A bigger problem than the rise of artificial intelligence is the decline of natural intelligence.

The real risk is cognitive offloading

We have always used tools to think, from notebooks to calculators. The difference with today's AI systems is scope and depth. When a system drafts our emails, plans our study path, or proposes an argument, we skip the slow work that builds understanding. Skipping once is harmless. Skipping by default changes how we think.

What the data already shows

International learning data shows broad declines in core skills after the pandemic, with evidence of longer-standing downward trends in some places. PISA 2022 reported unprecedented drops across the OECD in mathematics and reading, and multiple analyses argue the fall is only partly due to school closures. At the individual level, two patterns stand out. The Google effect shows that when information is easy to retrieve, people remember the location, not the content. And the mere presence of a smartphone can reduce baseline attentional performance, hinting at persistent cognitive costs in always-on environments.

How tools change what the brain practices

When we rely on GPS, our spatial memory atrophies. When we rely on autocomplete and summarization, our writing and comprehension muscles do less work. Studies of media multitasking and screen exposure show mixed but consistent risks for attention and executive function, especially with passive or excessive use. Early childhood evidence ties background television and age-inappropriate content to poorer cognitive and psychosocial outcomes, while co-use and purposeful engagement tend to help.

AI, immediate gains, and weaker retention

AI and search can boost short-term performance on lower-order tasks, but the advantage often fades at delayed tests. Work from Stanford's SCALE initiative finds that groups using ChatGPT or Google scored higher immediately, then fell back on retention, while control groups retained more on higher-order tasks. Meanwhile, emerging studies on AI-assisted writing report reduced engagement and weaker recall among heavy users, which researchers describe as cognitive debt. The risk is not that AI makes us unable to learn. It is that defaulting to help reduces the very practice that stores knowledge.

Where offloading helps, and where it harms

Offloading is useful when it frees capacity for deeper reasoning, for example, using AI to surface contrasting sources after you write a first pass, or to check logic after you build it. Offloading is harmful when it replaces the primary act of learning, for example, asking for a summary before you read, or accepting explanations without generating your own.

A simple rule for resilient intelligence

Do the hard part first, then use AI to improve the result. Write before you edit. Reason before you verify. Plan before you optimize. You keep the learning benefits, and you still gain the speed. Write the first draft yourself, then ask AI to critique clarity, examples, and structure. Read the source, then ask AI for counterarguments you may have missed. Solve a problem on paper, then compare with an AI solution and reconcile the differences. Keep phone-free zones for deep work. Batch your lookups so your working memory stays engaged rather than constantly interrupted.

The point

Intelligence is not a score. It is a habit of attention and effort. Tools that remove all friction feel like magic, but a life with no friction gives your mind nothing to push against. Use AI as a lever for better thinking, not as a substitute for it.

References