The Geography of Ideologies: From Human Societies to Artificial Intelligence

Liberalism, conservatism, socialism, nationalism, religion, markets, equality, authority and sovereignty are not combined in the same way across human societies. Left and right continue to structure a large part of political competition, but they are not sufficient to map the world. Ideologies are shaped by history, institutions, religions, economic structures, social classes, generations and territories. As artificial intelligence systems become universal intermediaries between individuals and knowledge, this diversity raises a new question: how can machines be designed to be politically neutral in a world that has no universal political center of its own?

There is a persistent temptation to represent politics as a line. At one end sits the left; at the other, the right; between them, a center supposedly forming a natural point of equilibrium. This representation is convenient. It allows parties, governments and voters to be positioned within a given political system. It becomes far more fragile when extended to the planet as a whole.

The problem is not that left and right do not exist. These categories continue to refer to real political traditions. The problem lies in assuming that they can, by themselves, provide a universal system of coordinates.

An American voter who supports publicly funded healthcare may be described as strongly progressive within the US debate while defending a fairly ordinary position in parts of Europe. A European center-right party may support levels of taxation, labor protection and social welfare that would be regarded as highly interventionist in some American political debates. A nationalist movement may support free markets or, on the contrary, protectionism and the nationalization of strategic assets. A government may promote private enterprise while exercising extensive political control. A society may favor economic redistribution while remaining socially conservative on moral or family issues.

The political map of the world is therefore not a line. It is a multidimensional space.

LEFT AND RIGHT: A HISTORICAL INVENTION THAT BECAME A GLOBAL LANGUAGE

The distinction between left and right emerged in a very specific context: the French Revolution. In the National Assembly, defenders of a stronger role for the monarchy increasingly sat to the president’s right, while supporters of deeper political transformation sat to the left. A spatial arrangement became political vocabulary.

During the nineteenth and twentieth centuries, these concepts changed meaning. The left gradually became associated with political and social equality, the labor movement, socialism and later different forms of progressivism. The right became more closely associated with order, property, institutional continuity, conservatism and, depending on the period, nationalism or market economics.

Yet this history was never linear. Liberalism, revolutionary in its confrontation with absolute monarchy during the nineteenth century, may today be placed at the center or center-right in contemporary welfare democracies. Nationalism has successively been revolutionary, anti-colonial, conservative, socialist and authoritarian. Social protection, once associated primarily with labor movements, has been incorporated into most European conservative platforms. Environmental issues, largely absent from the old industrial divide, have become a major political dimension.

In Left and Right, Norberto Bobbio attempted to identify a more durable criterion beneath these historical changes: differing sensitivities toward equality and inequality. The distinction remains useful, but it is still insufficient to explain every dimension of contemporary politics.

To understand the world, several axes must be considered simultaneously: attitudes toward economic equality, individual liberty, the role of the state, authority, nation and sovereignty, religion, social change, markets, globalization and now environmental constraints.

The World Values Survey offers an empirical illustration of this complexity. Research by Ronald Inglehart and Christian Welzel identified two major dimensions of cultural variation: traditional versus secular-rational values, and survival versus self-expression values. This framework is not itself a map of ideologies, but it shows that a society cannot be reduced to an economic position on a left-right axis. Development, religion, family, authority, individual autonomy and perceptions of security evolve together while remaining deeply shaped by historical trajectories.

Ideologies are therefore better understood as combinations than as fixed points.

LIBERALISM: ONE FAMILY, DIFFERENT MEANINGS

Liberalism is probably the clearest example of the traps hidden in political vocabulary. Historically, its fundamental proposition has been the limitation of arbitrary power. The rule of law, individual rights, property, representation, freedom of conscience, freedom of expression and legal certainty form the core of the modern liberal tradition. But liberalism fragmented into several branches.

