The original GPS, conducted in 2012 as part of the Gallup World Poll, gathered responses from more than 80,000 people across 76 countries representing roughly 90 percent of the world’s population and economic output. It measures six experimentally validated preference domains that form the backbone of modern economic theory: risk-taking, patience, positive reciprocity, negative reciprocity, altruism and trust. These are not abstract curiosities. Patience predicts saving behaviour, educational attainment and entrepreneurial entry; trust correlates with financial development and growth; reciprocity underpins contract enforcement and workplace cooperation. If a language model could reproduce this preference landscape reliably, it could in principle generate counterfactual data for policy analysis, easing one of the deepest problems in causal inference — that we can never observe what would have happened under conditions that never occurred.

The research team, led by Fadong Chen and Qijun Wang of Zhejiang University together with Sudeep Bhatia of the University of Pennsylvania, took an unusually careful approach. Rather than asking models to answer surveys in the abstract, they built one-to-one ‘digital twins’: for each of 67,201 matched human respondents across 74 countries, they constructed a persona embedding that person’s age, gender, cognitive ability and country, then prompted each model to answer the identical GPS questionnaire. Crucially, the prompts were delivered in the respondents’ original survey languages, with payoffs denominated in local currencies — a Chinese respondent’s donation question, for instance, was framed in RMB. Twelve main model configurations, spanning OpenAI’s GPT family, Google’s Gemini models and DeepSeek-V3, generated roughly 27.8 million answers across about 1.39 million persona completions.

The headline finding is one of weak but genuine alignment. Pooling all countries, 63 of 72 correlations between model-predicted and human preferences — that is, 87.5 percent of the six preference domains across twelve configurations — were significantly positive. But the magnitude tells a different story. After normalizing both human and synthetic preference scores, the mean squared error between them averaged 1.827, nearly twice the variance of the human benchmark itself. Regression analyses showed that LLM-predicted preferences explained just 1.3 percent of the variance in individual human responses and, even after aggregating to country averages, only 4.2 percent of country-level variance. In plain terms, knowing what the AI ‘thinks’ a person would say tells you almost nothing about what that person actually said.

Perhaps the most striking pattern is that the models agree with each other far more than they agree with humans. Between-model mean squared errors were substantially lower than model-to-human errors at both the individual and country levels. This convergence suggests that today’s language models share a common output regularity — a kind of statistical accent — that systematically departs from the human benchmark. Notably, newer and larger models did not outperform older ones: GPT-5.4 showed no measurable advantage over GPT-3.5-turbo, and Gemini-2.5-Flash did not beat Gemini-2.0-Flash. Scaling alone, at least across these generations, does not appear to close the human-machine gap in economic preferences.

The audit then probed whether models reproduce the well-documented demographic gradients in preferences — differences by gender, age and cognitive ability. Here the results were more encouraging. Using a confidence-interval-based consistency criterion, models recovered 77.8 percent of the demographic patterns found in the GPS data, with cognitive-ability gradients matched in 84.7 percent of cases and gender gradients in 81.9 percent. Risk-taking and patience gradients were recovered perfectly across all twelve configurations. But cultural variation proved far harder. When the team tested whether models captured differences along fifteen cultural dimensions drawn from the World Values Survey, Hofstede’s framework and Schwartz’s value orientations, the consistency ratio collapsed to 48.1 percent — statistically indistinguishable from a coin flip. Only three of twelve models recovered the link between Hofstede’s long-term orientation and patience, and just one recovered the relationship between uncertainty avoidance and risk-taking.

Why do models handle demographics but stumble on culture? Ablation experiments, in which pieces of the persona prompt were systematically removed, provided a diagnostic answer. Stripping demographic information degraded alignment substantially, while removing the country label barely changed the results at all — individual-level error was 1.838 for full personas versus 1.961 when only country information was retained. The models, it seems, condition effectively on traits like age and gender but largely ignore the country tag, treating a persona in Nairobi much like one in Oslo. Demographic effect sizes in the ablation conditions fell from an average of 0.303 to nearly zero when demographic cues were absent, confirming that the recovered gradients depend on explicit prompt information rather than deep cultural knowledge.

