The real question

With AI, are we getting dumber or smarter?

The most useful answer is: both outcomes are possible. AI removes friction. What matters is what you do with the attention it gives back.

If that extra bandwidth goes to verification, synthesis, better decisions, and deeper learning, you get sharper. If it goes to passive acceptance of plausible output, you slowly lose edge.

That is exactly the lens of Cognitive Load Theory (CLT).

A quick evidence-backed frame

CLT says working memory is limited, and mental effort can be split into three types:

  • Intrinsic load: the real difficulty of the task.
  • Extraneous load: avoidable friction from bad tools, poor process, noisy context.
  • Germane load: productive effort used to build understanding and mental models.

Classic references still hold: Sweller’s synthesis in Cognitive Science (2016), and Paas & van Merrienboer’s review in Current Directions in Psychological Science (2020).

In practical terms, AI usually reduces extraneous load first: drafting, summarizing, reformatting, boilerplate, first-pass exploration. That part is mostly good news.

The risk appears in how people react to fluent output. Lee et al. (CHI 2025, survey of 319 knowledge workers) report that higher confidence in GenAI is associated with lower self-reported critical thinking effort, while higher self-confidence is associated with more critical thinking (DOI).

So no, AI does not “make people dumb” by default. But it can normalize cognitive offloading without evaluation if teams do not build habits around review and accountability.

Where the load goes (illustrative)

This first chart is illustrative, not measured data. It shows a CLT-consistent pattern: the same task keeps similar intrinsic load, while behavior changes what happens to germane load.

Illustrative cognitive budget split for similar tasks under different AI-use behaviors.

Mental work map (interactive)

The scatter plot below is a discussion tool. X is sustained mental intensity; Y is overall difficulty. Positions are judgment calls, not a formal dataset.

The key pattern is simple: many modern roles keep high intrinsic difficulty even as tooling gets easier. In other words, less friction does not mean less responsibility.

How to stay sharp in day-to-day work

A practical way to use AI without losing your judgment:

  • Think before prompting: write 3-5 evaluation criteria first.
  • Treat outputs as drafts: verify claims, assumptions, and edge cases.
  • Do regular no-AI reps: keep core skills active (debugging, reasoning, writing).
  • Review in public: ask “what did we verify?” before approving.
  • Protect attention: reduce context switching and tool sprawl.

These habits keep germane load alive while still enjoying the productivity gains.

Bottom line

AI is not a brain upgrade or a brain killer by itself. It is a cognitive load reshaper.

Used well, it removes noise and gives you more space for judgment. Used poorly, it becomes fluent autopilot and weakens critical thinking over time.

The advantage goes to teams that deliberately convert saved effort into better thinking.