Resources
Research, data and regulation
Peer-reviewed studies, official epidemiological data, court rulings and legislation from a dozen countries on the effect of digital environments on mental life.
Each reference sets out its finding, the limits of its interpretation and its degree of scientific consolidation. The library can be filtered by field and by certainty, and every entry links to the original source.
Levels of certainty
- Established evidenceReplicated result or stable official data.
- Open debateSound research with an unresolved methodological disagreement among specialists.
- Emerging signalRecent finding, without independent replication or not yet peer reviewed.
-
Open debate
Links between social media use and mental wellness in youth are an artifact of other…
Youth mental health · Experimental study
In a sample of thousands of British youths, raw correlations between social media use and wellbeing explained 1–4 % of the variance, and practically 0 % once controls were introduced.
Ferguson · 2026 · Current Psychology
Caveat
It is cross-sectional and single-sample, and controlling for everything carries its own risk: some variables may be mediators rather than confounders, artificially absorbing a real effect.
https://link.springer.com/article/10.1007/s12144-026-09205-3 · DOI 10.1007/s12144-026-09205-3
-
Open debate
Planned Social Media Ban and Depression and Anxiety in Adolescents
Youth mental health · Epidemiological data
In 365 Australian adolescents aged ten to fifteen surveyed before the ban, the age of starting on social media — eleven on average — was associated with anxiety and depression symptoms more strongly than merely holding an account.
Bilston, Brown, Qi, Werner-Seidler and Schweizer · 2026 · JAMA Network Open
Caveat
It is cross-sectional and self-reported, with a small and unrepresentative sample: it cannot establish causation or rule out the reverse, that those with more symptoms start earlier. Thirty-five per cent said they would circumvent the ban.
https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2853381 · DOI 10.1001/jamanetworkopen.2026.31107
-
Open debate
Parasocial Engagement With Social Media Influencers and Mental Health Outcomes…
Youth mental health · Meta-analysis
Across seventeen studies and fifty-two effect sizes, parasocial bonding with creators is associated more strongly with positive outcomes (r = .38) than negative ones (r = .24), and far more with one-off interaction than with sustained relationship.
Li, Liu and Liu · 2026 · JMIR Mental Health
Caveat
The evidence base is almost entirely cross-sectional, so causal direction is undetermined. It is a useful corrective to the reading that all mediated bonding is harmful, not proof of harmlessness.
-
Emerging signal
Addictive Screen Use Trajectories and Suicidal Behaviors, Suicidal Ideation, and Mental…
Youth mental health · Longitudinal study
In 4,285 children followed for four years, trajectories of compulsive phone and social media use doubled the risk of suicidal ideation and behaviour, while total screen time was not associated with any outcome.
Xiao, Meng, Brown, Keyes and Mann · 2025 · JAMA
Caveat
It is not causal: a compulsive pattern may be a consequence of distress as much as a precursor. Its main contribution is shifting the focus from how much is used to how it is used.
https://jamanetwork.com/journals/jama/fullarticle/2835481 · DOI 10.1001/jama.2025.7829
-
Established evidence
The Effect of Deactivating Facebook and Instagram on Users’ Emotional State
Youth mental health · Experimental study
In two randomised experiments with more than 35,000 users, deactivating the account for six weeks improved emotional state by 0.06 standard deviations, concentrated among young women.
Allcott, Gentzkow, Wittenbrink et al. · 2025 · National Bureau of Economic Research / American Economic Journal
Caveat
The sample is adult, so it cannot support child policy; the effect is real but small, and measures individual disconnection, not what would happen if a whole cohort left.
https://www.nber.org/system/files/working_papers/w33697/w33697.pdf · DOI 10.1257/pol.20240806
-
Established evidence
A focus on adolescent social media use and gaming in Europe, central Asia and Canada
Youth mental health · Epidemiological data
Across nearly 280,000 adolescents in 44 countries, problematic social media use rose from 7 % in 2018 to 11 % in 2022, and 12 % show a risk of problematic gaming.
World Health Organization, Europe (HBSC study) · 2024 · WHO Regional Office for Europe
Caveat
It measures the prevalence of problematic use, not its impact: it establishes no link with depression or self-harm. The 2018–2022 jump is also sensitive to the pandemic.
https://www.who.int/europe/publications/i/item/9789289061322
-
Established evidence
Youth Risk Behavior Survey Data Summary & Trends Report: 2013–2023
Youth mental health · Epidemiological data
Among US high school students, persistent sadness rose from 30 % to 40 % over a decade and suicidal ideation from 17 % to 20 %; both indicators fell slightly between 2021 and 2023.
