For most of the 20th century, human cognition behaved like a system on a reliable upgrade cycle. Every generation shipped a slightly better build than the last. Researchers called it the Flynn Effect, and for nearly 80 years it held: IQ scores climbed by roughly three points a decade, a slow but steady version bump traced back to better nutrition, longer schooling, smaller families, and increasingly complex environments.
Then, sometime in the mid-2000s, the curve bent the other way.
A study led by Northwestern University social scientist Elizabeth Dworak, alongside co-authors William Revelle and David Condon, pulled cognitive test data from nearly 400,000 American adults collected between 2006 and 2018, published in the journal Intelligence. The dataset came from the Synthetic Aperture Personality Assessment (SAPA) Project, an open online research platform, and the tasks came from the International Cognitive Ability Resource (ICAR) — an open-source intelligence battery covering matrix reasoning, logical sequences, verbal reasoning, and spatial rotation.
The result reads less like a single downward trend and more like a system quietly reallocating its processing power.
The Data Doesn’t Lie, But It Also Doesn’t Explain
Composite scores dropped across the 13-year window, and the decline wasn’t confined to one demographic. It showed up across age brackets, education levels, and genders. The steepest drops hit the youngest adults — 18- to 22-year-olds — and people with less formal education, a pattern that should worry anyone tracking the pipeline of future technical talent.
Dworak herself didn’t expect the result. She told New York magazine she spent weeks re-checking her code, convinced she’d made an error, until her adviser confirmed the math held up.
But treat this like a system diagnostic, not a verdict, and the picture gets more interesting. Break the composite score into its component subsystems, and the decline isn’t uniform at all:
- Matrix reasoning and logical sequencing — the tasks closest to raw abstract problem-solving — took the biggest hits.
- Verbal reasoning barely moved. The drop was too small to register as statistically meaningful.
- Spatial rotation — the ability to mentally manipulate 3D objects — actually improved over the same period, and the gain showed up across most of the adult age range.
| Cognitive Subsystem | Measured Trend (2006–2018) | Hypothesized Driver |
|---|---|---|
| Matrix Reasoning | Steep decline | Reduced engagement with unstructured abstract logic |
| Letter & Number Series | Steep decline | Shift from long-form structured text to bite-sized media |
| Verbal Reasoning | Slight decline, below the study’s significance threshold | Daily linguistic exposure holding relatively steady |
| Spatial Rotation | Measurable increase | Possibly video games and 3D interfaces — a proposed explanation, not a confirmed one |
That last row is the one worth sitting with. Researchers floated video games as a possible driver — the study didn’t test for it directly — which would make this less a story of cognitive decline and more a story of cognitive specialization: one mental skill traded for another, shaped by which tasks a decade of screens and interfaces actually rewarded.
The Suspects Nobody Can Fully Convict
The obvious modern villains don’t hold up under a timeline check. The decline predates the pandemic, so lockdowns and COVID-related disruption are off the hook. It also predates mainstream generative AI by more than a decade, which rules out ChatGPT-as-cognitive-crutch as the origin story, whatever the headlines eventually claim.
What lines up chronologically is the rise of smartphone adoption in the late 2000s — the same stretch when scores started sliding. Standardized tests tell a parallel story: both the ACT and SAT have posted steady score declines over the past decade, independent of this dataset.
Times of London columnist James Marriott has pushed a specific version of this theory: that reading on a phone trains a fundamentally different — and weaker — kind of thinking than reading in print. Print, he argues, forces an argument to cohere logically from start to finish in a way that feed-based, clip-based content never has to. It’s a compelling narrative. It’s also a hypothesis, not a finding — and Dworak’s study, built from repeated cross-sectional snapshots rather than tracking the same people over time, can identify that scores shifted without proving why.
The Real Story Isn’t Decline. It’s Reallocation.
Framing this purely as humanity “getting dumber” misses what the subject-level data is actually showing. A population that’s uniformly losing intelligence should see every measured skill fall together. This one didn’t. Abstract and sequential reasoning weakened. Verbal ability held steady. Spatial cognition strengthened. That’s not decay — that’s a tradeoff, the kind you’d expect from a species whose daily cognitive workload has been restructured by the tools it now spends most of its waking hours touching.
It’s a pattern with an obvious analog in tech: optimize hard for one kind of task, and performance on a different, less-exercised task tends to drift. Whether the underlying cause is smartphones, changing educational priorities, screen-based attention patterns, or something not yet on anyone’s radar remains genuinely unresolved. What the data does confirm is that human cognition isn’t static, and its trajectory isn’t guaranteed to run in one direction anymore — up, down, or otherwise.
The bigger question the study leaves open: as AI tools increasingly absorb the exact tasks where scores are falling hardest — abstract pattern recognition, logical sequencing — does that trend accelerate, stabilize, or reverse? The same verification-bottleneck logic that decides which tasks are safe to hand to AI may end up applying to entire categories of human reasoning, not just individual work tasks. Dworak’s data stops in 2018, right before the tools most likely to reshape the next chapter of this story became part of daily life. The next version of this study is the one that actually matters.
Related: Think For Yourself In the Age Of AI (Survival Guide)
