A survey result becomes dangerous when it arrives without the people inside it.
"Thirty-eight per cent" can look like a clean fact on a slide. It does not show the eighteen-year-old whose social life, study, entertainment and first experiments with generative AI all pass through the same screen. It does not show the older worker who sees AI as distant from daily life, or the person who says digital tools strengthen their closest relationships while also feeling exhausted by the hours spent inside them.
The OECD's 2025 policy brief How do people experience new technologies and generative AI? is useful precisely because it does not describe one digital public.
It reports a cross-country poll of 14,611 people in fourteen countries, collected in early 2025 through the OECD Digital Well-being Hub in collaboration with Cisco. The sample spans Australia, Brazil, Canada, France, Germany, India, Italy, Japan, Korea, Mexico, the Netherlands, South Africa, the United Kingdom and the United States.
The report is a snapshot of experience and perception. It is not a causal map of what technology does to human beings.
That boundary should remain visible throughout.
What The Poll Found
Age is one of the strongest dividing lines in the report.
Younger adults report the highest use of social networking and generative AI. Remote work is concentrated more strongly among adults in prime working ages. Generative AI adoption is particularly high among adults aged eighteen to thirty-five, with substantial differences between countries. The report describes higher engagement in India, Brazil and South Africa and lower reported adoption in countries including Germany, France, Italy and Japan.
Screen time is also uneven. About 38 per cent of respondents reported more than five hours of daily recreational screen use, with the highest proportion among younger adults. That number should not be read as a diagnosis. "Screen time" combines activities with very different meanings, and self-report has limits. But the scale is large enough to make digital experience a serious well-being question rather than a niche concern.
Relationships complicate the picture further. Thirty-nine per cent of respondents said digital tools had strengthened relationships with those closest to them. Younger respondents were more positive about this than older respondents. Technology therefore appears neither simply isolating nor simply connecting. Its relationship to human well-being varies with age, purpose, context and what people are actually doing through the medium.
The same complexity appears around generative AI.
More than three quarters of respondents under thirty-five regarded AI as useful. Younger people also reported higher trust and ethical confidence than older people, though strong confidence was not universal. AI-related training was most common among younger working-age groups and among people with tertiary education. The OECD reports 36 per cent participation in AI-related learning among people with tertiary education compared with 18 per cent among those whose highest attainment was upper secondary education.
That is not only an adoption story. It is a capacity-distribution story.
The Unequal Distribution Of Readiness
When a new technology enters society, access is only the first inequality.
There is also inequality in time, confidence, training, institutional support, language, occupational exposure and the ability to judge what the tool is doing. The OECD data show that training uptake is not evenly distributed. Younger and more highly educated groups report more AI-related learning. If this pattern persists, the people most exposed to technological change may not be the same people receiving the strongest opportunities to interpret it.
This is where the report becomes relevant to Inner Technology.
The Institute's concern is not whether everyone should use generative AI. It is whether people have sufficient human and institutional capacity to remain agents around systems that increasingly generate, recommend, classify and persuade.
The OECD brief does not measure discernment, embodied judgment, agency or attention directly. But it shows the environment in which those capacities now operate: high digital exposure, uneven AI adoption, demographic differences in trust and unequal participation in training.
The bridge from those findings to an Inner Technology framework is synthesis, not an OECD conclusion.
A Poll Of Experience Is Not A Trial Of Effects
The OECD is unusually explicit that the findings should be interpreted with caution because of the nature of the poll.
That matters.
A cross-sectional poll can describe associations and reported experience. It cannot establish that high screen time caused tiredness, that generative AI caused changes in trust, or that AI training caused better career outcomes. People who use more technology differ from people who use less technology in many ways that the poll cannot fully control.
Self-report introduces further uncertainty. Respondents estimate their own screen time, describe emotional responses and rate usefulness, trust and anticipated career effects. These are meaningful data because lived experience matters. They are not the same as behavioral logs, clinical measures, longitudinal trajectories or experimentally identified effects.
Even the phrase "nationally representative" needs careful handling. Representation improves the usefulness of a sample, but it does not remove measurement error, response bias, question-framing effects or the limitations of a twenty-question poll.
A strong research brief does not weaken a study by naming these limits. It makes the study easier to use correctly.
Digital Well-being Is Not One Outcome
One of the report's strengths is its refusal to treat technology exposure as a single good or bad quantity.
The OECD well-being approach considers health, social connections, civic life, work-life balance and other dimensions. Its earlier literature review on digital technologies and well-being also emphasizes mixed outcomes. Digital tools can increase access, coordination and social connection while also introducing risks such as sedentary behavior, fatigue, harassment, compulsive use or unequal access.
This multidimensional view is important because public debate often collapses into two crude positions: technology is harming us, or technology is empowering us.
Both can be true for different people, at different times, through different designs.
A video call with a distant parent is screen time. So is compulsive scrolling at two in the morning. A generative AI system can widen access to language support. It can also produce a convincing answer that a user lacks the knowledge to verify. A digital platform can create community and train comparison in the same afternoon.
The unit of analysis has to become more precise.
From Exposure To Capacity
For the Institute, the most useful question opened by the OECD data is not "How many hours are people online?"
It is "What capacities do people need inside different forms of digital exposure?"
