Source-linked AI summary
Global AI Ethics: A Review of the Social Impacts and Ethical Implications of Artificial Intelligence
Alexa Hagerty, Igor Rubinov
TL;DR
The paper addresses globally U.S.-centered AI ethics research through a multilingual social-science review across five regions. It finds that AI’s social impacts and meanings vary by context, while AI systems can entrench inequality globally, motivating on-the-ground research.
Problem
Global AI ethics research is U.S.-centered and lacks evidence from regions outside the U.S. and Western Europe, limiting understanding of culturally varied AI impacts.
Method
Researchers reviewed more than 800 academic articles and monographs in over a dozen languages, alongside policy papers, government reports, and media from five global regions.
Results
AI’s social impacts and ethical meanings vary across geographical and cultural settings, while AI systems show a pattern of entrenching and amplifying social inequality globally.
Takeaways & Limitations
Rigorous, on-the-ground research is needed to identify inequality-amplifying AI systems and support responsible development, implementation, monitoring, and regulation.
Takeaways & Limitations
A trait-list approach to culture can oversimplify human experience and distract from practical socioeconomic factors shaping behavior.
Abstract
from arXiv · showhide
The ethical implications and social impacts of artificial intelligence have become topics of compelling interest to industry, researchers in academia, and the public. However, current analyses of AI in a global context are biased toward perspectives held in the U.S., and limited by a lack of research, especially outside the U.S. and Western Europe. This article summarizes the key findings of a literature review of recent social science scholarship on the social impacts of AI and related technologies in five global regions. Our team of social science researchers reviewed more than 800 academic journal articles and monographs in over a dozen languages. Our review of the literature suggests that AI is likely to have markedly different social impacts depending on geographical setting. Likewise, perceptions and understandings of AI are likely to be profoundly shaped by local cultural and social context. Recent research in U.S. settings demonstrates that AI-driven technologies have a pattern of entrenching social divides and exacerbating social inequality, particularly among historically-marginalized groups. Our literature review indicates that this pattern exists on a global scale, and suggests that low- and middle-income countries may be more vulnerable to the negative social impacts of AI and less likely to benefit from the attendant gains. We call for rigorous ethnographic research to better understand the social impacts of AI around the world. Global, on-the-ground research is particularly critical to identify AI systems that may amplify social inequality in order to mitigate potential harms. Deeper understanding of the social impacts of AI in diverse social settings is a necessary precursor to the development, implementation, and monitoring of responsible and beneficial AI technologies, and forms the basis for meaningful regulation of these technologies.
INTRODUCTION
Global AI ethics research remains limited and U.S.-centered, while the available literature indicates that AI’s social impacts vary by cultural setting and can amplify inequality worldwide.
- INTRODUCTION: Existing global AI analyses are biased toward U.S. perspectives and lack research outside the U.S. and Western Europe, limiting understanding of diverse social contexts.The authors identify regional differences as a major blind spot in AI ethics research.
- INTRODUCTION: The review finds that AI’s social impacts and interpretations are likely to differ markedly across geographical and cultural settings.Local social contexts shape how AI is understood and implemented.
- INTRODUCTION: The literature indicates that AI-driven technologies entrench social divides and exacerbate inequality, particularly among historically marginalized groups, on a global scale.Low- and middle-income countries may be especially vulnerable and less likely to share in AI’s benefits.
- INTRODUCTION: Rigorous, independent ethnographic research is needed to identify how AI affects societies and where systems may amplify inequality or cause harms.The authors present on-the-ground research as particularly important for understanding and mitigating these effects.
- INTRODUCTION: The project addresses a global research gap through a review of more than 800 publications across five regions, fourteen countries, and over a dozen languages.The review also considered policy papers, government reports, local and regional media, and adjacent technologies where direct AI research was scarce.
PART I: DEFINING THE TERMS
The paper treats AI as a technosocial system shaped by cultural context, participation, and data, while cautioning against simplistic cultural generalizations.
- PART I: DEFINING THE TERMS: AI is a technosocial system whose social values and assumptions shape its perception, design, use, and cultural interpretation.Technologies emerge from social visions and can also influence how societies imagine the future.
- PART I: DEFINING THE TERMS: Cultural imagination influences responses to AI and robotics, but its effects are not predictable: Japanese acceptance of some robots coexists with reluctance toward robotic eldercare.The contrast illustrates why cultural context should not be reduced to a single generalized pattern.
