Positive Technology

Are We Losing the Ability to Think?

are-we-losing-the-ability-to-think

There’s an interesting change taking place in the way we think, and in my opinion, we aren’t discussing it enough. Suppose you’re in a vacant room, with no distractions and no one else present, and all of a sudden you have an idea. You take out your phone or laptop and begin to write. Your thoughts are flowing, and the sentences are forming, and for a while you’re actually enjoying the act. However, at that moment something stops you. You break off and ask yourself if the sentence you’ve just written is grammatically correct. Whether the words have been arranged properly. And whether the idea makes sense to somebody other than yourself. And before you realise it, you’ve gone to ChatGPT. 

You ask it to fix your grammar, to reorganise your sentences, or maybe to improve the way your writing sounds. It carries out these tasks in a matter of seconds, and you feel a sense of relief. But have you ever stopped to think about why you needed that kind of assurance in the first place? When did we become so doubtful about our own ability to construct a sentence? Did we really become so dependent on machines that we now rely on them to confirm something that we are completely capable of doing ourselves?

It’s not just a matter of artificial intelligence taking over jobs or replacing human creativity. Rather, it is something far more personal. It raises an important question. What happens when we gradually begin to outsource the very processes involved in how we think, remember, learn, and make decisions? To be fair, a lack of confidence is not the only reason we reach for AI. Often it is simply convenience; it is quicker. Sometimes it is time constraints, a deadline and an inbox that won’t wait. And sometimes it is cognitive efficiency, a perfectly sensible decision to save our mental energy for the parts of the task that matter more. 

Cognitive Offloading: When Our Brain Gets an External Assistant 

There is a term that psychologists use for this phenomenon; it is known as cognitive offloading. What cognitive offloading basically amounts to is transferring part of our mental effort to an external device so that our brain does not have to carry out all the tasks on its own. For example, writing a list of groceries rather than trying to remember ten items has been commonplace for a long time. Setting reminders instead of remembering our appointments is also common. Similarly, using a calculator rather than working out calculations in our heads has been a common practice.

As Evan Risko and Sam Gilbert point out in their 2016 paper in Trends in Cognitive Sciences, people intentionally make use of external resources in order to lower the cognitive demands placed on them. This is not always a negative development. Since our brains have a limited capacity for working memory, offloading some of the tasks can make us more efficient.  It helps to separate ordinary cognitive offloading from excessive dependence on AI. Ordinary offloading is occasional and deliberate, and it leaves us in charge: we hand over a routine job and keep hold of the thinking that matters. Excessive dependence is habitual and automatic: we reach for the tool before we have made any attempt ourselves and find it hard to carry on without it. 

Now this is where it becomes interesting. There is a distinction to be made between using a tool to remember something and using a tool to carry out the kind of thinking that should have enabled us to remember it initially. A study carried out in 2011 by Betsy Sparrow, Jenny Liu and Daniel Wegner and published in Science gave rise to what is commonly referred to as the Google Effect.

Their experiments showed that people are less likely to remember information itself when they expect it to be available online. Instead, they are more likely to remember where they can find it. The internet has thus become an external memory system.  Maybe AI is going one step beyond this. Instead of just asking Google where something is, we are now asking AI to explain it, interpret it, organise it, write it, and at times even decide what we should think about it. 

From Remembering Information to Outsourcing Understanding 

In the past, when we wanted to grasp a complicated topic, we had to read various sources. We had to struggle with new concepts and link one idea to another. We also made errors. In the end, we reached our own understanding. Nowadays, however, we can ask an AI programme to summarise the whole subject in just a few seconds. The response could be excellent and might even be better organised than anything we could have put together ourselves. But then again, how much of that understanding actually becomes our own?  To see why this matters, we have to look at how long-term learning works.

Active engagement, meaning that we do something with the material rather than just read it, helps information to stick. Retrieval practice, which is trying to recall something from memory without looking, strengthens memory more than simply reading it again. Elaboration, where we explain an idea in our own words and link it to what we already know, builds the connections that deepen understanding. When AI summarises a subject for us, it can skip all three, and the effortful parts that help learning take root are the very parts that get offloaded. That is the link between cognitive offloading, memory retention and deep learning. 

A randomised controlled trial was carried out in 2025 by André Barcaui. It was published in Social Sciences & Humanities Open. The study involved 120 undergraduate students who were studying AI. It examined how much they remembered about the subject after 45 days. The group that had used ChatGPT achieved 57.5% in the retention test, whereas those who had studied using traditional methods scored 68.5%. This finding is specific to this study. It involved one group of students, one subject, and one way of using ChatGPT. Therefore, it cannot be taken as evidence that AI generally reduces retention.

