There was a time that we were all being told that watching too much television would give us square eyes and turn our brains to jelly. I well remember being told to go outside and do something less boring instead. Then when computer games came along, we were all being told that playing games would damage us (although in this case it was often concern about the violence in some video games). Later concerns were raised about spending too much time online on the internet or on your phone! More recently people have told me that if I can’t remember something – try harder to remember – don't use Google – its making idiots of us all!
So, is the rising concern around the impact of AI reducing mental capacity in humans (Cognitive Atrophy a.k.a. AI Atrophy) just the next cause celebre of those perpetually worried doomsayers?
Let's take a look at what happens when we submit to over-reliance on AI and the impact on human learning.
Don’t just Google it ...remember it!
I well remember being at a dinner party, about 10 years ago, and failing to recall some detail relevant to a conversation we were having; I pulled out my phone to Google the fact in question. However, I was heavily berated by several other dinner guests who insisted that I work it out for myself rather than rely on Google. They claimed that such reliance was hindering us, leading to ‘digital amnesia’ or the ‘Google effect.’ They were concerned about the shift toward people allowing themselves to forget information, safe in the knowledge they could easily look it up some other time. 
In this case, I did not remember the fact I wanted until it popped into my head at 4:30am whilst lying in bed! So, from a personal point of view I would have appreciated Google’s help here. Of course, even before Google I didn't remember every item of information ever imparted to me ...oh how I wish that had been the case! My O Levels and A levels would have been so much easier! Instead, back then I might have had to look something up in an Encyclopaedia, reference book or the library. Which of course would have been much slower, less convenient and may or may not have helped me set me up for easy long-term retrieval.
Well, here’s the thing: apparently it might well have had a much better chance. Research has shown that when people expect to have easy, future access to information, they tend to have lower rates of retention and recall for that information. That is, instead of remembering the facts, our brain will instead remember the 'meta' information - the where and how to find it. In addition, the practice of actively attempting to recall information before looking up that information increases the memory’s ability and improves future retention.
AI Atrophy / Cognitive Atrophy
Now if googling facts reduces your ability to retain those facts, what about Generative AI tools such as AI chatbots that search, analyse and summarise information for you? At least in the ‘old’ days, people generally had to look up different online sources and come to a conclusion / summarise information for themselves. That is, they had to do some sort of thinking and research, even if it was an order of magnitude quicker and easier than the old manual approach of searching for a book in a library.
However, in the Gen AI world, what are we left with? The AI system does the searching; the AI system does the ‘analysis’ and generates the summary which is often what the human user ends up reading, and the result, as claimed by numerous studies, is that humans experience AI Atrophy or Cognitive Atrophy. This refers to the decline in intellectual, creative or practical skills resulting from an over-reliance on these tools. 
Not only is the AI tool being used to find information, it is handling tasks such as writing, coding, presentation development, image creation and decision making. The human then just takes and uses the result with little or no input or analysis. This weakens their own analytical, critical and creative abilities as well as reducing their ability to retain information and apparently reduces their ability to concentrate. This phenomenon leads to the erosion of basic skills in adults, the inability to come to their own conclusions and, perhaps more worryingly, can also affect the development of foundational skills in children.
AI Atrophy and Coding Skills
As someone with rather a lot of skin in the game, I can't help but wonder how this impacts the world of the programmer or coder in particular. The claims still appear to hold; excessive reliance on AI coding tools such as ChatGPT and Copilot appears to threaten a ‘skills atrophy’ in developers. That is, they start to lose the foundational coding skills, debugging skills, testing skills and general problem-solving abilities originally key to their roles.
In fact a study from Anthropic found that AI assistance impacts the formation (and retention) of coding skills. They created an experiment that examined
how quickly software developers picked up a new skill (in this case, a Python library) with and without AI assistance; and
whether using AI made them less likely to understand the code they’d just written.
The Anthropic research group found that “using AI assistance led to a statistically significant decrease in mastery. On a quiz that covered concepts they’d used just a few minutes before, participants in the AI group scored 17% lower than those who coded by hand, or the equivalent of nearly two letter grades. Using AI sped up the task slightly, but this didn’t reach the threshold of statistical significance.” 
Other work has also found that general problem-solving and critical thinking, key to many aspects of software systems development and application are also affected. For example, if the AI handles much of the ‘hard work’ in designing and structuring a system, exploring the required logic and behaviours, then users who did not flex these cognitive muscles had a reduced ability to understand the systems created – and also to solve future problems. 
Dealing with Generative AI systems
We can’t put the genie back in the bottle. Generative AI systems are here to stay, so what are the available options? Most experts agree that treating the Generative AI system as an assistant rather than as a replacement for the humans’ own abilities is preferable. Use the AI system as one would a colleague, that is, as a helper to you work with the understand a new library, solve a complex problem or work on an application together. One could argue that the pair programming approach might be the best analogy, where neither programmer is subordinate to the other and both analyse each other’s work.
Through this, humans can continue to develop their skills so that they do not become reliant on the support of AI to the exclusion of their own abilities.
Interestingly, one of Anthropic conclusions regarding the use of AI assistants is that the information they imparted was less well retained than those who did not use the Ai assistant. That does not mean that nothing was learnt, but it was nonetheless diminished – akin to that "Google vs memory” issue where you learn where you got the information rather than what that information was.
In my opinion, there is still nothing to replace the great instructor-led training that has helped nurture so many developers over the last 40 years or more. To really ensure that information is imparted in the correct way and will be retained to the greatest extent, I have no doubt that in-person, human instructor-led training wins every time. That does not mean to say that AI systems cannot assist in this or that they can’t be an extension to the core training provided, but they shouldn’t be a replacement for it.
My advice?
Beware the AI-driven digital dementia that shows all the signs of becoming yet another mainstream modern malaise, alongside the likes of doomscrolling, FOMO and popcorn brain.
Cognitive Atrophy is real and it is already affecting many people. Keep your skills sharp, keep your abilities fresh and use the tools available to you to your advantage and not as your replacements.