Centers for Disease Control and Prevention · 2024 · CDC, United States
Caveat
The survey contains no measure of screen use, so the trend cannot be attributed to technology. Citing the decade-long rise without the recent fall skews the reading.
https://www.cdc.gov/yrbs/dstr/pdf/YRBS-2023-Data-Summary-Trend-Report.pdf
-
Open debate
The Anxious Generation: How the Great Rewiring of Childhood Is Causing an Epidemic of…
Youth mental health · Essay or commentary
Argues that replacing a play-based childhood with a phone-based childhood from 2010 onward is the main cause of declining adolescent mental health, and proposes four protective norms.
Haidt · 2024 · Penguin Press
Caveat
It is an authored essay, not peer-reviewed literature: the source of a hypothesis, not of a result. No scientific consensus supports it, and it should be read alongside its critics.
https://www.penguinrandomhouse.com/books/729231/the-anxious-generation-by-jonathan-haidt/
-
Open debate
The great rewiring: is social media really behind an epidemic of teenage mental illness?
Youth mental health · Essay or commentary
A critical reply to the previous thesis: the available data are correlational and compatible with reverse causation, and focusing on screens diverts attention from well-documented determinants.
Odgers · 2024 · Nature
Caveat
It is a review, not a systematic review: it brings no new data. Nor does it show social media is harmless, only that the evidence does not support strong causal attribution.
https://www.nature.com/articles/d41586-024-00902-2 · DOI 10.1038/d41586-024-00902-2
-
Established evidence
Social media use and its impact on adolescent mental health: An umbrella review of the…
Youth mental health · Systematic review
A review of twenty-five earlier reviews: the aggregate association between social media use and adolescent mental health is weak and inconsistent, with correlations between r = .02 and r = .17.
Valkenburg, Meier and Beyens · 2022 · Current Opinion in Psychology
Caveat
A small average effect does not mean no harm. The authors estimate that 10–15 % of adolescents experience negative effects and a similar share positive ones; the average cancels them out.
https://pure.uva.nl/ws/files/69982711/1_s2.0_S2352250X21001500_main.pdf · DOI 10.1016/j.copsyc.2021.08.017
-
Open debate
Windows of developmental sensitivity to social media
Youth mental health · Longitudinal study
With 84,011 British participants, social media use predicts lower life satisfaction a year later only within specific windows: ages 11–13 in girls, 14–15 in boys, and again around 19.
Orben, Przybylski, Blakemore and Kievit · 2022 · Nature Communications
Caveat
The authors present it as exploratory and it has not been consistently replicated. On its own it cannot justify a legal minimum age.
https://www.nature.com/articles/s41467-022-29296-3 · DOI 10.1038/s41467-022-29296-3
-
Established evidence
Social Media and Mental Health
Youth mental health · Experimental study
Exploiting Facebook’s staggered rollout across universities between 2004 and 2006, the platform’s arrival raised reported severe depression by 7 % and anxiety disorder by 20 %.
Braghieri, Levy and Makarin · 2022 · American Economic Review
Caveat
It is the most cited causal evidence, and the most overstretched: the Facebook of 2004 had neither a recommender algorithm nor a phone, and the population was undergraduates, not minors.
https://www.aeaweb.org/articles?id=10.1257%2Faer.20211218 · DOI 10.1257/aer.20211218
-
Established evidence
Preliminary findings on the addictive design of Instagram and Facebook
Design and attention · Legal action
On 10 July 2026 the Commission charges Meta over infinite scroll, autoplay, notifications and highly personalised recommendation as a systemic risk to minors.
European Commission · 2026 · European Commission, press release IP/26/1579
Caveat
It is the first time a regulator has treated addictive design as an offence in itself rather than a consequence of content. Fines may reach 6 % of worldwide turnover, but there is as yet no final decision and no case law defining the threshold.
-
Open debate
Algorithmic recommendation and adolescent mental health: a grounded theory study of…
Design and attention · Qualitative study
With twenty-seven British young people interviewed over screenshots of their own feeds, it identifies “algorithmically structured exposure” as the central mechanism, sustained by system reinforcement, emotional feedback loops and little perceived control.