- PART I: DEFINING THE TERMS: Understanding AI across cultures therefore requires attention to who designs systems, whose data they contain, and which cultural assumptions their designs encode.The paper frames these questions as central to analyzing AI’s social context across development and use.
- PART I: DEFINING THE TERMS: Global inequality in participation excludes many people from AI design, with women in the Global South especially affected by gaps in education and specialized skills.These inequities raise questions about the long-term consequences of developing AI without their full participation.
- PART I: DEFINING THE TERMS: AI datasets reflect social histories, while thin digital footprints leave low- and middle-income populations underrepresented and potentially increasingly marginalized.The paper contrasts frequent household data generation in the United States with roughly zero digital data points for an average household in Mozambique, where about 90% lack internet access.
Implementation and Use
AI technologies enter existing social conditions, so their effects depend on how they are deployed and interpreted in particular contexts. Oversimplified cultural assumptions can obscure economic, political, and institutional factors shaping those effects.
- Implementation and Use: Deployment under existing social conditions can reinforce status quo inequalities even when the technology itself is not biased or flawed.In Saudi Arabia, a movement-alert app enforced male guardianship, while women used other technologies to resist it.
- Implementation and Use: Technologies used in complex social worlds can produce unintended consequences ranging from political liberation to repression.Their effects depend on the surrounding social and political conditions, not only on technical design.
- Beyond Simple Slogans: Cultural analysis should avoid treating ethnicity, nationality, or language as fixed traits because that approach stereotypes groups and simplifies human experience.Anthropological critiques reject claims such as “the Chinese believe this” or “the Americans believe that.”
- Beyond Simple Slogans: Attributing differences to culture can obscure other causes, including socioeconomic constraints that require different solutions.A Mexican immigrant missed clinic visits because he could not afford to miss work, not because of distinct cultural beliefs.
- Beyond Simple Slogans: A useful account of culture treats it as a dynamic repertoire of shared understandings inseparable from history, politics, and economics.This framing supports analysis of technology experiences alongside factors such as literacy, broadband access, and economic status.
1. Culture is shared
Culture is shared through explicit and implicit understandings, but it is neither homogeneous nor uniformly accepted. Internal differences across social groups and individuals make simple cultural generalizations unreliable.
- 1. Culture is shared: Culture includes explicit rules and public beliefs as well as implicit assumptions and common sense.Shared cultural experience does not require agreement, as debates over immigration and healthcare illustrate.
- 1. Culture is shared: Cultures contain diverse subcultures, social groups, and individual differences rather than forming seamless wholes.Processes can differ even among closely connected groups.
- 1. Culture is shared: Age, gender, class, education, religion, ethnicity, region, and personality can produce substantial variation within a national culture.The contrasting public figures used in the paper illustrate why nationality alone cannot define a homogeneous culture.
3. Culture is always changing
Culture changes over time and remains connected to historical, political, and economic structures. These dynamics shape how technologies are imagined, adopted, resisted, and ethically understood.
- 3. Culture is always changing: Cultural change occurs at different speeds, with entrenched inequalities and institutional representation changing more slowly than tastes or fashions.The paper contrasts rapidly changing preferences with persistent wealth, gender, and elite-representation disparities.
- 3. Culture is always changing: Historical legacies such as colonialism, apartheid, and political repression remain part of the contexts through which cultures and technologies are understood.These histories shape contemporary relationships and collaborations.
- 3. Culture is always changing: Technological interpretations depend partly on cultural background and personal experience, so the same practice or system can evoke safety, fear, normality, or scandal.The paper uses examples involving housing, food, clothing, and armed patrols to show this contextual variation.
- 3. Culture is always changing: Colonial histories can generate distrust and disputes over ownership, as shown by an African research partnership disrupted by foreign patents.The case raises concerns about intellectual-property appropriation relevant to AI development.
- 3. Culture is always changing: Cross-cultural ethics includes formal philosophical and religious frameworks as well as politics, history, law, customs, common sense, and individual experience.Because these elements are linked, the paper treats ethics and culture as inseparable.
- 3. Culture is always changing: Understanding ethical practice requires studying how people act within cultural contexts, making close, on-the-ground research vital.The paper concludes that AI is shaped by social context throughout development and use.
PART II: ASKING HARD QUESTIONS
AI principles do not translate cleanly across borders because value-laden terms carry different linguistic, cultural, and social meanings. These differences shape how concepts such as fairness and privacy are understood and can affect technology’s practical consequences.
