R3 3.2 February 24, 2025 Cognitive Offloading, Cognitive Abilities, and AI

What new research says (and does not say) about impacts of using generative AI on critical thinking.

I’ve been keeping a close eye out for new empirical work that focuses on generative AI, as I know many of you have as well. But, as I’ve observed before, development of a deep and substantial research body is going to take time. We’re still at an early stage as academic work goes, and so it’s perhaps not surprising that a lot of what I’m seeing in the journals is more along the lines of talking about AI than studying it directly: discussing potential uses, risks, offering conceptual frameworks for understanding it and so on.

This is good, useful work (and I have done a fair amount of such opining and predicting myself, for example here and here). But I do pay special attention to work that does fall more in the direct-investigation category, and so I was keenly interested in a new piece from SBS Swiss Business School researcher Michael Gerlich – even more so given that the work relates to my other current focus, critical thinking.

This article is going to continue to make waves, and certainly has motivated me to take a close look at the findings. In today’s post, I’m going to go off of my usual format and instead discuss the details in a more narrative form.

Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 1–28. https://doi.org/10.3390/soc15010006

From the title of the article alone, it’s clear why so many folks in education are taking notice. There were actually two main components of the study, one quantitative and one qualitative (involving semi-structured interviews that were analyzed for major themes and insights). I’m going to focus just on the quantitative one in this post, but definitely take a look at the original if you’d like to get a sense of what people said about their perceptions of AI in those interviews.

Starting off with a brief overview of the methodology and findings, the research team surveyed individuals residing in the UK who were recruited via social media. They gathered responses from 669 people in age categories ranging from young adult (17-25) to older adult (55+). They asked respondents a short but targeted set of questions designed to get at three things in particular: their critical thinking prowess, how much they used generative AI tools in day to day life, and the degree to which they engaged in cognitive offloading.

The critical thinking items were adapted from two established measures of general critical thinking ability, the Halpern Critical Thinking Assessment and questions developed by Patrick Terenzini and colleagues. The subset of questions used were as follows (all rated on a scale of 1, Never, to 6, Always):

· How often do you critically evaluate the sources of information you encounter?

· How confident are you in your ability to discern fake news from legitimate news?

· When researching a topic, how often do you compare information from multiple sources?

· How frequently do you reflect on the biases in your own thinking when making decisions?

· How often do you question the motives behind the information shared by AI tools?

· I analyse the credibility of the author when reading news or information provided by AI tools.

· I compare multiple sources of information before forming an opinion based on AI recommendations.

· I question the assumptions underlying the information provided by AI tools.

The other key survey items were scored in a similar way. Because I think it’s important to get a good sense of exactly what was asked, I’m going to list these items in their entirety as well. For AI tool usage, the questions were:

· How often do you engage in activities that require deep concentration and critical thinking outside of AI tools? (e.g., reading books, solving puzzles, engaging in debates)?

· How often do you use AI tools (e.g., virtual assistants, recommendation algorithms) to find information or solve problems?

· To what extent do you rely on AI tools for decision-making?

· I find AI tools help me save time when searching for information.

· I trust the recommendations provided by AI tools.

· I often cross-check information provided by AI tools with other sources.

Lastly, cognitive offloading was measured with these items:

· How often do you use search engines like Google to find information quickly?

· Compared to the past, do you feel that finding information has become faster and more convenient with technology?

· How often do you use your smartphone or other digital devices to remember tasks or information?

· When faced with a problem or question, how likely are you to search for the answer online rather than trying to figure it out yourself?

· On a scale of 1 to 6, how dependent are you on digital devices for day-to-day tasks and information retrieval?

Gerlich analyzed the resulting data with an eye to addressing two main research questions: how AI tools impact critical thinking skills, and whether cognitive offloading mediates that role (meaning in this context, whether offloading explains the relationship between AI and critical thinking). Several statistical approaches were used, including ANOVA and regression, and all pointed toward the same pattern of results: a strong, negative correlation, meaning that as AI use goes up, critical thinking scores go down. Cognitive offloading was a significant mediator as well.

This is all a fascinating set of findings, and certainly one that we in education ought to be aware of as we seek to understand how students are already using tools like ChatGPT, and as we seek to shape how they use it going forward. However, the implications are not nearly as straightforward as a quick pass through the study description might suggest.

Alongside these main findings, for example, was another set of interrelationships involving age. Younger respondents both depended on AI tools more, offloaded more, and scored lower on the critical thinking questions. None of these correlations are perhaps surprising in and of themselves, and in a study such as this one, it’s possible to tease out age to understand the AI-critical thinking relationship independent of the age variable. It did strike me, though, that casual observation in the world wouldn’t offer that same perspective – perhaps creating an outsized impression of young people with thinking abilities stunted by technology, when the actual reasons are a lot more complex than that.

This leads me to the part of the interpretation that raised the most questions in my mind, the parts having to do with cognitive offloading. The implied claim here seems to be that AI users shift critical thinking to these tools in a global, ongoing way, and like a muscle that isn’t used, critical thinking ability withers away. However, readers need to remember that this research never claimed to track this process happening in in real time. It’s not clear how these differences in critical thinking practices, as they are measured in this study, would play out in real world situations where critical thinking would be engaged. To be sure, generalizability to the real world is always an issue when a study involves surveys or contrived laboratory procedures. But the generalizability problem is a bigger-than-normal concern in this case given that according some experts, critical thinking is not a singular capacity but rather something that is fundamentally tied up with the context where it is taking place. Yes, test batteries like the ones used as a basis for the survey items probably do reflect meaningful differences in how people take in information from the world and use that to solve problems. But given this situated nature of critical thinking, it’s risky to assume that a person’s answers to hypothetical, abstract questions about thinking mean that they would approach an actual problem or decision in their field of expertise in a more- or less-effective way. In other words, respondents might not be offloading the exact cognitive processes that they would use in the real world, making it less clear what the actual impacts of the AI connection would be.