Winstone, Parsonage, Cross et al. · 2026 · BMC Public Health
Caveat
The sample is self-selected, British and skewed towards girls, with no systematic record of mental health history: it offers conceptual transferability, not statistical generalisation. It locates harm in commercial design rather than individual behaviour.
https://link.springer.com/article/10.1186/s12889-026-28300-5 · DOI 10.1186/s12889-026-28300-5
-
Open debate
Problematic Social Media Use and Attention-Deficit/Hyperactivity Disorder Symptoms in…
Design and attention · Longitudinal study
In 11,286 adolescents from the ABCD cohort followed for five years, problematic social media use predicted parent-reported ADHD symptoms a year later, with only a partial reverse relationship and a stronger effect in boys.
Nagata et al. · 2026 · JAMA Network Open
Caveat
The result depends on who reports: symptoms self-reported by the adolescents showed no association at all, which suggests it may be measuring parental perception rather than behavioural change.
https://pubmed.ncbi.nlm.nih.gov/42467432/ · DOI 10.1001/jamanetworkopen.2026.23748
-
Established evidence
Manipulative online practices found on 148 out of 399 online shops screened
Design and attention · Institutional report
Of 399 shops screened across twenty-five countries, 148 showed at least one manipulative pattern: fake countdowns, interfaces steering users to costlier subscriptions, or hidden costs.
European Commission and national consumer authorities · 2023 · European Commission
Caveat
A sweep is a screening, not a ruling: the Commission speaks of potential breaches. Sites were chosen by national relevance, so the 37 % is not a market-wide rate.
https://ec.europa.eu/commission/presscorner/detail/en/ip_23_418
-
Open debate
Defining and Identifying Attention Capture Deceptive Designs in Digital Interfaces
Design and attention · Systematic review
A review of 43 studies establishing eleven attention-capture patterns — infinite scroll, time fog and recapture notification among them — each linked to the vulnerability it exploits.
Monge Roffarello, Lukoff and De Russis · 2023 · ACM CHI Conference on Human Factors in Computing Systems
Caveat
It is a conceptual taxonomy, not a meta-analysis of effects: cataloguing a pattern is not the same as proving how much harm it causes.
https://iris.polito.it/bitstream/11583/2978380/2/Defining_and_Identifying_Attention_Capture_Deceptive_Designs_in_Digital_Interfaces.pdf · DOI 10.1145/3544548.3580729
-
Open debate
Arizona ex rel. Mayes et al. v. Meta Platforms
Design and attention · Legal action
The complaint alleges Meta designed Instagram and Facebook to maximise minors’ time on platform through variable reward, infinite scroll, autoplay and notifications, while concealing internal evidence of harm.
Thirty-three US state attorneys general · 2023 · US District Court for the Northern District of California
Caveat
These are party allegations, not proven facts: trial began in August 2026 and there is no judgment. Always cite as “the attorneys general allege”, never as “it is proven”.
-
Open debate
How do social media feed algorithms affect attitudes and behavior in an election…
Design and attention · Experimental study
Replacing the algorithmic feed with reverse-chronological order for three months sharply cut time on platform and changed what users saw, but did not significantly alter polarisation or political knowledge.
Guess, Malhotra, Pan, Barberá, Allcott et al. · 2023 · Science
Caveat
A null result at three months does not prove absence of effect, and the study drew a formal challenge in Science over emergency measures Meta applied to the algorithm during the control period.
https://www.science.org/doi/10.1126/science.abp9364 · DOI 10.1126/science.abp9364
-
Established evidence
Dark Patterns at Scale: Findings from a Crawl of 11K Shopping Websites
Design and attention · Experimental study
An automated crawl of eleven thousand online shops identified 1,818 dark patterns across 1,254 sites, grouped into fifteen types, and twenty-two vendors offering them as a paid service.
Mathur, Acar, Friedman, Lucherini, Mayer, Chetty and Narayanan · 2019 · Proceedings of the ACM on Human-Computer Interaction
Caveat
It measures prevalence, not effect: it says nothing about how much money or autonomy users lose. The data are from 2019 and cover English-language commerce only.
-
Open debate
Brain Drain: The Mere Presence of One’s Own Smartphone Reduces Available Cognitive…
Design and attention · Experimental study
The original study reported that the mere presence of one’s phone, switched off and unchecked, reduces available cognitive capacity. Two preregistered replication attempts found no difference at all.