- Implications: Meaningful global AI ethics therefore requires attention to regional differences rather than assuming that principles have a uniform interpretation.The article frames cultural translation as a central challenge for cross-border AI ethics.
- Linguistic and Cultural Translation: The translation of terms can produce practical consequences, as Brazilian indigenous land-rights activism was affected by how Facebook’s “like” button was rendered.The issue was not simply mistranslation: reluctance to “enjoy” negative events influenced algorithmic filtering.
- Linguistic and Cultural Translation: AI principles contain value-laden terms whose meanings and connotations vary across languages and cultures.Even accurate translation cannot eliminate culturally specific interpretations of concepts such as fairness and privacy.
- Privacy Across Cultures: Privacy reflects different cultural ideas about the individual, family, government, and corporations.U.S. discourse emphasizes individual privacy, while Chinese understandings have more often connected privacy to the family; European “data protection” also signals concerns distinct from U.S. “privacy.”
AI PRINCIPLES IN PRACTICE
AI principles are intended to guide practice, but AI systems take different forms as they encounter local cultures. Existing examples show both beneficial applications and serious risks, including surveillance, data exposure, and large-scale failures that disproportionately affect vulnerable people.
- Global Variation: AI systems are expected to take different forms in different regions as technologies encounter local cultures.The article describes these encounters as producing tensions, friction, and new possibilities rather than simple technological transfer.
- Global Applications: Examples span anti-corruption monitoring in Brazil, tuberculosis screening in India, data-driven farming in Nigeria, facial recognition in São Paulo, and predictive policing in Delhi.These cases illustrate the wide range of social purposes and institutional settings in which AI is deployed.
- Biometric Identity Programs: Biometric identity systems can expand access to formal identification and services while also enabling surveillance, political suppression, and exposure of sensitive data.The risks include centralized data vulnerabilities and documented data breaches connected to Aadhaar.
- Biometric Identity Programs: At massive scale, even a 2% biometric error rate could affect millions of people and has been linked to failures in obtaining food rations.The article links Aadhaar-related ration problems to several deaths by starvation.
- Amplifying Inequality: AI systems have a documented pattern of entrenching social inequality, disproportionately affecting historically disadvantaged and vulnerable groups.Examples include biased recruiting, discriminatory credit algorithms, and bias in language and image technologies.
A Global Pattern
Global evidence suggests that AI’s inequality risks are especially acute in low- and middle-income countries, where existing political and social vulnerabilities can magnify harm. Yet rigorous on-the-ground research outside the U.S. remains limited, making the way these systems travel difficult to assess directly.
- A Global Pattern: Low- and middle-income countries may be more vulnerable to AI’s negative social impacts and less likely to benefit from positive outcomes.The World Economic Forum identifies especially high risks of discriminatory machine-learning outcomes in these countries.
- Vulnerable Groups: Historically persecuted groups face particular risks from AI-enabled social-media distribution, surveillance, policing, and sentencing systems.Examples include Rohingya oppression, surveillance of Uighur communities, and biased U.S. judicial technologies affecting African Americans.
- Research Gap: Evidence about how fairness and bias travel across settings remains limited because rigorous, on-the-ground research outside the U.S. is scarce.The article therefore treats conclusions about cross-border effects as an area requiring further investigation.
- Deepening Digital Divides: AI can deepen digital divides when communities lack infrastructure, representation in training data, or opportunities to develop specialized skills.These exclusions limit participation as creators and can further marginalize communities.
AI Labor Exclusions
AI’s economic and social effects are likely to be unevenly distributed, with low- and middle-income countries and lower-skilled workers receiving fewer gains while facing greater pressures. Because inequality can destabilize societies, localized harms may produce broader social and geopolitical consequences.
- AI Labor Exclusions: Low- and middle-income countries are expected to receive modest benefits while bearing substantial AI-driven job losses.The article connects this distribution to widening regional and gender divides.
- AI Labor Exclusions: Most projected AI-generated wealth is expected to accrue to China and the United States, leaving poorer countries with limited opportunities to harness these technologies.PwC projects that 70% of the $15.7 trillion in global AI-generated wealth by 2030 will accrue to China and America.
- AI Labor Exclusions: Higher-skilled workers are positioned to benefit from AI, while low- and medium-skilled workers face downward pressure from increasingly capable machines and software.These pressures may exacerbate existing global income inequality.