Classical liberalism emphasizes individual autonomy, property and limited government. Economic liberalism gives particular importance to markets, competition and private initiative. Social liberalism, by contrast, argues that the effective exercise of liberty may require public intervention in areas such as education, welfare, healthcare or the regulation of certain economic relationships.

This distinction explains part of the misunderstanding between the United States and Europe. In contemporary American vocabulary, “liberal” generally refers to a progressive or center-left orientation supportive of civil rights and a larger social state than conservatives favor. In several European countries, “liberal” is more directly associated with market economics, individual freedoms and more restrained government intervention.

Two political leaders may therefore both describe themselves as liberal while defending substantially different economic policies.

That alone should encourage caution whenever a study claims to determine whether a person, a country or now an artificial intelligence is “liberal” without specifying the political system of reference.

CONSERVATISM: THE POLITICS OF CONTINUITY

Conservatism is often defined too quickly as opposition to change. Its intellectual tradition is more subtle. It rests in part on skepticism toward the idea that society can be entirely reconstructed from abstract principles.

Inherited institutions, order, historical continuity, responsibility, family, community, property, authority and religion may all occupy important places within conservative thought. Yet there is no single conservatism, just as there is no single liberalism.

British conservatism historically developed within a particular relationship to parliamentary monarchy and inherited institutions. American conservatism combines constitutionalism, economic individualism, religion, federalism and several national traditions. European Christian democracy reconciled market economics, private property, family and the construction of an extensive welfare state. Conservatisms in India, Japan, Russia, Latin America or the Muslim world are embedded in very different religious and national histories.

Two parties described as conservative may therefore disagree profoundly on protectionism, social spending, foreign policy or the proper role of government.

Conservatism refers more to a family of attitudes toward change, authority and institutions than to a universal economic program.

Socialism developed during the nineteenth century amid industrialization, urbanization and the transformation of labor. Its central critique concerned inequalities of power generated by ownership and the organization of capital.

The socialist question can be framed simply: is legal equality sufficient when individuals possess highly unequal economic resources? The answers given to this question produced radically different systems.

Social democracy gradually prioritized representative democracy, collective bargaining, public services, progressive taxation and redistribution while preserving market economics. This formula deeply shaped postwar Europe.

Democratic socialism generally retains a greater ambition for economic transformation while rejecting the abandonment of political pluralism.

Marxist-Leninist communism, in the institutional forms it took during the twentieth century, produced one-party states characterized by extensive public ownership, economic planning and political centralization. Conflating these traditions is therefore a mistake.

Contemporary Sweden is not the Soviet Union. A strongly redistributive tax system is not a collectivized economy. Conversely, the presence of private firms within an economy does not necessarily make its political or institutional architecture liberal. Contemporary China makes this distinction particularly clear.

CHRISTIAN DEMOCRACY: THE MODEL THAT BLURS THE CATEGORIES

Christian democracy is especially important for understanding why American political categories travel poorly. Historically influenced by Christian social teaching, it has simultaneously defended property, market economics, community, family, subsidiarity and social protection.

Center-right parties played a major role in building European systems of social insurance, co-determination, family support and public services.

In an ideological framework in which public intervention and conservatism are necessarily opposites, this combination would appear incoherent.

In European political history, it is not incoherent at all. Christian democracy demonstrates that doctrines can combine economic and social dimensions that other traditions separate.

ANARCHISM: THE RADICAL CRITIQUE OF AUTHORITY

Anarchism is another major political tradition often neglected in contemporary maps. Its various currents share a fundamental distrust of coercive hierarchies and, above all, the state. Some anarchist traditions are socialist and collectivist; others place greater emphasis on individual autonomy.

This tradition reminds us that the economic axis is insufficient. It is possible to be radically egalitarian while rejecting a powerful centralized state.

It is precisely this distinction between economic power and political power that led certain frameworks, such as the Political Compass, to separate an economic axis from an authoritarian-libertarian axis. The intuition is useful: attitudes toward equality and attitudes toward authority are not the same thing.

