Two studies, two conclusions that look like mirror images of each other. A new PLOS ONE study of more than 1,000 people found that people who play more video games test cognitively better — independent of their mental health. Meanwhile, a growing, separate body of research on AI use keeps finding the opposite pattern: people who rely heavily on AI to do their thinking for them test worse on measures of the skills they've outsourced. Same underlying mechanism, opposite behavior, and the contrast is worth taking seriously rather than treating as a coincidence.
TL;DR: what both bodies of research actually show
| Finding | Gaming study | AI-reliance research |
|---|---|---|
| What's measured | Cognitive test performance (memory, verbal, reasoning) | Skill retention, comprehension, code understanding |
| Direction of effect | More engagement → better performance | More passive reliance → worse performance |
| Sample/scale | 1,000+ people, cross-sectional online study | Multiple independent studies across coding, homework, and professional skill domains |
| Mechanism proposed | Active, effortful cognitive engagement | Cognitive offloading — skipping the effortful step entirely |
| Is it causal? | Partially — cross-sectional here, but causal evidence exists elsewhere | Partially — same caveat, but the pattern is consistent across independent studies |
What the new study actually measured
The study, led by Sydni Paleczny, Conor Wild, Alex Xue, Roger Highfield, and Adrian Owen, set out to answer a genuinely open methodological question: exercise and video gaming are both widely promoted as ways to protect or enhance brain function, but it's been unclear which specific aspects of brain health each one actually affects. The researchers recruited over 1,000 participants globally for a cross-sectional online study, collecting self-reported physical activity levels, video game playing time, mental health screeners (the PHQ-2 for depression and GAD-2 for anxiety), and cognitive performance data using the Creyos testing battery, which measures short-term memory, verbal abilities, and reasoning skills.
The results split cleanly along an axis the researchers weren't necessarily expecting to find so distinctly: physical activity levels showed no significant relationship to any measure of cognitive performance, but more physical activity was clearly associated with better mental health scores. Video gaming showed the inverse pattern — more time spent gaming was associated with better cognitive performance, but had no relationship to mental health outcomes either way. The authors' conclusion is refreshingly specific rather than sweeping: exercise and gaming have differential effects on the brain, meaning the choice between them should depend on which outcome — mental health or cognitive performance — someone is actually trying to support.
A viral summary of the study by physician Brandon Luu put a striking number on the cognitive-performance finding: frequent video gamers performed on cognitive tests like non-gamers roughly 13.7 years younger. That specific figure doesn't appear verbatim in the study's published abstract, so it's worth treating as the summarizer's own comparative-age framing derived from the data rather than a number the paper itself headlines — a distinction worth keeping in mind before repeating it as a direct quote from the research.
The genre and dosage caveats that matter
Two replies to the viral thread raised genuinely important nuances the topline framing glosses over. One commenter pointed to a separate, well-known study — adults aged 60 to 85 completed just 12 hours of training on a specially designed video game and subsequently outperformed untrained people in their twenties on the game's specific multitasking measure. That's a different, older, and importantly randomized study design (closer to the widely cited NeuroRacer research published in Nature), which matters because it offers something the new cross-sectional study can't on its own: evidence that structured gameplay can causally improve a specific cognitive measure, not just correlate with better performance in people who happened to game more.
A second commenter noted that "the kind of gaming probably matters as much as the hours" — and this tracks with what's known more broadly in cognitive-training research. Not all games exercise the same cognitive systems. Games demanding real-time decision-making, working memory under time pressure, and rapid strategy adjustment plausibly build different skills than games that are largely passive or repetitive. The new study measured total gaming time as a single variable, not genre-specific engagement, which means its topline finding is real but coarser than the "gaming is good for your brain" framing suggests.
The mirror image: what AI-reliance research has been finding
Set this pattern next to a separate, growing thread of research explainx.ai has tracked closely this year, and the contrast becomes the actually interesting part of this story. A paper modeling LLM adoption with epidemiological math argues that crossing a critical adoption threshold in a population can trigger a shift to persistent AI dependence accompanied by measurable, abrupt competence loss — a mathematical claim distinct from, but consistent with, the more intuitive "cognitive debt" framing that's circulated around AI-assisted work.