A high-use environment increases the importance of attentional control, but attention is not the only issue. People need metacognitive awareness to notice when fluency is being mistaken for certainty. They need digital competence to understand provenance, privacy and system limitations. They need relational capacity to distinguish connection from availability. They need agency to choose when a tool is serving a purpose and when the purpose has quietly reorganized itself around the tool.
These are hypotheses about developmental need. The OECD poll does not test them directly.
The report can therefore function as environmental evidence: it shows patterns of use, perception and training that make further human-capacity research urgent.
What The Evidence Does Not Show
The OECD poll does not show that more screen time necessarily causes lower well-being. It does not establish that young adults who trust AI more understand it better. It does not show that AI-related training is high quality, that training improves discernment, or that people with tertiary education are intrinsically more capable of responsible AI use.
It does not support a universal story about "the digital generation." Younger respondents are heterogeneous, countries differ substantially, and the fourteen-country sample is not the world.
It also does not contain Nordic countries. France, Germany, Italy, the Netherlands and the United Kingdom provide European comparison, but Sweden, Denmark, Norway, Finland and Iceland were not part of the reported sample. Nordic implications therefore require other evidence.
Most importantly for this Library, the brief does not validate the category of Inner Technology or the Institute's Human Readiness Model. Those are interpretive frameworks brought to the data.
Why This Matters In Northern Europe
Northern European countries often combine high digitalization with strong public institutions and ambitious digital-policy agendas. That makes questions of well-being, competence and human agency especially relevant, but the OECD poll cannot be localized to the Nordic region by assumption.
The appropriate use is comparative. European institutions can ask whether the age and education gradients seen in the OECD sample appear in national datasets. They can examine whether AI training reaches workers outside professional and highly educated groups. They can distinguish access from capability and capability from confidence.
The OECD framework also encourages policymakers to examine the purpose and quality of digital engagement rather than treating connectivity as an outcome in itself.
That orientation fits a Northern European public-interest tradition without proving any particular policy choice.
The Human Being Behind The Dashboard
The most important contribution of the Digital Well-being Hub may be methodological rather than dramatic.
It asks institutions to include human experience in the measurement of digital transformation.
That sounds modest. It is not.
Technology policy has often counted infrastructure, adoption, productivity and access more readily than fatigue, trust, relationship, agency or the felt quality of interaction. The OECD work does not solve the measurement problem, but it makes the omission harder to defend.
The person behind the percentage is still sitting somewhere with a body, a social world, a working day and a finite field of attention.
A serious digital future has to be able to measure systems without losing sight of that person.
What Should Be Measured Next?
The poll is most useful when treated as the beginning of a measurement programme rather than a verdict on digital life.
Future research could separate quantity of use from quality of use more sharply. Five hours of recreational screen time can include social connection, creative production, passive consumption, gaming, news, pornography, exercise guidance, language learning and background video. These activities have different relationships to sleep, mood, movement, attention and belonging. Duration alone is a blunt exposure variable.
Longitudinal data would also help distinguish selection from effect. Do people who are already tired use certain technologies more, does the technology contribute to tiredness, or do both respond to a third factor such as work conditions? Cross-sectional reports cannot resolve that sequence.
Generative AI introduces additional measurement needs. Frequency of use says little about delegation depth. One person may use a model for translation while retaining full authorship and verification. Another may delegate synthesis, judgment and interpersonal communication. Both count as users, but the developmental implications differ.
The quality of AI training also deserves attention. A twenty-minute compliance module and a sustained programme in verification, provenance, task design and ethical judgment should not be treated as equivalent simply because both respondents report "AI-related training." Measures need enough granularity to distinguish exposure from capability development.
Finally, perceived trust should be compared with calibration. The ideal is not maximal trust or maximal suspicion. It is appropriate trust: confidence that changes with system reliability, task risk and available evidence. A person who distrusts every AI system is not necessarily more discerning than a person who trusts every one.
These distinctions matter if the Digital Well-being Hub is to inform policy rather than merely describe sentiment.
Why Heterogeneity Is The Finding
Policy summaries naturally look for averages. The deeper lesson in the OECD data may be variation.
Age, education, country and activity change the pattern. That means a universal intervention is unlikely to fit. Older adults may need different forms of support from younger adults. Workers in routine occupations may encounter AI through employer systems rather than personal experimentation. People with less formal education may face a training gap while also having less discretion to choose which tools enter their work.
For human-capacity research, heterogeneity should therefore be designed into the question. Not "Does AI affect attention?" but "Which uses, for which people, in which settings, through which mechanisms, with what consequences?"
That sentence is less dramatic than a headline. It is also closer to the evidence.
Research references
- OECD. How do people experience new technologies and generative AI? Insights from a few countries worldwide. OECD Policy Insights on Well-being, Inclusion and Equal Opportunity No. 23. 4 December 2025. https://doi.org/10.1787/49b8d10e-en
- OECD. Full report: How do people experience new technologies and generative AI? https://www.oecd.org/en/publications/how-do-people-experience-new-technologies-and-generative-ai_49b8d10e-en/full-report.html
- Lee S, Zarnic Z. The impact of digital technologies on well-being. OECD Papers on Well-being and Inequalities No. 29. 2024. https://doi.org/10.1787/cb173652-en
- OECD. How's Life in the Digital Age? 2019. https://doi.org/10.1787/9789264311800-en
- European Commission, Joint Research Centre. DigComp 3.0. 2025. https://doi.org/10.2760/0001149