LIBERTARIANISM: THE PRIMACY OF INDIVIDUAL AUTONOMY

Contemporary libertarianism, particularly influential in parts of the Anglo-American world, pushes the principle of individual autonomy very far. It generally combines extensive economic freedom, strong property rights, low taxation and limited government with broad support for personal liberties.

This combination can place libertarians on the economic right while making them socially permissive. Again, the left-right line becomes insufficient.

A libertarian may agree with a conservative on taxation while disagreeing fundamentally over the state’s authority to regulate private behavior. This distinction becomes important when political tests are applied to artificial intelligence models.

NATIONALISM AND SOVEREIGNISM: IDEOLOGIES THAT CROSS LEFT AND RIGHT

Nationalism is probably one of the political forces that resists simple classification most strongly. Its basic intuition is to place the nation at the center of the political community and to defend its capacity for collective self-determination. Almost everything else can vary.

Nationalism can be civic or ethnocultural. It can support free markets or protectionism. It can be conservative, socialist, revolutionary or anti-colonial. It can aim at emancipating a dominated people or consolidating the power of an already established state.

During the twentieth century, countless national liberation movements in Africa, Asia and the Arab world combined nationalism with left-wing ideologies. In contemporary Europe, nationalism is far more often associated with the right.

Sovereignism follows a related logic without being identical to nationalism. It emphasizes political, legal or economic control exercised by the state or national community against supranational institutions, external dependencies or constraints associated with globalization. It too can appear on the left or the right.

Industrial, energy, food and technological sovereignty are now invoked by governments belonging to almost every political family.

FASCISM: RADICAL NATIONALISM, MOBILIZATION AND THE REJECTION OF PLURALISM

Fascism belongs to another historical category. Research by Robert Paxton, Roger Griffin and other historians converges around several characteristics even though no completely consensual definition exists: ultranationalism, exaltation of the national community, rejection of liberal pluralism, political mobilization, valorization of authority, pursuit of national regeneration and acceptance of political violence as a legitimate instrument.

It is important to distinguish historical fascism from ordinary conservatism, nationalism or authoritarian politics more broadly. The words “fascist” and “fascism” are now frequently used as polemical labels. Their analytical value depends, on the contrary, on sufficiently restrictive definitions.

The same caution applies to all ideological categories: the more words are used as political weapons, the less useful they become for understanding the phenomena they claim to describe.

POPULISM: THE PEOPLE AGAINST THE ELITES

Populism belongs to a somewhat different category. A large part of contemporary academic literature, especially the work of Cas Mudde, defines it as a “thin-centered” ideology opposing a supposedly authentic people to an allegedly corrupt elite and arguing that politics should express the general will of the people.

Because it is ideologically thin, populism can attach itself to fuller doctrines. There are left-wing populisms centered on inequality, economic power or oligarchy. There are right-wing populisms emphasizing immigration, nation, order or sovereignty.

Populisms may themselves be economically interventionist or economically liberal. Treating populism as a single ideology with a fixed policy program therefore obscures more than it clarifies.

POLITICAL ECOLOGY: INTRODUCING PHYSICAL LIMITS INTO POLITICS

A large part of modern political thought developed during a period when industrial growth was the shared horizon. Capitalists and socialists disagreed radically over ownership, yet both could share a belief in increasing production.

Political ecology changes this structure. It raises questions about physical limits, climate, biodiversity, future generations and the sustainability of economic organization.

In much of Europe, green parties have moved closer to the cultural and social left. But ecology is not philosophically condemned to belong to the left. Land preservation, conservation, localism and critiques of industrialization also exist within various conservative traditions. Ecology therefore adds another axis cutting across older political families.

RELIGION AND IDEOLOGY: A GLOBAL DIMENSION

In parts of highly secularized Western Europe, religion may appear to have become a secondary political variable. At the global level, that would be a serious mistake.