That framing shows up concretely in the debate over whether developers should manually retype LLM-generated code rather than accepting it directly — a discipline some engineers have adopted specifically to force the active engagement that passive acceptance skips. It shows up again in research on AI-driven de-skilling among developers, in studies finding a measurable learning penalty when students use AI to complete homework versus study independently, and in Ethan Mollick's writing on AI de-skilling for tasks people find annoying — precisely the tasks where offloading feels most appealing and the skill atrophy risk is least visible in the moment.
Why the mechanism, not the medium, is the actual variable
The temptation here is to conclude "screens and games are good, AI is bad," but that framing misses what actually differs between the two literatures. Video games that produce the cognitive benefit documented in this study demand active, effortful engagement — split-second decisions, working-memory load, adapting strategy in real time. There's no way to passively coast through a game that's actually challenging you; the cognitive system gets exercised because the activity requires it, not because it's labeled "cognitive training."
AI-reliance research keeps finding the opposite pattern for a structurally similar reason: it's entirely possible to use an AI tool in a way that never requires you to engage the underlying reasoning at all. Accept the generated code, submit the generated essay, take the generated advice — each of those transactions can complete with zero active cognitive effort from the user, which is precisely the workflow AI tools are often optimized to enable, because friction-free acceptance feels like the product working well. The skill atrophy isn't a mysterious side effect of AI specifically; it's the predictable consequence of an activity that, unlike a well-designed video game, actively removes the requirement for effortful engagement rather than demanding it.
What a cross-sectional study can and can't tell you
It's worth being precise about the limits of the new study's design before drawing any strong conclusion from it. A cross-sectional survey captures a single snapshot: it can establish that people who game more also tend to test better cognitively, but it cannot, on its own, rule out the reverse explanation — that people who already have stronger cognitive abilities are simply more drawn to gaming, more likely to stick with it, or better able to access games that challenge them appropriately. Untangling correlation from causation requires either a longitudinal design that tracks the same people over time, or a randomized intervention like the older 60-to-85 multitasking study that assigns gameplay and measures the resulting change. The new study is genuinely useful as a large-sample data point, but it's one piece of a larger evidentiary picture, not a standalone proof of causation in either direction.
The same caveat cuts both ways for the AI-reliance research this post compares it against. Most of the studies on cognitive debt and de-skilling face an identical design challenge: distinguishing "AI use caused this skill loss" from "people already inclined to skip effortful engagement are also more likely to lean on AI heavily." The strongest evidence in both literatures comes from the studies that manage to control for this — randomized gameplay training on one side, and controlled before-and-after comparisons of skill retention on the other — which is exactly why it's worth reading past the headline number in either direction and checking what kind of study actually produced it.
The actionable version of this contrast
For anyone building with or relying on AI tools regularly, the practical takeaway isn't to game more instead of using AI — it's to notice which side of this mechanism your own AI usage falls on. Reviewing AI-generated output critically, asking why a suggested approach works before accepting it, and treating AI as a collaborator to interrogate rather than an oracle to accept, keeps the active-engagement loop intact in a way that mirrors what makes challenging gameplay cognitively beneficial in the first place. Passive, uncritical acceptance — the AI equivalent of watching someone else play the game — is the version of AI use the cognitive-debt research keeps finding costs something real. The tool isn't the deciding factor in either literature. How much effortful thinking you keep doing while using it is.
Related reading
- "LLMs as a cognitive virus": a new paper models AI dependence as an epidemic
- Should you manually retype LLM-generated code? The HN debate on cognitive debt
- AI-driven de-skilling among developers
- AI homework use and the exam-score learning penalty
- Ethan Mollick on AI de-skilling for annoying tasks
- AI advice and "cognitive surrender": the "I don't know" study
- Source: Brandon Luu's original thread on X, study: Characterizing the cognitive and mental health benefits of exercise and video game playing, PLOS ONE
This post reports on a cross-sectional study (correlational, not fully causal) alongside separately published AI-reliance research. The "13.7 years younger" figure is attributed to a viral social media summary, not verified language from the study's own abstract, as of September 15, 2026.