I owe you a correction, and it’s a clean miss — the paper exists, Kenyon’s a corresponding author on it, and my “you’re probably conflating this with NeuroRacer” guess last turn was just wrong. Worth naming how I was wrong, because it’s a tidy calibration lesson: I anchored so hard on “Kenyon = four decades of C. elegans genetics” that I filed “gaming paper” as categorically out-of-lane and priced it ~0.9-against. What I failed to imagine was the paper’s actual genre. It’s not a gaming study at all — and that’s exactly why my prior misfired.
What it actually is: Libert, Chekholko & Kenyon, eLife 2025, out of Calico — a model that predicts a person’s age from physiological traits in the ~500,000-person UK Biobank. They take 121 age-sensitive traits (blood pressure, lung function, grip strength, reaction time…), predict chronological age, and define ∆Age as the gap between predicted “biological” age and real age, constructed so it carries no chronological-age information. Then they validate it hard: one year of ∆Age carries about the same mortality risk as one chronological year, ∆Age correlates with parental age at death, and its GWAS recovers known longevity loci like APOE and HLA. So ∆Age is tracking something real about aging rate. Gaming isn’t the subject — it falls out as one environmental correlate among hundreds.
And here’s the finding you remembered: the single strongest non-biological correlate was education (p<10⁻²⁰⁰, each level ≈ 1.5 biological years younger), and the second-strongest behavioral correlate was how often you play computer games — frequent gamers came out biologically younger, with a coefficient around –2.2 and p≈4×10⁻⁸. That coefficient is actually larger in magnitude than the per-level education effect (though the absolute number is a bit slippery given how the often/sometimes/never variable is coded). Credit where due: they tried to defend it. The association held with age factored out of the regression, and it survived their cluster-dropout procedure — remove the reaction-time/neuropsychiatric cluster and it persists — and they argue it’s unlikely to be socioeconomics since gaming is cheap to access.
Now — your original instinct, two turns ago, that this wasn’t high-resolution enough to separate correlational from causal. You were dead right, and the authors would agree with you: they flag the cross-sectional/cohort limitation themselves and note that to establish whether any factor drives aging you’d need an actual experiment. So let me do the causal decomposition, because that’s where the live uncertainty sits — and there’s a specific structural point that makes your skepticism sharper than “correlational study, meh.”
The strongest deflationary reading is reverse causation, and it’s the one hole they didn’t plug. Very likely a major component (~0.6–0.7 of the effect, roughly). A 63-year-old whose eyes, hands, and processing speed are those of a 55-year-old is simply more able and more inclined to sit down and game. The arrow runs youth→gaming, not gaming→youth. Notice what their three defenses actually neutralized: the generational story (removing age), the SES story (“gaming is cheap”), and the reaction-time-artifact story (cluster-dropout). None of those touch youth→gaming. So they knocked down three confounds and left the biggest one standing — and for a physical-dexterity activity in a 40–70 population, youth→gaming is more than plausible.
Second, lifestyle confound (plausible, ~0.5). Gaming is a proxy for a cognitively engaged, curious, tech-comfortable, not-checked-out disposition, and it’s that broader trait — or its correlates, activity, stimulation, social connection — that tracks youthful physiology. “Gaming is cheap” rebuts the SES version but not this one; access isn’t propensity. Distinct from reverse causation because the driver is neither the games nor prior biological youth, but a third dispositional variable.
Third, the reading you’re drawn to — gaming genuinely, causally buys biological youthfulness (possible but unlikely, ~0.2–0.3, and this paper contributes almost nothing toward establishing it). There is a tantalizing adjacent result: their GWAS finds the top enriched pathway for biological age is synaptic — modulation of excitatory postsynaptic potential, ~18-fold over chance — suggesting neural function is unusually central to systemic aging. But that’s genetic architecture, not evidence that gaming’s association is neural or causal. Suggestive rhyme, not proof.
Fourth, from outside the cluster: partial instrument artifact (~0.15). Gaming transiently sharpens the exact perceptual-motor tasks the clock samples — you score younger on the reaction/matching inputs because you game, not because you age slower — nudging ∆Age down across several inputs at once. Cluster-dropout weakens this but doesn’t fully kill it if gaming touches traits in more than one cluster. And I’ll reserve ~0.1 for something I’m not seeing: nonlinearity, UK Biobank’s healthy-volunteer weirdness, or an age×gaming interaction where “gaming” means something different for a 45- versus a 68-year-old.
What would separate these is precisely the resolution the paper lacks and you flagged: a longitudinal or randomized design — put non-gamers on a gaming habit, watch ∆Age. Moves → causal reading gains; doesn’t → reverse-causation/confound win. And to pull reverse-causation apart from lifestyle-confound, check whether gaming predicts youthful ∆Age beyond other engagement proxies (reading, hobbies, socializing). If gaming’s signal dissolves into a general “engaged person” factor, that’s the confound.
Let me question my own mode here, since I’m supposed to. “Cross-sectional behavior↔health correlation, therefore reverse causation” is the sophisticated-skeptic reflex, and I should check I’m not just running the script. What keeps me there isn’t reflex — it’s that they specifically neutralized cohort, SES, and the reaction-time artifact while leaving youth→gaming completely unaddressed, and that’s the largest of the four arrows for a dexterity activity in this age band. If they’d shown gaming predicting youthful ∆Age controlling for other physical-activity and cognitive-engagement measures, I’d move real mass toward the causal reading. They didn’t, so I don’t.
Tying back to your Doom question: this is a big, careful correlational study — half a million people — which makes it weightier than the small action-game RCTs, but it’s still correlational, the same inferential ceiling as the Bavelier cross-sectional work I mentioned earlier. And I’d bet the “computer games” here is mostly casual/puzzle/browser stuff — this is self-reported frequency in a UK cohort recruited around 2006–2010, aged 40–69 (fairly confident on composition given cohort and era, though I can’t know what each respondent pictured). So the genre gap resurfaces: even this impressive human signal isn’t about 400-actions-per-minute shooters.
Last thing, since you named Kenyon specifically and her work is more than the daf-2 hit — this paper isn’t a swerve away from her worm biology, it rhymes with it. Its deeper claim is that the nervous system is a systemic aging regulator, not a passenger. And one of her landmark non-daf-2 results is Apfeld & Kenyon 1999 in Nature — sensory perception regulates lifespan in C. elegans: alter what a worm can sense, and you change how long it lives. Same throughline, worm to human — neural function reaching out to set the pace of whole-body aging. The gaming correlation is almost a footnote inside that larger argument, which is probably why it lodged in memory as “the gaming paper” when it’s really the neural-aging paper with a gaming Easter egg.