It’s also important to remember that when cognitive offloading was first floated as a concept, it wasn’t generally seen as a bad thing. To be clear, Gerlich does acknowledge early on in the article that it can be beneficial under certain circumstances. But that nuance could easily be eclipsed by the multiple other points where the phenomenon is characterized as a barrier to deep thought of all kinds. Especially because of this discrepancy in how offloading is talked about, I’m hoping that this article inspires readers to take another look at the now-classic Sparrow et al. article on search engines and cognitive offloading, work that is cited fairly heavily in Gerlich’s article. Notably, their research was less about the degradation of cognitive abilities across the board, and more about how deeply intertwined our memories and knowledge seem to be with digital sources. Studies 1 and 2 in fact mostly focus on the subconscious activation of computer-related ideas when we think about finding information and not on memory performance per se. Study 3 directly assesses offloading, but – and this is crucial – the key finding is that offloading hampers encoding of particular facts in and of themselves, not memory overall.

Granted, the pattern revealed in the Sparrow studies is still concerning, and (as I argued in my second book) these subtle effects could aggregate into larger and more noticeable impacts if students habitually search rather than attempting to remember factual material in a given field. However, it in no way supports the cognitive-atrophy trope, which is something that I tend to eschew anyway given the many ways in which minds and muscles are not alike. Nor does the Sparrow work get into critical thinking or problem solving at all, even though one could come away from Gerlich’s article assuming that it did. Lastly, Sparrow’s group measured cognitive offloading in an entirely different way than Gerlich did – through direct measurement of performance rather than through self-report - and did in the end characterize offloading more as a neutral or even positive phenomenon, stating in the concluding sections of the article that “[o]ne could argue that this is an adaptive use of memory—to include the computer and online search engines as an external memory system that can be accessed at will.”

And, speaking of critical thinking, we all need to remind ourselves of what we can and cannot conclude from a correlational study like the AI and critical thinking survey. It’s easy to skim through the various “interpret with caution” and “more research is needed” qualifiers typically proffered at the end of articles like this one, and still end up convinced that one variable caused changes in the other. The author of this study did not claim any such thing (although I’d quibble with the use of the term “impact” to describe the relationship given that the term seems to me to imply causality). There are plenty of alternative explanations and additional factors that could potentially link lower critical thinking scores to higher AI use, and I’m hoping that this study catalyzes more work that can either identify what those other factors are, or rule them out in favor of the direct cause-and-effect interpretation.

Among other things, I would be interested to know whether the inclination to engage in cognitive offloading could cause greater reliance on AI, rather than the other way around as proposed in the article, or whether there is some as-yet-unmeasured factor involving technology use or attitudes that could be a missing causal link. This particularly came to my mind given that trust in AI was queried by items in both the critical-thinking and the use-of-AI measures. I’m also now quite curious about whether certain ways of using AI are more predictive than others, in light of some interesting previous research by Nathaniel Barr and colleagues showing negative correlations between smartphone use and approaches to a problem requiring systematic, deliberative thought. Crucially, this work didn’t find that it was smartphone use across the board that mattered; levels of use for entertainment and media were far less predictive than using one’s phone specifically to find information online.

I’ll offer one last observation, which is that there seems to be a recurring dynamic where a finding about technology is initially interpreted positively, but is reinterpreted negatively later on for reasons that aren’t entirely clear. I think back to another classic, a 2009 article titled Your Brain On Google: Patterns of Cerebral Activation During Internet Searching. It reports on one of the first and only studies to look at changes to patterns of brain activity as a function of technology exposure, in this case, finding that learning to use search engines altered brain activity observed while searching among middle-aged and older study volunteers. In the original article, the authors interpreted these brain changes as essentially neutral or even positive (given that engaging with a search engine increased the observed activity in a number of brain regions). In later discussions of the work, though, these same authors recast the findings in a negative light, one I’d describe as “technology is doing creepy and unwanted things to human brains.” Could cognitive offloading be the latest example of this kind of abrupt reinterpretation?

Of course it's always fine for researchers to revise their conclusions when they have compelling reasons to do so, and maybe those authors did have good reasons. But it does seem sometimes like history is repeating itself when alarmist interpretations are elevated over other plausible ones. Perhaps we will see more researchers building on Gerlich’s framework, and maybe they will track down more proof that AI directly harms thinking, through the phenomenon of cognitive offloading. But I would argue that what we have now is a long, long way from a smoking gun.

The work I’ve discussed here is complex to say the least, and I suspect there’s a good chance that I’ve glossed over something or simply gotten it wrong. Again, this article is a valuable contribution on a timely subject, and I’m glad to see it in print – but it opens up space for a lot of misinterpretations, ones I’m already seeing flying around social media. So even more than with the other posts I make here, please let me know what you think, either in a direct message or by putting your thoughts in the comments below.

If you’d like to hear more about important issues relating to critical thinking in higher education, along with some suggestions for approaching AI in teaching, check out my interview with Tea for Teaching that will come out in March 2025. Also, if you liked having the audio version of this post, let me know and I might do more of them in the future!