Ward, Duke, Gneezy and Bos; replication by Ruiz Pardo and Minda · 2017 · Journal of the Association for Consumer Research
Caveat
We include it precisely because it is the field’s most circulated and least supported citation. Publishing it without its failed replication would be the error this section exists to prevent.
https://forrt.org/flora-replication-atlas/doi/10.1086/691462/ · DOI 10.1086/691462
-
Emerging signal
Screen Time Is Associated with Altered Neural Maturation in a Massive Adolescent Cohort
Development and neuroscience · Longitudinal study
Using two-timepoint structural and diffusion imaging in 2,666 adolescents from the ABCD cohort, more screen time was associated with microstructural alterations in eight of thirteen regions, consistent with accelerated maturation in the reward circuit and slowed maturation in cognitive control.
Atalay, Newman and Druzgal · 2026 · Children
Caveat
Screen time is parent-reported and was averaged across the two timepoints rather than measured as change, and the model omits weighty confounders such as puberty, sleep and psychiatric symptoms. Microstructural effects of this kind have small effect sizes and a history of poor replication.
https://www.mdpi.com/2227-9067/13/7/969 · DOI 10.3390/children13070969
-
Open debate
Guidelines on screen use in early childhood
Development and neuroscience · Institutional guideline
The WHO advises against any screen use before age two and caps it at one hour between two and four. Spanish paediatrics goes further: zero screens until six. The US body is more permissive.
World Health Organization; American Academy of Pediatrics; Spanish Association of Paediatrics · 2025 · WHO (2019), AAP (2016), AEP (2024-2025)
Caveat
The WHO itself rates the quality of the evidence behind its thresholds as “very low”. That two paediatric bodies disagree using the same data shows the dispute is normative, not empirical.
-
Open debate
Screen Time at Age 1 Year and Communication and Problem-Solving Developmental Delay at…
Development and neuroscience · Longitudinal study
In 7,097 mother–child pairs from a Japanese cohort, more screen time at age one was associated in dose-dependent fashion with communication and problem-solving delay at two, persisting at four.
Takahashi, Obara, Ishikuro et al. · 2023 · JAMA Pediatrics
Caveat
Motor and social domains showed no persistent association, so one cannot speak of “developmental delay” wholesale. Reverse causation is plausible: a struggling child may be given more screen time.
https://jamanetwork.com/journals/jamapediatrics/fullarticle/2808593 · DOI 10.1001/jamapediatrics.2023.3057
-
Established evidence
Associations Between Screen Use and Child Language Skills: A Systematic Review and…
Development and neuroscience · Meta-analysis
Across 42 studies and 18,905 children, more screen time was associated with poorer language skills, while educational content and co-viewing with an adult were associated with better skills.
Madigan, McArthur, Anhorn, Eirich and Christakis · 2020 · JAMA Pediatrics
Caveat
Citing only the negative figure misrepresents the source: the same meta-analysis shows coefficients of opposite sign depending on content and co-viewing. Quality weighs as much as quantity.
https://jamanetwork.com/journals/jamapediatrics/fullarticle/2762864 · DOI 10.1001/jamapediatrics.2020.0327
-
Emerging signal
Associations Between Screen-Based Media Use and Brain White Matter Integrity in…
Development and neuroscience · Experimental study
In 47 children aged three to five, a higher screen-use index correlated with lower white matter integrity in tracts associated with language and emergent literacy.
Hutton, Dudley, Horowitz-Kraus, DeWitt and Holland · 2020 · JAMA Pediatrics
Caveat
It is the most over-interpreted study in the field: 47 children, a single time point, no control group and no baseline. It does not license the headline that screens change children’s brains.
https://jamanetwork.com/journals/jamapediatrics/fullarticle/2754101 · DOI 10.1001/jamapediatrics.2019.3869
-
Established evidence
Don’t throw away your printed books: A meta-analysis on the effects of reading media on…
Development and neuroscience · Meta-analysis
Across 54 studies and 171,055 participants, print reading outperformed digital in comprehension, with the advantage concentrated under time pressure and in informational texts, and nil for narrative.
Delgado, Vargas, Ackerman and Salmerón · 2018 · Educational Research Review
Caveat
The effect is small and highly conditional: it does not apply to narrative reading nor, largely, to untimed reading. It does not support the claim that digital is inferior for every purpose.
https://www.uv.es/lasalgon/papers/Delgado%202018%20dont%20throw%20away%20your%20printed%20books.pdf · DOI 10.1016/j.edurev.2018.09.003
-
Established evidence
Association Between Portable Screen-Based Media Device Access or Use and Sleep Outcomes
Development and neuroscience · Meta-analysis
Across twenty studies and 125,198 children, having a device at bedtime doubled the risk of insufficient sleep; the mere presence of the device in the bedroom, unused, already raised it by 79 %.