- Global Risks: Inequality amplified by AI may increase instability, migration, or disruption to international financial markets beyond the groups initially harmed.The article links social inequality with political and economic instability, authoritarianism, civil unrest, and violent conflict.
- Social Organization and Control: AI can support elections, transportation, and disaster response, but it can also provide tools for intensive surveillance and other non-democratic ends.The same capabilities can serve both pro-social organization and social control.
CASE STUDY: SINGAPORE, SURVEILLANCE, AND SARS
Surveillance technologies are evaluated differently across societies, with Singapore and China showing broad acceptance alongside emerging signs of resistance and contestation. These cases also illustrate that surveillance can constrain movement while enabling bottom-up accountability.
- Singapore and surveillance: State surveillance is widely accepted in Singapore, where CCTV, social-media monitoring, peer surveillance, and extensive SARS-era controls became established practices.SARS responses included visitor screening, school closures, home quarantine with video monitoring, telephone surveillance, and electronic wrist tags.
- Tolerance, acceptance, and resistance: Surveillance ethics vary across societies: some tolerate large-scale government monitoring, while others respond with resistance shaped by local concerns about privacy, immigration, or parole.Facial recognition may be experienced as convenient or protective by some people and threatening by others.
- Tolerance, acceptance, and resistance: In China and Singapore, surveillance is often accepted as an exchange for security and stability, although outside media access and slower economic growth are exposing growing cracks in that acceptance.Public discussion of facial recognition ethics in China marks an unusual consideration of technology’s double-edged character.
- Surveillance and contestation: Surveillance can move from top-down monitoring toward bottom-up accountability, as captured images of Israeli soldiers circulated online and generated demands for personal accountability.The case shows that surveillance infrastructures can be repurposed by people under surveillance.
- Surveillance and contestation: Saudi Arabia’s movement-monitoring app constrained women’s travel while ordinary women contested it, including Manal Al-Sharrif’s public alerts about her husband’s notifications.The case links technologically mediated surveillance with both state control and citizen resistance.
CASE STUDY: SOCIAL CREDIT SYSTEM IN CHINA
China’s social credit system is a heterogeneous and still experimental set of public-private initiatives rather than the unified dragnet often portrayed abroad. Available research indicates that many Chinese citizens view it favorably, while its longer-term consequences remain unsettled.
- System scope and public perception: The system broadly aims to enforce existing laws, improve financial credit, reduce market fraud, and discourage behaviors such as environmental pollution and academic plagiarism.Numerical scores are used only in some limited and still experimental settings.
- System scope and public perception: Many Chinese citizens perceive the social credit system as beneficial because it represents state action against fraud, although perspectives from blacklisted entities remain limited.Younger, better-educated urban residents were most inclined to favor the system.
- System scope and public perception: China’s social credit system is a patchwork of public-private pilots that remains a work in progress, not yet the unified dragnet feared by Western media.The system is neither simply a sinister dragnet nor a ruse, and its tools may have far-reaching consequences.
- Global interpretation: Research on China’s social credit system demonstrates why AI-related social impacts require culturally situated analysis, because public perceptions and consequences vary by regional context.The article’s review covers five world regions and fourteen countries while emphasizing significant regional variation.
The Case for Ethnography
The paper argues for on-the-ground ethnographic research because AI’s effects emerge where technologies meet unequal, imperfect social worlds. Such research should connect global variation and vulnerability to practical decisions about designing, implementing, and monitoring AI.
- The Case for Ethnography: Ethnographic inquiry is needed to examine how AI technologies affect communities across the world, especially where systems may amplify inequality and harms require mitigation.The paper identifies the meeting of AI technologies and the real world as a research priority.
- The Case for Ethnography: Empirical research should follow AI from design through deployment, regulation, translation, interpretation, and daily use rather than relying on abstract principles alone.The authors emphasize close engagement with technologies wherever they are used.
- Vulnerability and inequality: AI has shown a pattern of exacerbating inequality, particularly in unequal societies and among vulnerable populations, making those populations important indicators of broader technological harms.The paper warns that impacts first felt by marginalized groups may eventually affect wider populations.
- From principles to practice: AI ethics requires practical action, not only principles, including deciding what technologies to produce and evaluating whether they support or undermine desired societies.The authors frame this as a human project for the digital age.
- From principles to practice: AI could reduce divisions, but achieving that promise depends on ethical, transparent, intentional implementation and attention to practical challenges on the ground.The paper balances concern about inequality amplification with the possibility of more inclusive outcomes.