Artificial intelligence and socioeconomic forces: transforming the landscape of religion

In the current global context, where artificial intelligence (AI) is becoming a pivotal part of everyday life, its influence on societal norms and individual liberties—particularly religious freedom—is a significant concern. Prior research highlights AI’s transformative potential in reshaping social behaviors and norms, including those associated with religious practices and freedoms. Rapid integration of AI in surveillance, communication, and data processing raises concerns about implications for religious expression and freedom, including risks from algorithmic bias and invasive surveillance that may restrict practices or discriminate against certain religious groups. Policy and governance challenges underscore how AI affects government policy and societal norms that can support or undermine religious freedom across contexts. Aligning AI development with protection and respect for religious freedom requires coordinated efforts among technologists, policymakers, religious leaders, and civil society to set ethical standards and frameworks that reinforce freedom, tolerance, and respect.

Study objective: Evaluate the impacts of AI and socioeconomic factors on religious freedom in 20 countries from 2000 to 2022 using a model with country and year specific effects. Main expectation: AI negatively affects religious freedom; economic growth, political stability, and education positively relate to religious freedom; digitalization may adversely affect it. System generalized method of moments (GMM) is used for robustness.

Contributions: (1) Provides empirical evidence that AI negatively affects religious freedom, offering a counterpoint to broadly optimistic views of AI’s social welfare impacts. (2) Identifies positive associations between religious freedom and economic development, political stability, and education, focusing specifically on religious liberty within the broader freedoms literature. (3) Introduces methodological rigor by employing system GMM to address endogeneity and unobserved heterogeneity beyond traditional regressions.

Organization: Section 2 reviews literature; Section 3 describes variables and models; Section 4 presents results; Section 5 concludes with insights and recommendations.

The literature on AI and religion is multifaceted, spanning practical, ethical, theological, and sociopolitical domains. Scholars note AI’s potential to augment religious practices by personalizing spiritual experiences, aiding interpretation of sacred texts, and broadening dissemination of teachings, potentially democratizing access to spiritual knowledge. Countervailing concerns highlight risks of oversimplifying spirituality, eroding communal aspects of worship, and creating AI-driven religious experiences that may compromise authenticity. Theological and philosophical debates probe AI’s implications for free will, consciousness, and the soul, challenging established doctrines. Ethical discussions stress developing AI with sensitivity to religious diversity, mitigating bias, stereotypes, and respecting cultural contexts through appropriate guidelines and frameworks.

Beyond AI–religion intersections, broader socioeconomic forces shape religious landscapes. Economic development can correlate with secularization as material values rise, yet some advanced societies exhibit religious resurgence driven by quests for meaning. Political stability can foster environments supportive of religious freedom and diversity, but in some regimes it may accompany state control over religious institutions. Education can lead both to questioning organized religion and to more nuanced understanding of doctrines and practices. Digitization revolutionizes access to religious information and participation, while also facilitating misinformation and superficial online experiences. Overall, interactions among economic development, political stability, education, and digitization are context-dependent and jointly influence religious beliefs, practices, and freedoms.

Variables:

• Dependent variable: Religious freedom, defined as the liberty to hold, not hold, change, express, and practice religious beliefs individually or communally, privately or publicly. This construct is motivated by prior work emphasizing legal, societal, and institutional dimensions of religious liberty and its role in pluralism and tolerance.
• Independent variable (AI): AI development/proliferation proxied by the number of AI patents, consistent with literature linking patent counts to technological progress, geographic clustering, sectoral adoption, temporal trends, and investment relevance.
• Control variables (socioeconomic factors): Economic development (GDP), political stability, education level, and degree of digitalization, each theoretically linked to both AI diffusion/management and religious freedom outcomes.

Data and transformations:

• Panel of 20 countries, annual data from 2000–2022. All data sourced from the World Bank. All variables are log-transformed to mitigate heteroscedasticity and stabilize variance.

Baseline model:

• Two-way fixed-effects panel specification with country and year effects: r_it = α0 + α1 a_it + α2 ec_it + α3 po_it + α4 ed_it + α5 di_t + η_i + μ_t + ε_it where r_it is religious freedom; a_it AI; ec_it GDP; po_it political stability; ed_it education level; di_t digitalization; η_i country fixed effects; μ_t year fixed effects; ε_it error term assumed white noise.
• Rationale: Country fixed effects capture time-invariant unobservables (e.g., culture, legal frameworks); year fixed effects capture global shocks and common trends.

Model selection and estimation strategy:

• Compared pooled OLS, panel OLS, country-FE, year-FE, and two-way FE models. Chow and Hausman tests guided rejection of pooled OLS and selection of the two-way FE model to address unobserved heterogeneity in both dimensions.