Carter, Rees, Hale, Bhattacharjee and Paradkar · 2016 · JAMA Pediatrics
Caveat
The pooled studies are mostly cross-sectional and self-reported. It is the best-established link in the field, but does not show that removing the device restores sleep to the same degree.
https://jamanetwork.com/journals/jamapediatrics/fullarticle/2571467 · DOI 10.1001/jamapediatrics.2016.2341
-
Established evidence
Decision no. 2026-911 DC, on the law protecting minors from social media
Regulation · Legal action
The Council struck down the article banning social media for under-fifteens as a disproportionate infringement of freedom of communication and for lacking safeguards on age verification.
Constitutional Council of France · 2026 · Conseil constitutionnel
Caveat
France has passed two minimum-age laws in three years and has none in force: the 2023 act never received implementing decrees owing to incompatibility with EU law, and the 2026 one was struck down.
https://www.conseil-constitutionnel.fr/decision/2026/2026911DC.htm
-
Established evidence
Settlement in the states’ litigation against Meta over harm to minors
Regulation · Legal action
After eight days of trial, on 26 August 2026 Meta agreed to end the proceedings by paying up to 17 billion dollars over ten years and accepting verifiable design obligations: age assurance, a default two-hour daily limit for teenagers, an overnight block, notification silencing, an available chronological feed, hidden likes and independent auditing.
State attorneys general and Meta Platforms · 2026 · US District Court, Northern District of California
Caveat
It is not a judgment: the company admits no liability and the settlement does not end the litigation. Florida stayed out and will go to trial, and thousands of personal-injury and school-district claims continue. Published figures differ, because the documentation distinguishes some 12.2 billion guaranteed from a 17.1 billion ceiling conditional on other platforms taking on equivalent obligations.
https://www.npr.org/2026/08/26/nx-s1-5944781/meta-settlement-child-safety-lawsuit
-
Established evidence
Preliminary findings against TikTok for failing to ensure safe accounts for minors
Regulation · Legal action
In July 2026 the Commission preliminarily finds that minors’ accounts on TikTok are too visible by default, and requires their visibility to be limited to accepted users and their exclusion from the recommender system.
European Commission · 2026 · European Commission
Caveat
These are preliminary findings, not a final decision: the company may respond and the case may close with commitments. It is part of a series: in April 2026 the Commission charged Meta with failing to prevent access by under-thirteens, and in July, with the addictive design of Instagram and Facebook.
https://ec.europa.eu/commission/presscorner/detail/en/ip_26_1679
-
Open debate
Organic Bill for the protection of minors in digital environments
Regulation · Legal instrument
It would raise the age of consent for data processing from fourteen to sixteen, require parental controls enabled by default on devices, and ban paid loot boxes for minors.
Spanish Congress of Deputies · 2025 · Official Gazette of the Cortes Generales, bill 121/000052
Caveat
It is not in force. As of September 2026 it remained in committee in the Congress, with no floor vote and the Senate stage entirely pending; amendments tabled in July would raise the age of access to social media to sixteen. Headlines claiming Spain has already banned it are false.
https://www.congreso.es/public_oficiales/L15/CONG/BOCG/A/BOCG-15-A-52-1.PDF
-
Established evidence
Guidelines on the protection of minors online, Article 28 of the Digital Services Act
Regulation · Legal instrument
They recommend minors’ accounts private by default, disabling addictive features such as streaks and autoplay, adjusting recommender systems, and age assurance proportionate to risk.
European Commission · 2025 · European Commission
Caveat
They are non-binding guidance, and Article 28 sets no minimum age. As of today there is no harmonised digital minimum age in the European Union, despite the Parliament having called for one.
https://digital-strategy.ec.europa.eu/en/library/commission-publishes-guidelines-protection-minors
-
Established evidence
Protection of Children Codes of Practice, Online Safety Act 2023
Regulation · Legal instrument
Since July 2025, services hosting pornography or suicide and self-harm content must apply highly effective age assurance, alongside strengthened moderation and safer recommendation.
United Kingdom, Ofcom · 2025 · Ofcom
Caveat
The two documented side effects are mass use of virtual private networks to circumvent it and defensive geoblocking: small services that would rather close UK access than comply.
-
Established evidence
Law 15.211, Digital Statute of the Child and Adolescent
Regulation · Legal instrument
In force since March 2026, it requires platforms, app stores and operating systems to verify age reliably — banning mere self-declaration — with fines of up to 10 % of group turnover.