The residual errors were not randomly distributed, either. The team constructed an ‘inconsistency index’ — the sampling-weighted within-country mean of individual prediction errors — and correlated it with national characteristics. In many models, prediction errors were significantly larger in countries with lower GDP per capita and weaker digital infrastructure, as measured by the ICT Development Index, while population size and internet-user counts showed no systematic relationship. The authors suggest two mechanisms: languages and content from wealthier countries receive disproportionate emphasis in model development, and digitized economies generate more of the high-quality structured data on which models train. The equity implications are uncomfortable — AI tools may simply work worse for the populations that stand to benefit most from careful policy analysis.

Practical replication tests drove the point home. When the researchers attempted to reproduce 35 published country-level findings from the GPS literature — such as the association between patience and GDP per capita, or gender gaps in preferences and gender equality — the best-performing model reproduced only 16, or 45.7 percent, with the same direction and statistical significance. Two additional biases emerged: women’s prosocial preferences were systematically overestimated across virtually all models, with the distortion largest in countries with stronger gender stereotypes, and model responses were far more concentrated than human ones, a homogenization that raising the sampling temperature only partially repaired.

The authors are careful not to declare LLM simulation useless. Synthetic respondents may still serve as tools for early-stage hypothesis generation, study-design diagnostics, robustness checks and ethically sensitive counterfactual exploration before human experiments begin, and the one-to-one digital-twin protocol itself offers a reusable benchmark for tracking how cultural alignment evolves across model generations. But the core conclusion stands as a warning against a seductive shortcut. Language models do not yet encode the cultural texture of human economic life, and their failures are patterned along lines of global inequality. Until that changes, understanding what real people want — in Nairobi, Oslo or anywhere else — still requires asking them.

Subject of Research: Validity of large language model-simulated survey responses as proxies for human economic preferences across 74 countries

Article Title: Simulated responses from large language models are weak proxies for human economic preferences across 74 countries

Article References: Simulated responses from large language models are weak proxies for human economic preferences across 74 countries. (n.d.). https://doi.org/10.1038/s41562-026-02611-4

Image Credits: AI Generated

DOI: 10.1038/s41562-026-02611-4

Keywords: large language models, economic preferences, Global Preference Survey, survey methodology, cross-cultural alignment, digital twins, synthetic data, behavioral economics, AI bias, Nature Human Behaviour, policy evaluation, cultural diversity

The research team, led by Firew Bekele Ejigu with colleagues Tariku Sime Gutu and Wudu Melese Tarekegne, surveyed 342 primary school teachers selected through stratified random sampling across the city, complementing the questionnaire data with in-depth interviews and school observations involving six purposively chosen participants. The study employed an embedded mixed-methods design, giving greater weight to the quantitative strand while using qualitative data to contextualize and enrich the statistical findings. Addis Ababa, home to roughly 5.46 million people and often described as a microcosm of Ethiopia, provided an ideal setting: the city’s population includes ethnic Amhara (47 percent), Oromo (19.5 percent), Gurage (16.3 percent), Tigray (6.2 percent) and numerous other groups, alongside Orthodox Christian, Muslim and other religious communities.

The theoretical backbone of the study draws on James A. Banks’s influential framework of multicultural education, which identifies five interlocking dimensions: content integration, the knowledge construction process, equity pedagogy, prejudice reduction, and an empowering school culture. Crucially, Banks’s model insists that multicultural education is not merely a matter of revising textbooks or adding festival days to the calendar. It demands a transformation in teachers’ beliefs, attitudes and everyday instructional practices, because even the most carefully designed curriculum is only as effective as the educator who enacts it in the classroom.

Measuring attitudes required a purpose-built instrument. The researchers constructed a 20-item Likert-type questionnaire ranging from strongly disagree to strongly agree, with negatively worded items reverse-coded during analysis. One sample item probed a common anxiety: whether addressing student diversity in class is dangerous for national unity. Before deployment, the scale was vetted by language instructors and diversity education researchers, then piloted with 15 teachers outside the main sample. The resulting Cronbach’s alpha of 0.754 indicated acceptable internal consistency, and skewness and kurtosis statistics confirmed the scores were approximately normally distributed, clearing the way for parametric testing.

