Generational Risk: AI Is Making Kids 55% Dumber

The headline number is alarming, and it deserves precision before anything else. According to research presented on the Prof G Markets podcast by host Ed Elson and research lead Mia Silverio, one study revealed students using AI tools for homework “experience a 55% reduction in overall brain activity.” That is a measurement of cognitive engagement during AI-assisted work, not a drop in IQ or permanent intelligence loss. The distinction matters, but the implication for how a generation learns is exactly what investors, parents and policymakers should be wrestling with.

The Generational Break

Elson’s segment leaned on congressional testimony from neuroscientist Dr. Jared Horvath, who told lawmakers that “every generation in history has been smarter than their parents’ generation except for Gen Z.” The supporting data is sobering. Roughly half of American 12th graders now score below basic math levels, and a third of American 12th graders lack basic reading skills. Horvath called it “the worst rating that we’ve seen among that group in 3 decades.”

The trend is global. Science, math, and reading scores among teenagers in high-income countries have fallen nearly 5% over two decades, and adult literacy rates have fallen roughly 3%. AI arrived on top of an existing decline.

The Brain on ChatGPT

The Prof G Markets framing of the cognitive cost is striking. When students use ChatGPT, “your brain is actually more impaired, more suppressed than if you were to be twice over the legal alcohol limit.” The follow-up finding is what should give educators and product designers pause: “the impairment effects of using AI appear to compound even after the AI tool is removed from usage.” Alcohol wears off. Cognitive offloading, according to this research, leaves a residue.

You can listen to the segment via the Prof G Markets podcast feed for the full discussion.

The Investor Takeaway

AI remains a powerful investment theme. The productivity gains for adult knowledge workers, software engineers, and enterprises are real and showing up in margin expansion across the hyperscaler complex. The Elson and Silverio thesis is narrower and more useful: the same tool that compounds output for an experienced professional may be compounding deficits for a developing brain.

That has consequences worth tracking. Education technology companies pitching AI tutors, large language model providers selling into K-12 districts, and consumer subscription services targeting students all face a research environment that is starting to quantify a downside. Regulation tends to follow congressional testimony like Horvath’s by a few years.

For households, the actionable point is simpler. How your kids use AI for homework this school year is a parenting decision now backed by neuroscience rather than a pure productivity question. Watch the policy response, and watch which education-exposed names get pulled into the conversation next.