Federative Republic of Brazil · 2025 · Brazilian Data Protection Authority (ANPD)
Caveat
It is the most ambitious framework outside Australia, but has been applied for only a few months. Content removal without prior judicial order has drawn freedom-of-expression objections.
-
Established evidence
Online Safety Amendment (Social Media Minimum Age) Act 2024
Regulation · Legal instrument
Since December 2025, designated platforms must prevent under-sixteens from holding an account, under fines of up to 49.5 million Australian dollars. Neither the minor nor the family is ever penalised.
Commonwealth of Australia · 2024 · Federal Register of Legislation of Australia
Caveat
It is the only such regime fully in force and its enforcement is deficient: in March 2026 the regulator identified compliance failures at the five largest platforms, and the first independent evaluation found that over 85 % of under-sixteens were still using them three months on.
-
Open debate
The Tragedy of the Cognitive Commons: How AI Could Disrupt the Regeneration of…
Artificial intelligence · Essay or commentary
Each organisation adopting artificial intelligence captures the efficiency gain, while the cost — erosion of the channel through which a profession regenerates its expertise — is spread across the whole field.
Lovett · 2026 · Human Resource Development Review
Caveat
It is a peer-reviewed theoretical framework with no data of its own: it cannot show the erosion is already happening. The author places observable consequences between 2030 and 2045.
https://journals.sagepub.com/doi/10.1177/15344843261470602 · DOI 10.1177/15344843261470602
-
Open debate
AI, Human Cognition and Knowledge Collapse
Artificial intelligence · Preprint, not peer reviewed
In a dynamic model, artificial intelligence improves immediate decisions but reduces the incentive to generate general knowledge, and welfare is maximised at intermediate, not maximal, accuracy.
Acemoglu, Kong and Ozdaglar · 2026 · National Bureau of Economic Research
Caveat
It is an economic model without empirical validation and without peer review. Its conclusions rest on contestable assumptions and it has already drawn academic rebuttals.
-
Emerging signal
Layer-sensitive cognitive offloading in generative AI-assisted writing: supported…
Artificial intelligence · Experimental study
Over eight weeks with 168 undergraduates, open collaboration with the model performed better with assistance, but bounded assistance performed better on the final unassisted task, with a negative indirect effect of cognitive offloading on independent thinking.
Chen · 2026 · Frontiers in Psychology
Caveat
Only six classes were assigned and p values hover around 0.05: the bounded condition mixes restricted delegation with mandatory reflection, so the design cannot attribute the effect to the restriction alone.
https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2026.1906199/full · DOI 10.3389/fpsyg.2026.1906199
-
Open debate
Cognitive Commons in the Age of Generative Intelligence: A Heterodox Appraisal of the…
Artificial intelligence · Preprint, not peer reviewed
It argues that the thesis of cognitive collapse through delegation to AI, though coherent, overstates the uniformity of human responses and understates adaptive capacity, and proposes governing the question by cognitive domain rather than wholesale.
Kallel and El Louadi · 2026 · Deposited on arXiv
Caveat
It is a preprint without peer review and without original data: an argumentative counterweight, not evidence. It is filed alongside Lovett because it directly rebuts it, and neither should be read without the other.
-
Emerging signal
Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for…
Artificial intelligence · Preprint, not peer reviewed
Those who wrote with a language model showed the lowest brain connectivity of the three groups, the least ownership of their text, and difficulty quoting what they had just written.
Kosmyna, Hauptmann, Yuan et al. · 2025 · MIT Media Lab, deposited on arXiv
Caveat
It remains unreviewed and has drawn a formal critical comment (arXiv:2601.00856) objecting to sample size, EEG methodology and reproducibility. It measures correlates during one specific task, not general cognitive capacity: it does not say AI damages the brain.
-
Established evidence
The Impact of Generative AI on Critical Thinking
Artificial intelligence · Experimental study
Across 319 knowledge workers and 936 real-world uses, greater confidence in the tool was associated with less critical thinking, and greater self-confidence with more.
Lee, Sarkar, Tankelevitch, Drosos, Rintel, Banks and Wilson · 2025 · ACM CHI Conference on Human Factors in Computing Systems
Caveat
Everything is self-reported: it measures perceived effort, not objectively measured capacity. It documents a shift in the kind of cognitive task, not necessarily a loss of faculty.
https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/ · DOI 10.1145/3706598.3713778
-
Established evidence
On the conversational persuasiveness of GPT-4
Artificial intelligence · Experimental study
In debates with nine hundred participants, a model with access to the opponent’s sociodemographic data was more persuasive than the human in 64.4 % of cases.
