What happens when you stop practicing the thinking you've handed off to a chatbot?
That's the question at the center of a BBC Future investigation published April 17, which examines research suggesting that habitual reliance on AI assistants may gradually weaken independent reasoning, memory, and problem-solving skills - the cognitive equivalent of losing muscle mass when you stop exercising.
The concern isn't entirely new. Psychologists have studied "cognitive offloading" - using external tools like notebooks, calculators, and GPS to handle mental tasks - for decades. The worry with AI is the scope. Unlike a calculator, which handles arithmetic you probably don't need to practice, chatbots like ChatGPT and Claude handle open-ended reasoning, writing, and analysis - skills that arguably matter more, and that you can lose.
When Fluency Becomes a Crutch
The mechanism researchers point to isn't complicated: skills atrophy when you don't use them. If you let an AI draft every email, summarize every document, and answer every research question, you're not practicing those skills. Over time, the gap between what you can do with AI assistance and what you can do without it widens.
This plays out differently depending on how you use the tools. A developer using AI to autocomplete boilerplate is offloading low-value work. A student using a chatbot to reason through a problem they haven't attempted yet is skipping the cognitive work that builds understanding. Same tool, different outcomes.
There's also the confidence problem. AI responses sound authoritative whether they're right or wrong. If you accept AI-generated answers without scrutinizing them, you gradually stop building the internal knowledge base that would let you catch errors. You become dependent on a system that can confidently give you incorrect information.
Does It Actually Affect You?
For daily AI users, the honest answer is probably: it depends on how you use them.
Using AI to accelerate work you already understand? Likely fine. The scaffolding isn't replacing the skill; it's speeding it up. Using AI to skip learning something in the first place? That's where researchers see real capability loss.
The GPS analogy holds up under scrutiny. Studies have shown that regular GPS use correlates with measurably reduced spatial navigation ability - particularly in people who previously relied on mental maps. The skill doesn't vanish overnight, but it weakens with sustained disuse. AI assistants are GPS for a much broader range of cognitive tasks.
None of this is an argument for stopping. But it is a reasonable case for occasionally doing things the slower way - drafting without assistance, thinking through a problem before querying a chatbot, deciding before asking for a recommendation. Not as a ritual, but as maintenance.