Christianity, Islam, Hinduism, Judaism, Buddhism and other religious traditions continue to influence representations of family, law, authority, political power and national identity. This does not mean that each religion produces a homogeneous political orientation.

Within the same religious tradition there may coexist conservative, reformist, social, nationalist and liberal interpretations.

Pew Research Center’s 2025 international survey on religious nationalism across 36 countries shows precisely how much the relationship between religion, national belonging, political leadership and legislation differs from one society to another.

Religion is therefore indispensable to any meaningful geography of ideologies. It is particularly important for understanding India, the United States, Latin America, much of Africa, the Middle East and North Africa.

AUTHORITARIANISM IS NOT A POSITION ON THE LEFT-RIGHT AXIS

One of the most persistent confusions is to equate authoritarianism with ideological orientation. They are different variables. Authoritarian regimes have claimed communist, nationalist, conservative, Arab socialist, developmentalist and military doctrines.

Likewise, pluralist democracies can be governed by socialist, liberal, conservative or nationalist parties. The question “who owns economic resources?” is not the same as “how is political power exercised?”

Economic organization and political pluralism must therefore be separated analytically. According to V-Dem’s Democracy Report 2026, using its own Regimes of the World methodology, the world at the end of 2025 contained 92 autocracies and 87 democracies, and roughly 74% of the global population lived in systems classified as autocratic. These figures are outputs of a specific methodological framework, not political facts independent of definition. They nonetheless remind us that classifying regimes is not the same thing as classifying ideologies.

International IDEA adopts a different, multidimensional framework. Its Global State of Democracy 2025 assesses representation, rights, rule of law and participation. It reported that 94 of the 174 countries studied had declined on at least one dimension of democratic performance between 2019 and 2024.

These two examples themselves demonstrate that there is no single scientific way to summarize an entire political system in one score.

THE GLOBAL GEOGRAPHY OF IDEOLOGIES

Political traditions become even harder to classify once they encounter geography. History, borders, wars, religions, economic structures and institutions have generated different regional combinations.

We should therefore speak of tendencies and trajectories, never of an ideology belonging naturally to an entire people or region.

WESTERN EUROPE: THE HISTORICAL LABORATORY OF MODERN IDEOLOGIES

Western Europe remains the space in which the traditional nomenclature probably retains the greatest historical coherence. Liberalism, socialism, communism, conservatism, Christian democracy and political ecology all generated durable parties, trade unions, institutions and political cultures. But postwar convergence transformed these families.

The governing left largely accepted market economics. The governing right broadly accepted the welfare state. Pension systems, healthcare and unemployment insurance became relatively stable components of European states even when their organization remained contested.

At the same time, new cleavages became structural: immigration, European integration, globalization, environmental policy, national identity, cultural questions and sovereignty. It is in this context that nationalist and populist right-wing parties have grown in several countries.

Comparative data published by Pew in 2026 identified a particularly regular pattern across the European countries it studied: people with less formal education were, on average, more sympathetic to several parties classified as right-wing populist than those with more education. The educational divide does not explain everything, but it shows that contemporary politics has also become a geography of educational and cultural capital.

The old opposition between workers and owners has not disappeared. It has been overlaid by new divides between graduates and non-graduates, metropolitan centers and peripheries, and perceived winners and losers from economic openness.

NORTHERN EUROPE: MARKETS, WELFARE AND CULTURAL INDIVIDUALISM

Nordic societies are especially useful for dismantling ideological caricatures. They combine advanced market economies, deep integration into international trade, powerful private firms and high levels of taxation, social expenditure and public services.

They correspond neither to the minimalist capitalism sometimes associated with Anglo-American liberalism nor to a socialist economy based on generalized public ownership.

Research based on the World Values Survey also places several Nordic societies among those where secular-rational and self-expression values are especially strong.

The same society can therefore combine markets, redistribution and social individualism. Such a configuration is difficult to represent accurately on a single left-right line.