Salvi, Horta Ribeiro, Gallotti and West · 2025 · Nature Human Behaviour
Caveat
The advantage depends on access to the interlocutor’s data: without personalisation it was not significant. It is a structured debate format, and persistence of the opinion change was not measured.
https://www.nature.com/articles/s41562-025-02194-6 · DOI 10.1038/s41562-025-02194-6
-
Established evidence
FTC Launches Inquiry into AI Chatbots Acting as Companions
Artificial intelligence · Institutional report
The US authority ordered seven companies — among them Alphabet, Meta, OpenAI, Snap and X.AI — to explain how they measure and monitor the effects of companion chatbots on minors.
US Federal Trade Commission · 2025 · Federal Trade Commission
Caveat
It is an information-gathering power, not an accusation or a penalty: it evidences regulatory concern, not corporate liability.
-
Established evidence
Generative AI enhances individual creativity but reduces the collective diversity of…
Artificial intelligence · Experimental study
Access to model-generated ideas made stories be judged more creative, especially among less creative writers, but the assisted texts resembled one another more closely.
Doshi and Hauser · 2024 · Science Advances
Caveat
It measures similarity between short stories in a one-off task, not ideas or beliefs. Without follow-up it is unknown whether the effect persists, accumulates or fades with habitual use.
https://www.science.org/doi/10.1126/sciadv.adn5290 · DOI 10.1126/sciadv.adn5290
-
Established evidence
AI models collapse when trained on recursively generated data
Artificial intelligence · Experimental study
When generative models are trained recursively on data produced by earlier generations, the tails of the distribution vanish and output converges towards progressive impoverishment.
Shumailov, Shumaylov, Zhao, Papernot, Anderson and Gal · 2024 · Nature
Caveat
The result assumes complete replacement of human data by synthetic data; accumulating rather than replacing attenuates the effect. It predicts training pipelines, not human culture.
https://www.nature.com/articles/s41586-024-07566-y · DOI 10.1038/s41586-024-07566-y
-
Established evidence
Recommendation on the Ethics of Neurotechnology
Neurorights · Legal instrument
Adopted in November 2025, it is the first global instrument on neurotechnology: it recognises mental privacy as part of dignity, requires explicit consent to process neural data, and bans manipulative use.
UNESCO General Conference · 2025 · UNESCO
Caveat
It is not binding: it invites states to legislate but creates no enforceable obligations or sanctions. It should not be presented as international regulation of neurotechnology.
https://www.unesco.org/en/legal-affairs/recommendation-ethics-neurotechnology
-
Established evidence
State laws on the privacy of neural data
Neurorights · Legal instrument
Five states have added neural data to their privacy regimes as a sensitive category, with staggered entry into force between August 2024 and July 2026: Colorado, California, Connecticut, Montana and Vermont.
Colorado, California, Montana and Connecticut · 2025 · US state legislatures
Caveat
They are not neurorights laws: they are amendments to privacy statutes adding a category. They do not regulate devices or create substantive rights, and definitions differ across states. In Vermont, the most recent, the Senate removed the prior-consent requirement and the private right of action, leaving the text without an enforcement mechanism.
https://www.neurorightsfoundation.org/advocacy/united-states
-
Established evidence
Girardi Lavín v. Emotiv Inc., case 105.065-2023
Neurorights · Legal action
The Court ordered the company to delete brain data stored in the cloud, reasoning that its use for research requires express, specific consent, distinct from consent to purchase the device.
Supreme Court of Chile · 2023 · Supreme Court of Chile
Caveat
It is a single case decided through an injunctive procedure, binding only the parties, and much of the reasoning rested on data and health rules predating the constitutional reform.
https://derechocienciaytecnologia.uc.cl/sentencia/girardi-emotiv-inc-rol-105065-2023-corte-suprema/
-
Established evidence
Law 21.383, constitutional reform on science and technology at the service of persons
Neurorights · Legal instrument
It adds to Article 19 of the Constitution a mandate that the law regulate technological development while specifically safeguarding brain activity and the information derived from it.
Republic of Chile · 2021 · Library of the National Congress of Chile
Caveat
It is a mandate to the legislature, not an operative regime. The implementing statute remained unpassed at the last available check, dated July 2025.
-
Established evidence
Towards new human rights in the age of neuroscience and neurotechnology
Neurorights · Essay or commentary
The field’s founding text: it proposes four neuro-specific rights — cognitive liberty, mental privacy, mental integrity and psychological continuity — as a legal response to advancing neurotechnology.
