AI Offloading: Legitimate Threat To Our Brains Or Overblown Moral Panic? We Asked A Neuroscientist For His Take

Like an athlete’s muscles, our brains develop through training, and having such a mental crutch could be seen as undermining that development. 

And what about school kids? They’re now appearing in classrooms as an everyday part of their education.

Are we not ushering in a future where people will be dependent on a thing to do tasks and processes humans have been doing on their own for centuries? It feels like another way to undermine children’s basic learning skills. It’s easy to see why some schools are implementing strict bans.

These sentiments are a rephrasing of genuine concerns that were raised in a New York Times article. And while you may think I was talking about the sudden ubiquitous appearance of artificial intelligence- (AI-) enabled programs in every aspect of our lives, I am actually talking about the humble calculator. 

Back in 1975, not long after the handy devices hit the market, some people raised concerns that are remarkably similar to those we are hearing today.

Over the last few years, large language models (LLMs) like ChatGPT, Gemini, Claude, and others have become the focus of much attention. For better or worse, we’re told, these technologies are here to stay. 

And they’re everywhere: every industry has been infiltrated by these technologies, from financial services to the health sector, and yes, even education is wrestling with the best ways to utilize these tools.

The rapid transition of AI from a specialized technology to an intimate everyday presence has brought with it significant debate, as we try to make sense of what it means for our lives. One point in particular keeps coming up: is it okay for us to offload our thinking to these programs?

Although researchers are still tackling this question, there is already considerable concern about “cognitive offloading” – the use of external aids to help alleviate mental effort to focus attention elsewhere. 

On the one hand, cognitive offloading is a well-known method we use to help delegate cognitive tasks to external resources to help reduce our mental demand. Think about making a list to help you remember what to buy at the shops. Or perhaps setting a date in your calendar so you don’t miss an appointment. Both of these are examples of cognitive offloading that few of us would regard with suspicion.

I'm sure you'd see different brain activity if people wrote down their shopping list compared with if they remember it themselves. But that doesn't mean that it's terrifying that people are using their phones to remember their shopping lists.

Dr Sam Gilbert

But the arrival of AI represents something different and altogether less known. In the last few years, there's been a worry that by relying on AI for increasingly diverse cognitive tasks, we are threatening our critical thinking skills and atrophying our overall cognitive ability.

As the above deliberately tongue-in-cheek concern about calculators demonstrates, this is not the first time such worries have accompanied the widescale use of new technology. 

Now this isn’t to say that AI is harmless – the research is ongoing – or that its gradual creep into more aspects of our lives is not problematic (it is). However, it does ask us to take a closer look at the situation.

Is AI doing something new?

Before weighing up the extent to which we should be concerned with cognitive offloading, it’s probably a good idea to examine where these concerns come from.

In June 2025, an influential article posted as a preprint by MIT’s Nataliya Kosmyna and colleagues explored the neural and behavioral costs of LLM-assisted essay writing for students. The study separated participants into groups – those who used LLMs, those who just used search engines, and those who relied on their “brains” alone.

Throughout these sessions, the researchers used electroencephalography (EEG) – a test that records electrical activity in the brain – to assess the cognitive load occurring as participants produced their essays. 

The results revealed distinct differences. 

Those who produced essays without additional assistance had stronger electrical activity in their brains, which was also more distributed. 

In particular, they showed greater activity in regions responsible for creative thinking, executive self-monitoring – mental management of focus, working memory, impulse control – and semantic integration, which is the ability to combine and interpret separate information from different sources.

I think it's very important to take a historical view here. And because these worries about technology are so ancient, that doesn't mean today’s worries are misguided. 

Dr Sam Gilbert

Those who used search engines showed moderate connectivity, and, as you may have anticipated, those who relied on LLMs had the weakest outcomes.

When participants switched method in session four, those who swapped from the LLMs to using their own brains showed lower alpha and beta connectivity, suggesting under-engagement. In contrast, those in the brain-only group who started using LLMs had higher memory recall and electrical activity in the areas of the brain associated with visual processing and decision-making, attention, and planning.

When it came to behavior, the results also revealed stark differences. In particular, those who relied on LLMs found it much harder to quote their own work, and those who did recall content quoted it incorrectly. They also felt less ownership of their essays compared to the other participants.

