alternative ways of saying something.
To find additional external sources relevant to what I am working on.
To map a new field — building a quick overview that gives me entry points to learn more.
To condense and edit for clarity — tightening my writing while keeping the core thought intact.
To create short-form derivatives of something I have already written — a social post from an article, a summary of a workshop.
For tedious administrative work — from navigating bureaucracy to filing a claim for a cancelled flight.
The pattern is consistent: AI comes in after the original thinking is done, to extend and support it, not to replace it.
How do I use AI without letting it use me?
The how matters as much as the what. I write prompts that are deliberate and as precise as I can make them. I critically evaluate everything before I include it — and I check original sources before quoting anything, without exception.
And I run a gut check before using any output. Quite often I dismiss a suggestion because it feels too cliché, unoriginal, or impersonal to be mine. Interestingly, the more I work with AI, the more appreciation I find for human complexity, creativity, and ingenuity — and the more I hope we do not trade them away for convenience.
Does anyone actually get this right?
I do not have it figured out, and my approach will keep evolving. I do not hold to my own guardrails perfectly either — I aim for something closer to an eighty-twenty ratio that accounts for the messiness of real life. I also do not believe in a single correct way to use AI responsibly, and I am not judging anyone who sees it differently.
But I do think we need to be conscious, and a little cautious, about how we use this technology and what we set in motion for ourselves. Every choice we do not make consciously is an unconscious acceptance of a choice made for us. For anyone leading others, that principle carries even further than their own desk.
The next piece in this series turns from guardrails to cultivation — the practices and environmental choices that help me access the higher-order abilities AI cannot reach.
Sources & further reading
Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. MIT Media Lab. arXiv:2506.08872 (preprint). https://arxiv.org/abs/2506.08872
Gerlich, M. (2025). AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1), 6. https://www.mdpi.com/2075-4698/15/1/6
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. Microsoft Research & Carnegie Mellon University (CHI 2025). https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/
Frequently Asked Questions
How can I use AI without eroding my own thinking?
Set clear guardrails. A workable rule of thumb is to avoid using AI for any ability you need to think, express yourself, and hold your own in a real conversation about something complex — such as initial ideation, structuring your thoughts, interpreting situations, and forming opinions. Bring AI in after the original thinking is done, to extend and support it rather than replace it.
What should I not use AI for?
Consider keeping AI out of core competences related to sense-making, self-expression, and relating to others: capturing your first ideas, structuring your own thinking, forming opinions on complex situations, writing personal messages, and ongoing learning you want to retain. Using it as a therapist or a source of metaphysical guidance is also risky, since these systems tend to reinforce whatever perspective they are given.
What are good uses of AI for knowledge work?
Useful applications include widening your perspective after you have formed your own view, expanding vocabulary, finding additional sources, mapping an unfamiliar field, condensing and editing for clarity, generating short-form derivatives of your own content, and handling tedious administrative tasks. The common thread is that AI supports thinking you have already done rather than substituting for it.
Is there evidence that AI use affects cognition?
Yes, and it is growing. A large MIT Media Lab study using EEG found that people who wrote with an AI assistant showed weaker, less connected brain activity, felt less ownership of their work, and often could not quote what they had just produced — an effect the researchers call “cognitive debt.” A study of over 600 people published in the journal Societies found a strong negative link between frequent AI use and critical-thinking scores, driven by cognitive offloading, and research from Microsoft and Carnegie Mellon found knowledge workers applied less critical thinking the more they trusted AI output. These are early findings rather than final verdicts, but the direction is consistent enough to use the technology deliberately.
How strictly should I follow AI guardrails?
Perfection is not the goal. A realistic target is something like an eighty-twenty ratio that accounts for the messiness of real life. What matters most is consciousness and a degree of caution — because every choice we do not make deliberately is an unconscious acceptance of a default chosen for us.