Society
Technology and society, honestly assessed
Attitudes are measured shakily, the labour-market evidence describes robots rather than AI, and every forward-looking number in this subject is a model output.
The strongest evidence about technology's effects on society describes the past; the loudest claims describe the future. Attitudes to AI are measured by online panels that skew hardest in exactly the countries where the results look most striking. The best labour-market evidence — automation explains between half and seven-tenths of the change in the US wage structure — covers 1980 to 2016 and describes robots and software, not generative AI. Estimates of AI's productivity effect differ by an order of magnitude. On long-run social effects nothing is known: these technologies are two to four years old. For a workplace-oriented example of technology measurement in practice, see workforce optimization software.
Global AI optimism is real, small in movement and shakily sampled
The best multi-country series shows a modest positive shift: 55% of people globally said in 2024 that AI products have more benefits than drawbacks, up from 52% in 2022, with 18 of 26 countries surveyed twice becoming more positive (Ipsos, reported in Stanford HAI, AI Index Report 2025). "Global" there means 23,685 adults across 32 countries recruited through online panels, so it is not a population estimate.
The same respondents report unease alongside the optimism: 54% globally agreed that AI products make them nervous, and the share trusting AI companies with personal data moved from 50% in 2023 to 47% in 2024 — three points in panel data, within ordinary noise. The previously most sceptical countries gained most: France 31% to 41%, Germany 37% to 47% and Great Britain 38% to 46% between 2022 and 2024 (Ipsos, in AI Index 2025). Two points two years apart is weak evidence of a trend.
The country-level spread from that survey is the most quoted finding here and the least trustworthy.
83% of people in China, 80% in Indonesia and 77% in Thailand see more benefits than drawbacks from AI, against 36% in the Netherlands, 39% in the United States and 40% in Canada.
Direction: Cuts both ways. Strength of evidence: Mixed.
Caveat Online-panel bias is far worse where internet penetration is low, so the high-optimism figures are the least reliable in the set; question translation and response styles also differ.
An online panel can only recruit people who are online. Where much of the population has no reliable connection, it reaches an urban, educated, higher-income minority — those most likely to benefit from new technology. In the Netherlands or Canada it reaches nearly everyone. A chart placing 83% next to 36% is therefore partly measuring who is still offline rather than what nations think, and it is routinely printed without that caveat.
The public and the people who build AI disagree sharply
The strongest attitude data here is national, because it uses a probability panel.
51% of US adults say they are more concerned than excited about increased use of AI in daily life, against just 11% of AI experts.
Direction: Cuts both ways. Strength of evidence: Strong.
Caveat The "experts" are a self-selected group with direct career and financial stakes in AI, not a representative expert population.
The gap runs through every forward-looking question: 56% of experts think AI will have a positive effect on the US over 20 years against 17% of the public, and on jobs 73% against 23%. The study offers no evidence that experts forecast outcomes more accurately, and they are paid by the industry whose effects they grade. Both fear too little regulation rather than too much: about 60% of adults and 56% of experts worry government will not go far enough (Pew Research Center, 2025).
Commercial surveys find the same shape with worse provenance: a 47-country study funded by KPMG, which sells AI advisory services, puts willingness to trust AI systems at 46% (Gillespie and colleagues, 2025).
The labour-market evidence is strong and about the last technology
The best-identified finding here concerns automation between 1980 and 2016.
Between 50% and 70% of the change in the US wage structure over four decades is accounted for by relative wage declines among worker groups specialised in routine tasks in automating industries.
Direction: Increase. Strength of evidence: Strong.
Caveat A structural model estimate rather than an experiment; the range depends on the task-displacement measure, and competing explanations are argued against rather than ruled out.
That finding is about robots and enterprise software; applying it to generative AI is an analogy, not an inference. The estimates that do concern AI are models, and the serious ones sit below industry forecasts.
Generative AI is estimated to raise total factor productivity by no more than about 0.66% over ten years, and under 0.53% once harder-to-learn tasks are accounted for.
Direction: Increase. Strength of evidence: Mixed.
Caveat A deliberately conservative projection rather than a measurement, extrapolating from micro-studies of simple tasks; the industry forecasts it argues against are projections too.
The same model expects AI to widen the gap between capital and labour income and finds no evidence it will narrow wage inequality (Acemoglu, 2024) — a modelled implication, not an outcome. Anyone reading a number about how many jobs automation will take, or any technology forecast, should ask whether it measures exposure or displacement.
The most-cited trust source is the weakest one
"Trust in technology is collapsing" is the shakiest claim here, and its usual source explains why.
Only 32% believe the next generation will be better off, and the 2025 edition reported 61% globally with a moderate or high sense of grievance.
Direction: Decrease. Strength of evidence: Weak.
Caveat Conducted and published by a public relations firm that sells reputation management to the companies whose trustworthiness it measures — a direct conflict of interest — using non-probability panels with limited published methodology.
A PR firm grading an industry it sells reputation services to is not a neutral instrument, and it publishes too little methodology to assess comparability across years or countries. It is quoted constantly in coverage of public feeling about platforms because it is annual, global and free — publication qualities, not evidential ones.
The short version
- 55% of people globally said in 2024 that AI has more benefits than drawbacks, up three points on 2022, from panels that are not population samples (Ipsos, in AI Index 2025).
- Reported AI optimism runs from 83% in China to 36% in the Netherlands, but panel bias makes the high figures the least reliable.
- 51% of US adults are more concerned than excited about AI, against 11% of AI experts, and both groups fear too little regulation (Pew Research Center, 2025).
- Automation accounts for 50% to 70% of the change in the US wage structure from 1980 to 2016, a finding about robots rather than generative AI (Acemoglu and Restrepo, 2022).
- The long-run effects of current AI systems are unmeasured rather than contested, and the most-quoted trust series comes from a PR firm with a stake in the answer.