The headline quantitative result was a grand mean score of 3.41 on the five-point scale, with a standard deviation of 1.33. That figure places teachers squarely in the moderate range: above the scale’s midpoint, signaling generally favorable orientations toward weaving diversity into curriculum experiences, but far from uniformly enthusiastic. The variability matters. Some teachers expressed genuine comfort with multicultural content, while others reported discomfort or uncertainty, and the interviews revealed why the average conceals a spectrum of understanding.

Those qualitative conversations were revealing. One teacher described a school where children speak Welayitegna, Afan Oromo, Amharic, Tigrigna and other languages, and where she regularly coordinates cultural shows and Nation and Nationalities Day events. Another participant, a school principal, painted a more cautious picture, noting that while some teachers hold positive attitudes, others avoid discussing diversity openly for fear of being perceived as politically motivated. Several teachers conflated multicultural education with inclusive education for students with disabilities, or equated it narrowly with religion-free schooling, suggesting that conceptual clarity remains a work in progress even among committed educators.

The inferential statistics told a story about what does and does not predict these attitudes. Gender made no significant difference, with male teachers averaging 3.39 and female teachers 3.47, a gap that failed to reach statistical significance. Academic qualification likewise showed no significant relationship, despite degree holders posting the highest raw mean of 3.73. More surprising was the weak, non-significant correlation between the number of languages teachers spoke and their multicultural attitudes, a finding the authors interpret through the lens of how languages are acquired: community-based, everyday multilingualism may matter more than a simple count of languages known.

What did matter was training and career stage. Teachers who had completed multicultural coursework or professional development scored significantly higher, at 3.54 versus 3.34 for those without such exposure, a difference the researchers report as statistically significant. One-way ANOVA results further showed that younger teachers, aged 22 to 35, and those with one to ten years of experience held significantly stronger multicultural attitudes than their older, more seasoned colleagues. The authors caution against attributing this generational pattern to training alone, since the data did not capture when or where teachers studied; generational socialization around diversity values, or younger teachers’ closer cultural proximity to their students, could also play roles. Still, the pattern echoes findings from Norway’s TALIS data, where professional development emerged as a significant predictor of multicultural teaching self-efficacy.

The study’s most sobering statistic may be that 50.6 percent of participating teachers had never received any training in multicultural education. That gap mirrors a continental pattern. Comparable studies in Nigeria, South Africa and Tanzania have documented a recurring attitude-practice divide, in which teachers express rhetorical commitment to cultural diversity but lack the preparation, resources and institutional support to translate that commitment into classroom practice. The Ethiopian findings thus reinforce a growing scholarly consensus that the weakest links across African education systems are content integration, equity pedagogy and empowering school culture, precisely the dimensions that require deliberate, structured intervention.

The implications reach well beyond Addis Ababa. The researchers recommend embedding multicultural education as a core component of teacher preparation, both through dedicated standalone courses and through infusion across subject-specific methods training, alongside continuous in-service workshops and experience-sharing programs. They also call for systematic textbook analysis against Ethiopia’s new Education Sector Development Roadmap and curriculum framework. Because children begin forming evaluative categories about social difference at remarkably early ages, and because teachers serve as the role models from whom young learners internalize attitudes toward out-groups, the stakes of this professional development extend far beyond individual classrooms. In a nation whose founding educational aspiration is unity within diversity, the study suggests that the surest investment is not in policy documents but in the minds of the teachers who open them every morning.

Subject of Research: Primary school teachers' attitudes toward multicultural curriculum integration in Addis Ababa, Ethiopia

Article Title: Primary school teachers’ attitudes towards integrating multiculturalism into curriculum implementation in Addis Ababa city, Ethiopia

Article References: Ejigu, F. B., Gutu, T. S., & Tarekegne, W. M. (2026). Primary school teachers’ attitudes towards integrating multiculturalism into curriculum implementation in Addis Ababa city, Ethiopia. Discover Education, 5(1), Article 1073. https://doi.org/10.1007/s44217-026-02166-6

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02166-6

Keywords: multicultural education, teacher attitudes, Ethiopia, Addis Ababa, primary education, curriculum implementation, professional development, teacher training, cultural diversity, inclusive education, James A. Banks, mixed methods