It does, however, raise a significant question as to what we lose when learning becomes too easy.  A study carried out in 2025 by researchers from Microsoft Research and Carnegie Mellon University, which was presented at the CHI Conference, looked into the practices of 319 knowledge workers and gathered 936 instances of AI-assisted work. The study showed that higher confidence in generative AI correlated with a reduction in the amount of critical-thinking effort reported. We are not simply stopping our thinking altogether; instead, we may be altering the kind of thinking we do. Rather than having to struggle to find an answer, we are increasingly assessing an answer that has already been provided to us. 

The Paradox of Trust: Why We Believe AI More Than Ourselves 

A study from 2024, which was published in Nature Machine Intelligence, looked into the way people assess the reliability of large language models. It discovered that users tend to overestimate the accuracy of the responses generated by AI, especially when the explanations are given with confidence or are longer. This situation leads to an interesting contradiction: we turn to AI precisely because we are doubtful about our own judgement, yet the confidence displayed by AI in its communications causes us to be less inclined to question its judgement. 

Psychology has a few ideas that help here. Automation bias is the tendency to favour suggestions from automated systems and to go along with them even when our own judgement or other evidence points elsewhere. Cognitive trust is the extent to which we rely on a source because we see it as competent and reliable. And metacognition, our awareness of how well we actually know or understand something, is what would normally tell us when to question a source. When AI sounds confident, it can gain our trust without having earned it, and our metacognition may be slower to raise the alarm: a polished answer feels like a correct one. 

Maybe it is in this way that dependence starts to develop. It doesn’t begin when we cease to have ideas, but rather when we stop thinking that our ideas are good enough on their own without needing external approval. In 2025, researchers from MIT published a paper called Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing. They looked at 54 participants who were divided into groups consisting of those who used an LLM, those who used a search engine, or those who used neither while writing essays.

By means of EEG measurements, they found differences in brain connectivity among the groups, with the group that relied only on their brains showing the most intense and widespread connectivity. The group that used the AI also reported less ownership of their essays and had more difficulty remembering their own writing. The study in question is a preprint, which means it has not yet been through full peer review, and it is based on a fairly small sample; therefore, it cannot be regarded as conclusive proof that AI is weakening our brains.

The differences in EEG connectivity show that the brain was working differently during the task; they do not establish cognitive decline or brain damage. Nevertheless, it does provide an interesting avenue for further research, namely, what occurs when the process of creating something has become so aided that we no longer feel fully connected to what we have created? 

The Thinking That Happens Between the Words: It’s How We Use AI, Not Why

This is something that applies directly to the subject of writing. Writing involves more than merely arranging words; it is a way of working out what we really think. There are occasions when we start to write without knowing exactly where a paragraph is going to end up. The effort involved in this process is not wasted; it is an integral part of thinking.

In a study from 2014, Pam Mueller and Daniel Oppenheimer discovered that students who took notes by hand did better on questions that required conceptual understanding than those who took notes on laptops, the reason being that handwriting promoted more processing rather than just a verbatim transcription. That study was about note-taking, not AI-assisted writing, so it does not directly establish the effects of using AI to write. The point is not that technology is in itself harmful, but that the manner in which we deal with information is important. 

At the same time, it would be unjust to present this article as being about how AI is destroying our cognitive abilities. There is no research to back up such a simple conclusion. A systematic review and meta-analysis carried out in 2025 by Ruiqi Deng and his colleagues, which looked at 69 experimental studies, found that ChatGPT was able to improve academic performance, motivation and the tendency towards higher-order thinking, as well as reduce mental effort. There were limitations too. The authors themselves pointed to weaknesses in many of the included studies, such as small samples, a lack of rigorous baseline controls and simple measures of learning, and they said it is unclear whether the gains in motivation last beyond the novelty of using a new tool.

Another meta-analysis in 2026 by Xinning Wu and his colleagues, based on 35 experimental studies and involving 4,193 participants, likewise found positive effects on student learning outcomes. And these results say little about what happens to long-term retention, which is what the earlier studies were concerned with.  In an experiment carried out in 2023 and published in Science, Noy and Zhang involved 453 professionals and found that having access to ChatGPT cut the time needed to complete writing tasks by 40% and at the same time improved the quality of the output by 18%. 

Maybe the issue doesn’t lie in cognitive offloading itself but in the kind of offloading that we choose to do. There is a distinction between productive offloading and outsourcing the core processes of learning. Productive offloading frees up mental effort for the work that builds understanding; tidying the grammar of a draft that we thought through ourselves is an example. Outsourcing the core learning processes, such as working out the argument, struggling with the concept or forming the opinion, removes the very effort that learning depends on.

