For most of the twentieth century, each new generation got better at IQ tests. The increase appeared across countries and across kinds of cognitive problems. James Flynn made the pattern famous, and it became known as the Flynn effect: roughly speaking, the test kept getting easier for the people born later.
Then the line bent. In Norway, Denmark, Finland, and several other wealthy countries, gains slowed, stopped, or went into reverse. A large Norwegian study found both the rise and the decline within families, between brothers born in different years. That is important. It points away from a story about genes changing across the population and toward a story about environments changing around it.
What the Flynn effect actually measured
The Flynn effect never meant that a baby born in 1990 arrived with a better brain than one born in 1930. The gains happened too quickly for that. Better nutrition, less disease, longer schooling, smaller families, and a world that demanded more abstract reasoning all plausibly helped. Modern life trained people for the kinds of abstraction modern intelligence tests reward.
The reversal should be read with the same care. It does not prove that an entire generation is simply “dumber.” IQ tests sample particular abilities, and trends differ by country, test, and subtest. But a sustained drop on the same measures is still a signal. If environments can raise cognitive performance, environments can also stop raising it.
We built an environment that exercised abstraction. Now we are building one that exercises interruption.
The timeline rules out the easy villain
Smartphones did not cause the original reversal. Norwegian scores peaked among men born in the mid-1970s, long before the infinite scroll. Social media cannot explain a decline that began before social media existed. Generative AI arrived later still. Anyone claiming the causal case is closed is skipping the dates.
But the dates do not make our current technology irrelevant. They change the question. Social media and AI may not explain why the Flynn effect first stalled; they may help determine whether it returns.
Recent longitudinal research on adolescents has linked some rising patterns of social-media use with lower performance in reading, vocabulary, memory, and overall cognitive scores. The findings are associations, not a verdict that every hour online removes an IQ point. Screen use is not one behavior, and children do not use screens in identical lives. Still, the mechanism is not difficult to imagine: less sustained reading, more task switching, worse sleep, and fewer long stretches in which a hard idea has to be held in mind.
AI changes the bargain again
Social media competes for attention. Generative AI can replace the cognitive act itself. It can summarize the chapter, choose the structure, write the paragraph, solve the equation, and explain the answer before the learner has produced one. That is useful when the goal is output. It is dangerous when the output was supposed to be evidence of practice.
We do not yet have the decades of data required to say what generative AI will do to generational intelligence. Early studies of AI use and cognitive offloading are suggestive, not decisive. But waiting for a cohort to reach adulthood before asking whether students still read, retrieve, calculate, and write unaided would be a strange standard of caution.
The useful distinction is not technology versus no technology. Calculators, search engines, and AI can extend what a practiced mind can do. The question is whether the extension comes before or after the mind has learned the move. A tool used after an attempt can diagnose, challenge, and accelerate learning. The same tool used instead of an attempt can make intellectual work disappear behind a polished answer.
A trend is not a destiny
The hopeful lesson of the Flynn effect is that cognitive performance responds to the world we build. The unsettling lesson is exactly the same.
We can design schools and tools that restore sustained attention, require retrieval, reward complete reasoning, and make students wrestle with an idea before a machine resolves it. Or we can optimize every surface for speed and engagement, then act surprised when patience and independent thought become harder to measure.
Lune Synth™ is our bet on the first path: do the work by hand, make the reasoning visible, receive specific feedback, and practice the part that failed. AI belongs in that loop—but after it has something human to respond to. The Flynn effect reminds us that environments train minds at population scale. We should be very deliberate about what today's environment is training.
Further reading: the Norwegian within-family study, the Danish cohort analysis, and a longitudinal study of adolescent social-media use and cognition.
Want AI that responds to thinking instead of replacing it? Join the beta waitlist on the Lune Synth home page, or write to griffin@lunesynth.com.