IQ Scores Are Falling. What That Means If You're Deploying AI
IQ scores climbed for most of the twentieth century, then started sliding in the 1980s. Norwegian data pins it on environment rather than genetics — and the leading suspect has direct consequences for anyone putting AI into how their organisation works.
For most of the twentieth century the trend line only went one way. Average IQ scores rose so reliably that test designers kept making the questions harder just to hold the average at 100. The pattern picked up a name — the Flynn effect, after the researcher James R. Flynn — and an easy explanation: better nutrition, better healthcare, more schooling, more demanding work. Each generation showed up a little sharper than the last.
Somewhere in the 1980s and 90s that stopped. Researchers spotted the turn first in Scandinavia, then in France, the UK and parts of the United States. In several of those places young adults now score below their parents.
What the Norwegian data actually shows
The most-cited evidence comes from Norway, a study covering more than 730,000 men. Scores rose for everyone born up to roughly the mid-1970s, then started sliding.
The part worth slowing down on isn't the decline. It's where the decline shows up. Brothers raised by the same parents in the same house scored differently, with the younger ones lower. Genetics can't explain a gap that opens up inside a single family. Whatever is driving this is environmental, and it moves fast enough to separate siblings.
Cognitive offloading is the leading suspect
The explanation researchers keep coming back to has a name: cognitive offloading. We hand our devices work our brains used to do alone.
When did you last pull a phone number out of memory, or work through arithmetic without reaching for a screen? There's nothing irrational about not doing either. The tools are faster and more accurate than we are. But unexercised skills weaken, and the ones going unexercised here happen to be exactly what IQ tests measure — working memory, sustained attention, grinding through a problem that won't yield to a quick answer.
Attention shows up as the other half of this. Digital life rewards fast scanning and constant switching between things. IQ tests reward staying put until the problem gives way. Verbal scores have slipped furthest, which lines up with a move from long-form reading toward headlines, captions and feeds. Pattern recognition has held up much better.
Schooling drifted the same direction — standardised testing, finding the answer and moving on, less time sitting with something hard. Then there are the unglamorous inputs nobody wants to talk about: sleep, diet, exercise, stress, pollution. None of them explains the trend alone. Stacked up, they plausibly do.
Why this matters if you are deploying AI
Every AI deployment is a cognitive offloading decision. That's the whole point of it. The analyst stops writing the first draft, the developer stops writing boilerplate, the support agent stops composing replies from scratch. The productivity case is real and I'm not arguing with it.
What the Norwegian siblings suggest is that offloading bills you later, and the bill lands on whatever skill you stopped practising. Route every draft, every summary and every judgement call through a model and the human capability underneath will soften — over years, not generations. That's an argument for choosing deliberately, not for going slower.
So choose. Let models draft, retrieve, summarise, translate. Keep people deciding what matters, catching the output that's confidently wrong, and holding the context the model was never given. Those skills erode quietly and cost a fortune to rebuild.
Two things that keep this honest
IQ measures a narrow band: pattern recognition, logic, processing speed. It's silent on creativity, judgement under ambiguity, emotional intelligence, collaboration, and how well someone picks up an unfamiliar tool. A fair number of researchers think intelligence is changing shape rather than shrinking, and that the tests simply haven't kept up.
The other thing is that almost everything on the list of causes is a choice. Reading, hard problems, sleep, actual conversation, learning something genuinely unfamiliar — all still there, all still free.
Which makes the interesting question less about whether we're getting less intelligent, and more about whether we still leave ourselves room to think slowly about things that deserve it. If you're putting AI into how your organisation works, you probably owe that one an answer.