This study generated significant attention when it was first published. It joined a body of work suggesting that cognitive offloading to AI has significant downsides. Outlets like the BBC compared the advent of LLMs with the “the Google Effect”, whereby people appeared to remember fewer details as they became more reliant on search engines. 

The difference here, however, was the scale of the offloading.

And as with all similar societal concerns, there were fears that young people are the most vulnerable to this offloading cost – in particular, the concern that AI technologies are already damaging younger user’s skills while making them more dependent on them.

Education and everyday critical skills are not the only areas of concern. Other research has raised questions about AI’s role within mental wellbeing. 

In this context, cognitive offloading is regarded as a potential positive: it offers people a way to regulate stress while democratizing access to potentially useful ideas and offering new ways to cope. 

But there is also a risk here of “cognitive overloading”. This is where users limit their personal introspection, becoming reliant on algorithmic feedback, which effectively outsources their resilience. 

There is also a danger that users become more anxious because AI devices, especially those that'selated to biometric scores, may make people fixate on optimization, making them feel like they underachieving.

Clearly AI is reshaping not only how we live and think but also how we approach the world, our work, and our health. But are the concerns grounded and reasonable?

Fears as old as time?

Although the term itself is new, the fears underpinning cognitive offloading are not. Plato, the ancient Greek philosopher, worried that writing – a form of cognitive offloading – would diminish people’s memories.

“People are talking a lot about AI at the moment, but cognitive offloading goes back tens of thousands of years, if not longer. The earliest example I can think of that’s believed to relate to cognitive offloading is about 42,000 years old,” Dr Sam Gilbert, Senior Research Fellow in Metacognition and Executive Function at UCL, London, told IFLScience.

During the 1970s, archaeologists excavating a cave in the Lebombo Mountains on the border of South Africa and Eswatini found a notched bone. The Lebombo bone, as it’s called, is believed to be the oldest known counting tool ever discovered. Which, as Gilbert explained, may make it the “earliest known artifact” that was “likely to have been a cognitive tool”.

“According to some people, language itself is a form of cognitive technology. Language is invented. It’s communicated socially and it simplifies certain mental computations.”

Cognitive offloading is, clearly, something humans have done for a long time, and concerns for its costs are potentially just as old.

There is a belief some people have that technology is likely to change the skills we have but not our core abilities. And I think that’s probably true.

Dr Sam Gilbert

“I think it's very important to take a historical view here. And because these worries about technology are so ancient, that doesn't mean today’s worries are misguided. It just means you need to provide a reason why what's happening now is different to what's happened all those other times that seem quaint now.”

It is possible that AI is actually attracting multiple related societal concerns – like an anxiety-filled babushka doll. Accompanying the concerns over cognitive offloading is the historically recurring view that we are living in unprecedentedly chaotic and busy times.

“There's a somewhat related point. It's not quite the same thing but not only have people worried about losing our mental abilities through technology for at least 2,000 years, but this sense that modern life is uniquely fast-paced and we have information overload is also a feeling that goes back centuries.”

Following the invention of the printing press in the 15th century, intellectuals worried that the quick appearance of thousands of new printed books would not only flood the market but also overwhelm readers. In short, there were suddenly far more books competing for people’s attention, more than a person could possibly master in a single lifetime. 

Similar concerns appeared with the advent of daily newspapers and the telegram. Even the emergence of railway travel generated related fears as people worried that faster travel would unhinge the mind.

We may see these historical examples as trivial now, but they nevertheless mirror the concerns currently manifesting in academic and public discourse.

“Every generation thinks there’s an information overload that didn’t exist previously,” Gilbert added, “but I think it’s probably not that different.”

Recurring intergenerational concerns may also be nesting among these other issues. This is the belief that the young are either weaker or too lazy compared to previous generations. Sociologists have shown that older generations have been criticizing younger ones for these traits for millennia.  

“[T]hat's a very old feeling that you can trace back for many thousands of years as well. So this intergenerational aspect, I think, is important,” Gilbert explained.

Balance in all things

What does this mean for AI offloading? None of the above suggests there isn’t a problem associated with overreliance on these technologies. 

However, it does demonstrate that, as we continue to wrestle with their ever-expanding place in our lives, we should be cognizant of the types of beliefs we rely on in the process.

Indeed, past research on cognitive offloading more generally – not in relation to AI – has demonstrated some cognitive costs.

“[I]t's true that for many tasks,” Gilbert added, “when we use a tool as part of our thinking process, we can do that task better than if we just do it with our bare brain. But there are some interesting phenomena relating to the consequences of doing that”.

