Last week, my editor, Chelsea, said something that stopped me in my tracks. She was worried about the ubiquity of AI, but not for the normal journalistic reasons: job losses, plagiarism, dull prose, etc. It was the possibility that by using AI, she might be sacrificing one of life’s most reliable small pleasures – the daily joy she gets from having an “Aha!” moment. “For me,” she says, “it’s almost a physical feeling, something spreading across my brain.”

She wondered what might happen if we start outsourcing an increasing amount of our idea generation to AI before wrestling with it ourselves. Would we get fewer dopamine hits that come with figuring things out? And if those “Aha!” moments become rarer, what else might our brains be losing?

It turns out those “Aha!” moments are indeed giving us more than just small pleasures; there is growing evidence that they change our brain entirely, shaping what we learn and remember, and perhaps even play a role in protecting our long-term brain health. Luckily, as we head into an AI-driven world, there is something we can do to protect ourselves from losing out, aside from cancelling our ChatGPT subscription altogether.

Chelsea’s description of pleasure spreading through the brain at the moment of insight wasn’t far off. “Although it does feel like you get a jolt of dopamine, we can’t say that every insight produces a dopamine hit,” says Carola Salvi at the Catholic University of the Sacred Heart in Italy. However, several lines of research strongly suggest that the dopamine system is involved when you have those mini-epiphanies.

For instance, in 2018, Martin Tik at the Medical University of Vienna in Austria and his colleagues found that when people solved problems designed to elicit a eureka moment while undergoing functional magnetic resonance imaging (fMRI), their brain scans showed small changes in activity in midbrain structures involved in releasing dopamine. Tik told me at the time that neural activity in those areas was highest during “Aha!” moments, and scans showed significantly lower activity when people arrived at a solution without that feeling of eureka.

But “Aha!” moments don’t just feel good. There is increasing evidence that they also have cognitive benefits for learning and memory, says Salvi. She believes that they function as a kind of internal “selection signal”.

By this she means that when a solution suddenly becomes coherent and pops into our head, the accompanying feeling of accuracy and satisfaction helps capture our attention. The brain, perhaps with the help of the dopamine system, flags the idea as important. According to Salvi’s models, this helps us prioritise certain ideas for learning and future use.

This makes sense when you consider that ideas that come into awareness as an “Aha!” moment are also more likely to be correct. Of course, it’s not totally infallible – we’ve all been seduced by ideas that feel brilliant and turn out to be ludicrous – but generally, that sense of eureka appears to be a useful signal.

There’s also empirical evidence for this. Several studies show that sudden insight, and even “D’oh!” moments – cousins of the “Aha!” moment experienced after failing to arrive at a conclusion and having it revealed – improve memory for information presented around the same time. In other words, the pleasurable feeling that Chelsea describes generates a state of activity in the brain that helps store memories around the moment. We’ve also seen this in action from brain scans taken while people are having sudden insights that show they fundamentally change neural networks involved in memory and vision, and that the extent of those changes are associated with how easily people later remember the information they have learned.

“From an evolutionary perspective, this makes sense,” says Salvi. “If the brain suddenly discovers a useful new pattern or solution, it would be adaptive for that information to become especially memorable.” That “Aha!” moment may be a mechanism for tagging discoveries as worth learning.

Which brings us back to AI. If we increasingly turn to large language models (LLMs) for ideas and solutions to even our smallest problems, are we depriving ourselves of an opportunity to learn, remember or perhaps something greater still?

For this, I turned to Hannah Critchlow, a neuroscientist at the University of Cambridge and author of The 21st Century Brain: How to future-proof your mind in the age of AI.   

She pointed me to a small but fascinating study published last year that compared the neural activity of 18 people who wrote essays either using brain power alone, with the help of a search engine or with ChatGPT. Those who used AI showed consistently lower brain activity than those who used Google or brain power alone. Across four sessions over four months, participants who used ChatGPT to write their essays this way struggled to accurately quote their own work and consistently underperformed neurally, linguistically and behaviourally on the task.

With only 18 people who completed the study, we should be cautious of over-interpreting the results. But they raise the provocative possibility that while LLMs may seem like they give speedy insights, they could actually be harming longer-term learning and memory.

So, aside from deleting ChatGPT from our lives, how do we defend against this? Critchlow draws attention to a separate body of research showing that when people discuss ideas together, in a manner that’s not competitive, their brainwaves start to synchronise with each other.

This could be an important clue as to what human insights offer that AI cannot replicate. Making sure your brain has opportunities to develop this synchrony seems to be beneficial. Critchlow says that how well your brain synchronises with others can be used to predict how healthy your brain will be in later life. “It seems to be potentially protective against dementia and is one of the most significant predictive factors for whether a teenager is going to flourish during adolescent periods – whether they are going to be able to form bonds with others and learn from them,” says Critchlow.

In other words, the solution isn’t to use LLMs less, but to increase human connection. Critchlow thinks that, in light of this, schools, universities and other learning environments may become more collegiate, with renewed emphasis on teaching people in smaller, face-to-face groups. “Perhaps paradoxically, these new tools will help us to appreciate that fundamental to our species’ success is our ability to connect with others and to communicate with them. To learn from them and to allow ideas to hop from mind to mind, so that we can get that satisfying ‘Aha!’ moment, and so that we can problem-solve together and benefit as a species”.

So, for anyone who shares Chelsea’s concerns, there may be a simple lesson here. While it may be tempting to turn to LLMs for instant insights, exercising your own mental muscle to get to the answer yourself, whenever possible, may be the better path to take – not only for your own quick dopamine hit, but for long-term learning and brain health too.