CENTRAL AND EASTERN EUROPE: THE MEMORY OF REGIMES

The end of the Soviet bloc did not erase earlier political trajectories. Central and Eastern European societies experienced simultaneous transitions toward political pluralism, markets, private ownership and renewed national sovereignty.

The social costs of some transitions, memories of communist rule, identity questions and geopolitical insecurity generated distinctive combinations.

Some populations may distrust state economic intervention because of communist experience while still wanting political power to assert sovereignty and national identity strongly. Others remain attached to social protections inherited from the past.

The World Values Survey continues to show that formerly communist societies form recognizable cultural clusters decades after those regimes disappeared. Institutions sometimes change faster than values.

THE UNITED STATES: TWO PARTIES, SEVERAL POLITICAL AMERICAS

The United States occupies a disproportionate place in the world’s political vocabulary, partly because English dominates much of digital information and the technology industry. Yet its political system is highly unusual.

The American institutional and electoral structure has encouraged two large coalitions. Democrats and Republicans must therefore aggregate groups that, in a multiparty parliamentary system, would probably be divided among several parties.

The Democratic coalition includes social liberals, progressives, moderates, voters strongly attached to social welfare, others focused more heavily on civil rights, and groups whose economic interests are not identical.

The Republican coalition brings together religious conservatives, market-oriented voters, nationalists, entrepreneurs, rural constituencies, institutional conservatives and an increasingly influential populist right.

American polarization therefore cannot be reduced to taxation. Religion, immigration, guns, abortion, foreign policy, conceptions of nationhood, the role of the federal government, attitudes toward universities and trust in institutions now help structure both coalitions.

Pew Research Center’s June 2026 political typology is instructive precisely because it moves beyond the simple Republican-Democrat distinction and divides Americans into nine groups constructed from thirty questions.

The data show, among other things, a strong generational gradient: people under 30 are more heavily represented in several groups on the left, while those over 65 are overrepresented in several conservative categories. This does not mean that every young American is progressive or every older American conservative. It means that generation is a meaningful dimension of American political geography.

The American “center” is therefore itself a moving average of several political societies existing side by side.

CANADA: A NEIGHBOR THAT ALREADY MOVES THE CENTER

Canada shares with the United States a deeply integrated economy, part of its cultural environment and much of its media language. Its political spectrum nevertheless differs.

Federalism, Quebec, bilingualism, healthcare, immigration, institutionalized multiculturalism and social protection have all contributed to the construction of a different political center.

Two neighboring, developed and predominantly English-speaking countries do not therefore define political normality in exactly the same way. This observation becomes crucial when we later ask what a “neutral” answer from an artificial intelligence should look like.

LATIN AMERICA: INEQUALITY, SOVEREIGNTY, RELIGION AND MEMORY

Left and right remain important categories in Latin America, but their content is deeply shaped by regional history. Concentrated ownership, inequality, coups, military regimes, popular movements, external dependency, US intervention and the influence of the Church have all structured political conflict. Latin American lefts cover a spectrum from institutional social democracy to revolutionary traditions and highly centralized political systems.

The right is equally diverse: religious conservatism, economic liberalism, nationalism, historical authoritarianism, defense of property and newer forms of populist mobilization.

The growth of evangelical Protestantism in several countries adds another dimension to older religious cleavages. The region especially reminds us that support for redistribution does not necessarily imply cultural progressivism, and that social conservatism can coexist with strong demand for economic intervention.

RUSSIA AND EURASIA: CAPITALISM, STATE AND SOVEREIGNTY

Contemporary Russia demonstrates the limits of another common association: the assumption that economic capitalism necessarily leads to political liberalism.

The country possesses a monetary and market economy, private firms and private capital holders. At the same time, the state plays a central role in strategic sectors and political power is highly concentrated.

