Around the year 2000, the two-century rise in Western IQ scores began to stall; Horvath, a former schoolteacher, argues in his book The Digital Delusion that the inflection coincides with the rollout of educational technology.
The Claim
Horvath has spent over 15 years studying how people learn. He traces the post-2000 stall to screen-heavy classrooms: in the US, edtech is a roughly $400 billion industry; in the UK, around £6.5 billion annually. Roughly 88% of US school districts now operate one-to-one laptop programs. School rollouts accelerated between 2010 and 2012, then went near-universal during COVID-19 remote learning. Horvath notes that post-pandemic, most schools never reverted.
Citing PISA waves from 2012, 2015, and 2018, he observes that "the more time students spent on screens at school, the further their scores fell," and concludes that screen-based tools underperform for the kind of deep, transferable knowledge he studies. He frames the moment as a generational cognitive setback rather than a methodological dispute.
The Counter-Signal
An eight-year Finnish cohort published this week complicates the picture. Researchers at the University of Jyväskylä tracked children into adolescence and found that greater screen time correlated with better cognitive processing during the teenage years — the opposite direction of Horvath's claim.
Doctoral researcher Petri Jalanko cautions against treating screen time as monolithically harmful. The decisive variable, per the Finnish data, is content: devices used for problem-solving, creativity, and active learning show a positive association with cognitive processing; passive consumption does not. The team also reported that light-intensity physical activity in girls and guided exercise in boys correlated with stronger working memory — reinforcing the case that engagement quality, not exposure per se, is what drives the variance.
Reading the Evidence
UCL Knowledge Lab's Rose Luckin, an emerita professor and recognised expert on AI in education, notes that "a great deal of educational technology was sold on promises rather than proof" and warns against confusing correlation with causation. That same gap — marketed outcomes versus measured performance — is where consequential misallocations occur, and the same evidentiary discipline applies to any rigorous evaluation of platform performance.
Three filters follow:
- Demand outcome data, not engagement metrics. Login counts and screen-time minutes are not cognition. PISA-style achievement deltas and longitudinal processing measures are.
- Stratify by content type. Aggregate screen-time bans discard the active-learning half of the variance.
- Triangulate before scaling. Single-source claims — pro- or anti-screen — should not drive policy in a $400 billion market.
The empirical question is not whether screens stay in classrooms — they will — but which configurations of device, content, and task produce measurable cognitive transfer. Horvath measures exposure; the Jyväskylä group measures activity type. For educators, parents, and clinical readers, the durable takeaway is that dose-response claims about screens require the same evidentiary bar as any neurocognitive intervention — mechanism, replication, matched cohorts.