Robustness: System GMM

• To address potential endogeneity (simultaneity, omitted variables, measurement error) and dynamics, employed system GMM (Arellano–Bond; Hansen). The specification includes a lagged dependent variable (autoregressive term) and treats potentially endogenous regressors with internal instruments (lagged levels and differences). Years are treated as strictly exogenous instruments to preserve first-difference properties.
• Implementation details: Use of Helmert (forward orthogonal deviation) transformations to handle fixed effects and optimize instrument structure; focus on identification, simultaneity, and exclusion restrictions. Validity of instruments assessed via Difference-in-Hansen tests for overidentifying restrictions and exogeneity of instrument subsets; standard AR(1)/AR(2) tests for serial correlation in first differences. System GMM estimates are compared with FE results for consistency in signs and significance.

Interpretation framework:

• AI may affect religious freedom through data processing, surveillance, and platform governance (e.g., hate speech monitoring, content moderation, state surveillance). Controls parse socioeconomic channels that co-move with religious liberties.

• Model selection: Chow tests reject pooled OLS; Hausman tests support fixed-effects over random-effects approaches. Two-way fixed-effects (country and year) model (Model 5) chosen as primary.
• Main AI effect: In Model 5, AI is negatively associated with religious freedom. A 1% increase in AI (patents) correlates with approximately a 0.011% decrease in religious freedom (coefficient −0.011, t ≈ −2.095, significant at 5%). Across alternative models (1–4), the AI coefficient remains negative and statistically significant at the 5% level, supporting robustness of direction and significance.
• Socioeconomic controls (Model 5): • Economic development (GDP): Positive and highly significant (β ≈ 0.144, t ≈ 3.875, p < 0.01). • Political stability: Positive, marginal significance (β ≈ 0.354, t ≈ 1.635, ~10%). • Education: Positive and significant (β ≈ 0.205, t ≈ 2.032, p < 0.05). • Digitalization: Negative, marginal significance (β ≈ −0.061, t ≈ −1.589, ~10%).
• Goodness of fit and tests (Model 5): R² ≈ 0.257. The suite of FE models (1–4) provides consistent sign patterns, acting as internal robustness checks.
• System GMM robustness: Dynamic panel estimates yield coefficients broadly consistent in sign and significance with FE results. Post-estimation diagnostics (including Difference-in-Hansen tests) support instrument validity; minor variations in magnitudes do not alter conclusions.
• Substantive interpretation: AI expansion is associated with modest but statistically meaningful reductions in religious freedom, consistent with channels involving surveillance intensification, content moderation/suppression of religious content, and potential bias/discrimination. Economic development, political stability, and education generally support religious freedom, whereas higher digitalization correlates with reductions—potentially via enhanced monitoring and control capabilities.

The study distinguishes direct restrictions on religious liberty from broader AI-driven surveillance effects, situating religious freedom within the wider ecosystem of civil liberties. The negative association between AI and religious freedom aligns with technological determinism and information control arguments: expanding AI capabilities can facilitate surveillance and platform moderation that chill or constrain religious practice and expression, and can embed biases against religious groups in public and private systems. Conversely, higher economic development, political stability, and education correlate with greater religious freedom, reflecting modernization, stable governance, and tolerance mechanisms. Digitalization’s negative correlation suggests that pervasive digital infrastructures can enable monitoring and regulation that impinges on religious practices. Overall, the findings address the research question by showing AI’s expansion is linked to diminished religious freedom, moderated by socioeconomic environments. The results underscore the need for ethical, legal, and design safeguards so that AI deployment enhances rather than erodes fundamental rights, and they position religious freedom within an interdependent set of civil liberties affected by rapid technological change.

The paper demonstrates that AI growth is associated with decreases in religious freedom across 20 countries (2000–2022), while economic development, political stability, and education are positively associated, and digitalization is negatively associated, with religious freedom. Two-way fixed-effects estimates and system GMM robustness analyses converge on these findings, indicating an intricate relationship shaped by socioeconomic contexts.

Policy implications:

• Establish comprehensive AI regulatory frameworks that explicitly safeguard religious freedom, including mechanisms to address violations arising from AI applications.
• Balance digital transformation with stronger privacy and surveillance laws to prevent misuse of digital tools against religious practices or groups.
• Promote economic growth and political stability as supportive conditions for religious liberty.
• Invest in education, including curricula that foster tolerance and literacy about religious freedom and the societal impacts of AI.

Future research directions:

• Expand country coverage to capture broader regional and cultural variation.
• Enhance measurement strategies for religious freedom and AI influence; complement quantitative analysis with qualitative case studies and interviews to illuminate mechanisms.
• Incorporate additional confounders (e.g., cultural variables; specific AI technologies and use-cases) and explore heterogeneous effects.

• External validity: Analysis covers 20 countries; results may not generalize globally across diverse cultural and political contexts.
• Measurement constraints: Proxies for religious freedom and AI (patents) may not capture all dimensions; cross-country and temporal data quality differences may affect estimates.
• Omitted variables: Despite controls, unobserved factors may remain. Future work should explore additional cultural, legal, and technological covariates and adopt mixed methods (qualitative case studies, interviews) to deepen mechanism understanding.