Institutional discourse emphasizes sovereignty, power, historical continuity, security and traditional values. This combination differs markedly from Anglo-American liberal conservatism. It demonstrates that market economics, political pluralism, social conservatism and nationalism are four distinct variables.

CHINA: THE SYSTEM THAT BREAKS THE GRID

China is probably the clearest contemporary example of the limits of the classic Western classification. It is governed by a Communist Party whose official doctrine continues to invoke Marxism-Leninism and socialism with Chinese characteristics.

The state exercises major influence over strategic industries, finance, infrastructure and planning. Yet China also has a vast private sector, markets, entrepreneurs, investors, multinational corporations and intense competition across many industries.

To these economic elements must be added political centralization, nationalism, technological sovereignty and a specific conception of state continuity.

Describing China simply as “left-wing” because it is ruled by a Communist Party provides very little information about how the system operates today. Describing it as “right-wing” because it extensively uses market mechanisms would be equally misleading. The Chinese system combines elements that Western classifications usually separate.

JAPAN AND SOUTH KOREA: ECONOMIC MODERNITY AND DISTINCT POLITICAL HISTORIES

Japan and South Korea are highly developed, industrialized and technologically advanced democracies. Their political trajectories nevertheless differ substantially from those of Europe or North America.

Japan has experienced unusually durable domination by a large conservative party in an economic system where the state historically maintained close relationships with industrial development.

South Korea experienced rapid industrialization, authoritarianism, democratization and the rise of powerful conglomerates, while remaining confronted with an exceptional national security environment because of the division of the Korean Peninsula. The left-right axis exists in both countries, but it is reshaped by distinct historical and geopolitical variables.

INDIA: POLITICS CANNOT BE SEPARATED FROM CIVILIZATION

India combines electoral democracy, market economics, an interventionist legacy, linguistic plurality, caste, religion, federalism and regional identities. Its economic trajectory itself changed substantially following liberalization reforms beginning in the 1990s.

Contemporary Hindu nationalism, however, cannot be understood through economic policy alone. It involves a conception of national identity, history and the relationship between the religious majority and the political community.

India therefore reminds us that religion is not always a private variable separate from ideology. In some societies, nation, civilization and belief are tightly intertwined within political language.

THE MIDDLE EAST AND NORTH AFRICA: SEVERAL HISTORIES UNDER ONE REGIONAL LABEL

The Middle East and North Africa provide another illustration of the limits of imported political cartographies. The region includes monarchies and republics, rentier and diversified economies, party systems and highly centralized regimes, as well as nationalist, liberal, Islamist and socialist traditions.

Arab nationalism, Arab socialism, different forms of political Islam, monarchical traditions and liberal movements have structured different periods.

Oil and gas have also altered the relationship between taxation, redistribution and power in several states. A government able to finance a significant share of its budget through hydrocarbon revenues does not build exactly the same political contract as a state financed primarily by direct taxation of its population.

Religion also has its own political structures. Political Islam itself includes very different families, ranging from movements participating in electoral competition to much more radical conceptions of political order.

Speaking of a single “Arab” or “Muslim” political ideology would therefore be as misleading as speaking of a uniform European political ideology.

MOROCCO: A COMBINATION THAT RESISTS IMPORTED CATEGORIES

Morocco provides a particularly instructive case. Its political system combines a monarchy with deep historical and religious foundations, representative institutions, party pluralism, an open market economy, international integration and substantial state capacity to guide development.

Industrial strategies, infrastructure projects, energy policy, social programs and public investment show that markets and state intervention are not regarded as mutually exclusive.

Moroccan society is simultaneously undergoing rapid demographic, educational, urban and cultural transformation without the disappearance of religious and family references.

Describing Morocco simply as “conservative” or “liberal” therefore fails to capture this combination. Arab Barometer surveys help make this complexity empirically visible.