The former helps to improve our ability; the latter might gradually lead to our becoming dependent on an ability that we never actually practise. I believe that we should focus on this distinction, since dependence doesn’t always appear in the form of laziness; at times it takes the shape of perfectionism. We hope that our writing is faultless, that our decisions are right, that our arguments are convincing and that our thoughts come across as intelligent. For this reason, we become so afraid of making a mistake that we begin to seek an external source for reassurance at each stage, and AI is always there to give that reassurance.

Conclusion: Learn to Think Unassisted Again 

We need not reject AI if we want to preserve our ability to think. Instead, all we have to do is ensure that we are not handing over all the unpleasant aspects of thinking to it. This is because the role of technology should be to give us more space to think rather than make thinking seem like an unnecessary thing to do. Perhaps the next time we find ourselves in an empty room with an idea, we should begin to write. We should then go ahead and complete that first draft without consulting the AI to find out if it’s good enough.

It can be grammatically flawed. The thoughts might be a bit scattered, and the sentences could sound like our own. We can always go on to improve them later on. But I wonder if we continually ask a machine to decide whether our thoughts are worth expressing.

Will we eventually reach a stage at which we can no longer see their value without the machine’s input? And maybe that is the real cost of cognitive offloading that we should be discussing. It’s not just about forgetting information, but about gradually losing the ability to trust the mind that originally produced it. 

References + 
  • Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002 
  • 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 
  • Risko, E. F., & Dunn, T. L. (2015). Storing information in the external world: Metacognition and cognitive offloading. Consciousness and Cognition, 36, 398–410. https://doi.org/10.1016/j.concog.2015.03.006 
  • Gilbert, S. J. (2015). Strategic offloading of delayed intentions into the external environment. Quarterly Journal of Experimental Psychology, 68(5), 971–992. https://doi.org/10.1080/17470218.2014.972963 
  • Meyerhoff, H. S., Papenmeier, F., & Huff, M. (2021). Individual differences in cognitive offloading: A review and framework. Psychonomic Bulletin & Review, 28, 923–940.  https://doi.org/10.3758/s13423-020-01836-8
  • Mueller, P. A., & Oppenheimer, D. M. (2014). The pen is mightier than the keyboard: Advantages of longhand over laptop note taking. Psychological Science, 25(6), 1159–1168. https://doi.org/10.1177/0956797614524581
  • 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. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3706598.3713778
  • Kosmyna, N., Hauptmann, E., Yao, Y., et al. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. MIT Media Lab (Preprint). https://arxiv.org/abs/2506.08872 
  • Barcaui, A. (2025). ChatGPT as a cognitive crutch: Evidence from a randomised controlled trial on knowledge retention. Social Sciences & Humanities Open, 12, 102287. https://doi.org/10.1016/j.ssaho.2025.102287 
  • Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6. https://doi.org/10.3390/soc15010006
  • Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187–192.  https://doi.org/10.1126/science.adh2586 
  • Deng, R., Jiang, M., Yu, X., Lu, Y., & Liu, S. (2025). Does ChatGPT enhance student learning? A systematic review and meta-analysis of experimental studies. Computers & Education. https://www.sciencedirect.com/science/article/pii/S0360131524002380
  • Pardos, Z. A., & Bhandari, S. (2024). ChatGPT-generated help produces learning gains equivalent to human tutor-authored help on mathematics skills. PLOS ONE, 19(5). https://doi.org/10.1371/journal.pone.0304023 
  • Polakova, P., & Klimova, B. (2024). The impact of ChatGPT feedback on the development of writing skills. Cogent Education.  https://doi.org/10.1080/2331186X.2024.2410101 
  • 1Zawacki-Richter, O., Marín, V. I., Bond, M., & Gouverneur, F. (2019). Systematic review of research on artificial intelligence applications in higher education – Where are the educators? International Journal of Educational Technology in Higher Education, 16, 39.  https://doi.org/10.1186/s41239-019-0171-0
  • Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19. https://doi.org/10.1093/analys/58.1.7 
  • Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285.  https://doi.org/10.1207/s15516709cog1202_4
  • Soderstrom, N. C., & Bjork, R. A. (2015). Learning versus performance: An integrative review. Perspectives on Psychological Science, 10(2), 176–199. https://doi.org/10.1177/1745691615569000 
  • Dunn, T. L., & Risko, E. F. (2016). Toward a metacognitive account of cognitive offloading. Cognitive Science, 40(5), 1080–1127. https://doi.org/10.1111/cogs.12273 
  • Ryan, R. M., & Deci, E. L. (2020). Intrinsic and extrinsic motivation from a self-determination theory perspective. Contemporary Educational Psychology, 61,101860. https://doi.org/10.1016/j.cedpsych.2020.101860
Exit mobile version