Preprimary education is central to the global education agenda, including Sustainable Development Goal 4.2. Despite global momentum in low- and middle-income countries, cross-national evidence on long-term outcomes associated with preprimary education legislation remains limited. Here we applied a matrix-completion approach with subgroup and sensitivity analyses using 495 nationally representative surveys from 1990 to 2024, covering nearly 6.7 million women across 117 countries. Eligibility under the legislation was associated with higher educational enrolment and completion, and with more gender-egalitarian attitudes, more favourable empowerment-related behaviours and greater reproductive autonomy. For example, eligibility was associated with higher female primary-school enrolment (1.21 percentage points (pp); 95% confidence interval (CI) 0.52 to 1.91; P < 0.001) and completion (0.75 pp; 95% CI 0.24 to 1.25; P = 0.004), greater disapproval of wife-beating (1.04 pp; 95% CI 0.53 to 1.56; P < 0.001), greater participation in decisions about husbands’ earnings (3.25 pp; 95% CI 2.50 to 4.00; P < 0.001) and greater ever use of contraception (2.96 pp; 95% CI 2.33 to 3.59; P < 0.001). Eligibility under the legislation was also associated with lower rates of adolescent childbearing (−2.41 pp; 95% CI −2.67 to −2.14; P < 0.001) and child marriage (−0.43 pp; 95% CI −0.71 to −0.15; P = 0.003). These findings provide cross-national evidence directly relevant to ongoing policy discussions surrounding Sustainable Development Goal 4.2 and broader global development agendas, while emphasizing the importance of implementation capacity and equitable system expansion. Lou et al. find that eligibility for preprimary education is linked to more gender-equitable attitudes, greater empowerment and reproductive autonomy, and lower levels of adolescent marriage and childbearing across low- and middle-income countries.
Prominent employers are increasingly dropping college degree requirements when hiring. Does this provide upward mobility for non-college workers? Matching job postings to 446,500 job starts in US administrative data from 2011 to 2021, we show that employers reduced the share of postings requiring a degree by more than 10%. Non-college workers who move to decredentialed jobs earn 12 log points more than matched workers moving to never-credentialed jobs (β = 0.124; 95% confidence interval, (0.042, 0.206)), around US$4,600 per year—a premium that endures for at least five years. These hires are more likely to be Black, Hispanic and female than the workers they replace. Employers also benefit, as removing degree requirements reduces labour costs. Yet most employers that drop requirements do not hire non-college workers (non-college hiring rises 1.2 percentage points; 95% confidence interval, (0.5, 2.0)). Decredentialing is a promising tool for social mobility, but only when employers follow through. Linking job postings to US Census earnings records from 2011 to 2021, Lyttelton et al. find that non-college workers hired into jobs that dropped degree requirements earned around US$4,600 more per year, yet employers rarely hired non-college workers after decredentialing.
People will often trust or distrust an artificial intelligence (AI) output because of how it makes them feel, rather than how reliable it is. AI governance frameworks should require that platforms disclose design choices that target user emotions, allow users control over these, and include features aimed to calibrate trust.
Large language models (LLMs) are transforming computational social science. This Review examines how LLMs now serve four distinct research roles: subjects that simulate human behaviour, assistants that automate analysis, collaborators that generate hypotheses and evaluators that audit and improve research quality. We analyse recent advances across sociology, political science, psychology, anthropology, economics and related fields, revealing how LLMs accelerate discovery while creating new methodological paradigms. This transformation arrives as society itself becomes saturated with artificial intelligence agents—making LLMs both our research tools and authentic research subjects. Yet fundamental challenges persist. LLMs struggle with causal inference and raise urgent ethical concerns about privacy and misinformation, and their ability to reproduce social complexity remains uneven and unproven. We identify critical research directions for harnessing LLMs’ potential while addressing their limitations. As artificial intelligence reshapes human society, understanding this transformation requires the very tools driving it—a recursive challenge that will define the future of social science. This Review from Evans et al. documents the distinct roles that LLMs now have in computational social science and provides an overview of recent LLM-supported advances across fields including economics, psychology, political science, sociology and anthropology.
As artificial intelligence (AI) becomes increasingly embedded in daily life, ascertaining whether an agent is human is critical. Here we systematically benchmark AI’s ability to imitate humans in three language tasks (image captioning, word association and conversation) and three vision tasks (colour estimation, object detection and attention prediction), collecting data from n = 636 humans and 37 AI agents. We also conducted 72,191 Turing-like tests with 1,916 human judges and 10 AI judges. Current AIs are approaching the ability to convincingly impersonate humans and deceive human judges in both language and vision. Even simple AI judges outperformed humans in distinguishing AI from human responses. Imitation ability showed minimal correlation with conventional AI performance metrics, suggesting that passing as human is an important independent evaluation criterion. The large-scale Turing-like datasets and metrics introduced here offer valuable benchmarks for assessing human-likeness in AI and highlight the importance of rigorous, quantitative imitation tests for AI development. Zhang et al. systematically compare humans and AI across three language and three vision tasks.