Ienca and Andorno · 2017 · Life Sciences, Society and Policy
Caveat
It is legal theory, not empirical evidence, and its thesis is contested: one strand holds that existing human rights already protect these goods and that multiplying rights dilutes them.
https://pmc.ncbi.nlm.nih.gov/articles/PMC5447561/ · DOI 10.1186/s40504-017-0050-1
-
Emerging signal
Is the iPhone Birth Control? Causal Evidence from AT&T’s 2007–2011 Carrier Monopoly
Bonds and demography · Preprint, not peer reviewed
Using the iPhone’s single-carrier exclusivity as a quasi-experiment, it estimates its arrival cut births by 4.5–8 % among teenagers, and attributes up to half of the fertility decline to the phone.
Myers and Hooper · 2026 · National Bureau of Economic Research
Caveat
We publish it as an unaccredited hypothesis. It is a preprint without peer review, does not observe who owned a phone, coincides with the Great Recession, and its magnitude is hard to sustain at under 6 % adoption.
-
Emerging signal
Wide and Shallow: Digital Technology and the Post-2007 Fertility Decline
Bonds and demography · Preprint, not peer reviewed
With a structural model calibrated to US data from 2007 to 2024, plus instrumental variables on broadband and 4G rollout, it estimates that freezing phone prices at 2007 levels would account for 43 % of the fertility decline; in-person socialising fell 22 % while digital leisure rose 93 %.
Hudson and Moscoso Boedo · 2026 · Working paper, University of Cincinnati
Caveat
It is not peer reviewed and its causal identification is concentrated between ages fifteen and twenty-nine: the aggregate 43 % figure rests on the model, not on the quasi-experimental estimates. It has drawn the direct criticism that it inverts causation, since the phone is also matchmaking infrastructure.
https://homepages.uc.edu/~moscoshn/Personal_webpage/papers/Wide_Shallow_web.pdf
-
Established evidence
The Puzzle of Falling US Birth Rates since the Great Recession
Bonds and demography · Essay or commentary
The Great Recession explains the initial fall in births, but the sustained decline reflects a cohort effect: the authors state they cannot identify any economic or policy factor that accounts for it.
Kearney, Levine and Pardue · 2022 · Journal of Economic Perspectives
Caveat
It is above all a well-founded declaration of ignorance. The unexplained residual is large, and that does not license filling it with anyone’s preferred hypothesis.
https://www.aeaweb.org/articles?id=10.1257%2Fjep.36.1.151 · DOI 10.1257/jep.36.1.151
-
Established evidence
Trends in Frequency of Sexual Activity and Number of Sexual Partners Among Adults Aged…
Bonds and demography · Epidemiological data
Between 2000 and 2018, annual sexual inactivity among men aged eighteen to twenty-four rose from 18.9 % to 30.9 %, and doubled among those aged twenty-five to thirty-four.
Ueda, Mercer, Ghaznavi and Herbenick · 2020 · JAMA Network Open
Caveat
It is descriptive and identifies no causes. Among young women the rise was not statistically significant, an asymmetry that undercuts simple explanations.
https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2767066 · DOI 10.1001/jamanetworkopen.2020.3833
-
Established evidence
The Decline in Adult Activities Among U.S. Adolescents, 1976–2016
Bonds and demography · Epidemiological data
Across 8.4 million US adolescents, adult activities — driving, working, dating, drinking, having sex — have been declining steadily since 1976.
Twenge and Park · 2019 · Child Development
Caveat
The series begins three decades before the first smartphone, and the authors themselves explain it by income, longer schooling and smaller families. It is evidence against technological attribution.
https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.12930 · DOI 10.1111/cdev.12930
-
Open debate
Does broadband Internet affect fertility?
Bonds and demography · Experimental study
Using German data and broadband diffusion for identification, internet access raised the probability of having children among university-educated women, with null effects in other groups.
Billari, Giuntella and Stella · 2019 · Population Studies
Caveat
It is the only peer-reviewed literature on connectivity and fertility, and it finds the opposite sign to the popular narrative. One country, and a different technology from the mobile phone.
https://www.tandfonline.com/doi/abs/10.1080/00324728.2019.1584327 · DOI 10.1080/00324728.2019.1584327
No reference matches those filters.
Methodological note
Each reference was verified against its primary source — scientific journal, institutional repository, official gazette or court ruling — and those that could not be confirmed were excluded. The library also carries findings that failed later replication and laws that were passed but never applied, marked as such. The last full review was in September 2026; to report an error or an outdated source, write to the Observatory.