One of these is the so-called Google Effect mentioned above. The effect, which is also known as digital amnesia, involves people forgetting specific facts they’ve seen using search engines. However, this is actually balanced by something called the “saving-enhanced memory effect” where the offloading of information actually frees space in our brains so we can remember additional information and enhances learning.

“So it's a balance. You're not losing the ability to remember necessarily, but you're reprioritizing what you put into your brain.”

For Gilbert, cognitive offloading is not regarded as a problem per se. For instance, machines are far better than humans at remembering information, so we can reasonably rely on them for our shopping lists, calendar events and telephone numbers. But there are still some areas of human activity that should remain ours.

“There are still many activities that humans are better at than machines: forms of creative thinking, forms of social intelligence,” and so on.

Of course, AI technologies are becoming more general purpose, so “there is a temptation to use them across the board for many different activities,” he says. 

“I think we should pay attention to the things tech is good at and what it’s bad at. And I think its intuitive to many people that something like creative thinking, well, you don’t need an answer from me on that.”

This is why it is important to fully understand the assumptions and cultural beliefs we may hold, possibly even unbeknownst to ourselves.

“There is a belief some people have that technology is likely to change the skills we have but not our core abilities. And I think that’s probably true. I don’t see any convincing evidence that we are losing our ability to pay attention or our ability to remember content. But technology certainly changes our incentive to pay attention [or to remember] certain things.”

Technology changes us based on how we use it. 

Dr Sam Gilbert

It is the question of current evidence that really crystallizes Gilbert’s point.

“[W]hat does the current evidence say? I don't think it says very much about the questions that we would most like the answers to.”

For instance, Gilbert explains that the now famous MIT experiment that is often cited as the more concerning example of AI causing issues, is less concerning when you look at its methods.

“What they did in this experiment was to get volunteers to come along – volunteers who got paid regardless of how they performed in the experiment – and asked them to write essays about topics for which they probably had no intrinsic interest.”

Should it be surprising then that the participants who used AI to help them were less likely to remember what was written? Gilbert doesn’t think so.

“I don't think that sounds problematic at all. I think, in fact, it would be masochistic of those participants to behave otherwise, because if you're told to come to an experiment, you're going to get your $10 whatever happens, you don't care about what you're asked to do, and you can get ChatGPT to do it for you, then great. It's a problem if using ChatGPT interferes with the goals that people actually have, but that wasn't demonstrated in the study”.

Gilbert also stresses that you can’t place a moral or value judgement on brain activity.

“It is what it is. And so just because the pattern of brain activity was different doesn't mean it's better or worse.”

“I'm sure you'd see different brain activity if people wrote down their shopping list compared with if they remember it themselves. But that doesn't mean that it's terrifying that people are using their phones to remember their shopping lists.”

To be worthy of the kind of societal concerns currently being articulated in the public sphere, researchers would need to test for other things, Gilbert says. 

For instance, researchers would need to ensure that people have a genuine goal of doing something that can either be done by a person or by an LLM. They would then need to understand how using AI can change someone’s ability to perform on a subsequent task that is not the same as the one they performed earlier. 

Without this “transfer effect”, Gilbert says, “you don’t know anything about how [AI] is harming your abilities.

At the start of July 2026, a new study was published in Trends in Cognitive Science showing that relying on AI doesn’t necessarily impact our ability to perform mental tasks. Although it is possible that overreliance on these tools for long periods of time – across decades – or their introduction at an early stage in a person’s development could have negative effects, they were not inevitable.

The authors conclude that it is still too early to draw meaningful conclusions about the long-term effects of LLMs, but they stress that it ultimately comes down to how we use them. If used as collaborative tools, the study suggests, they can enhance learning.

None of this is to say that we should not engage in critical conversations about AI or its impacts on our lives. But there simply isn’t enough evidence about offloading at this point. 

However, we can still ask questions about the extent to which we want to rely on these technologies and what they mean for the type of society we wish to live in. This is also a very important question.

“There's this great quote, [what is known as Kranzberg’s First Law], saying that technology is neither good, nor is it bad, nor is it neutral,” Gilbert said. “Technology changes us based on how we use it. It’s clearly a sense that depending on the technology we have available to us, it encourages us into certain forms of thinking.”

So how do we want this technology – as well as those who produce it – to influence our world, both mental and actual?