In Wave VIII, conducted between December 2023 and January 2024 among more than 2,400 Moroccans, 73% of respondents agreed that democratic systems may have problems but remain preferable to alternatives. Roughly 68% considered a multiparty parliamentary system appropriate or very appropriate for the country.

At the same time, answers to other questions concerning authority, religion and political effectiveness revealed preferences that do not fit neatly within a classical Western liberal framework.

Even more important for the subject of this article, Arab Barometer observed sharp divisions in Moroccan attitudes according to socioeconomic status. The survey referred almost to “two Moroccos”: one wealthier and more educated, the other poorer and less educated, whose perceptions on many questions differed significantly.

This illustrates the central argument of this article perfectly: the geography of ideologies does not only run between countries. It cuts through societies internally.

SUB-SAHARAN AFRICA: PARTIES, STATES AND POLITICAL IDENTITIES

Politics in sub-Saharan Africa also has its ideological traditions: pan-Africanism, African socialisms, independence nationalism, liberalism, religious conservatism, democratic movements and different doctrines of development. But political parties are not everywhere organized around a stable left-right doctrinal axis as they are in Western Europe.

State formation, territorial or communal belonging, natural resources, security, patronage networks, public service capacity and the legacy of liberation movements may weigh more heavily than certain abstract ideological divisions.

This does not mean Africa lacks ideology. It means its political systems must be understood through their own trajectories.

National maps remain insufficient. In many cases, political differences within a country are becoming almost as important as those separating states.

A new geography runs across societies: metropolitan centers and peripheries, graduates and non-graduates, asset holders and people dependent mainly on wages, older and younger generations, internationally connected professions and locally rooted occupations.

These variables never mechanically determine an opinion. They change the experience of the world from which opinions are formed.

METROPOLISES AND PERIPHERIES

In many advanced economies, large cities concentrate universities, higher-value services, finance, technology, media, intellectual professions and populations deeply integrated into international exchanges. This material structure creates a particular social environment.

Mobility is greater. Exposure to cultural diversity is part of daily life. The benefits of international openness are more visible. Professional sectors depend more heavily on global flows of capital, data or skills.

A declining industrial region, a smaller city or a rural area may experience the same transformations very differently.

The globalization that produces high-skilled employment and property appreciation in a capital city may elsewhere be associated with offshoring, disappearing industries or demographic decline. Ideological conflict is therefore partly a conflict between different experiences of the same economic system.

EDUCATION AS A NEW POLITICAL FRONTIER

Education has become one of the most visible determinants of voting behavior in several democracies. This does not mean that a degree mechanically turns someone into a progressive voter.

Higher education correlates with other variables: occupation, mobility, income, urban location, social networks and international exposure. It also produces specific languages and cultural references.

Pew’s 2026 data on several European right-wing populist parties showed consistent differences in sympathy according to education. Across the ten parties compared, people with lower levels of formal education expressed more sympathy than those with more education.

This does not summarize the entire electorate of those parties, but it shows that the old politics of class has partly transformed into a politics of qualifications and cultural capital.

WEALTH COMPLICATES THE IDEA OF CLASS

Social class can no longer be measured by monthly income alone. Two households earning similar amounts may occupy profoundly different economic positions. One may own its home, hold financial assets and possess inherited family wealth. The other may pay high rent, own no meaningful assets and depend entirely on wages.

Their views of interest rates, taxation, inflation or property prices may therefore diverge. In societies where property and financial assets have appreciated strongly, the divide between asset owners and non-owners has become politically important. The historic opposition between capital and labor has not disappeared. It has fragmented.

WORK PRODUCES DIFFERENT VISIONS OF THE WORLD

Professional environments also shape political priorities. An entrepreneur encounters regulation, taxation, financing and risk directly. A civil servant sees every day the situations in which public action is required. A farmer lives under the pressure of prices, climate and territory. A soldier thinks more often in terms of capability, hierarchy and security. An academic operates in an environment shaped strongly by internationalization and scientific norms. An industrial worker is directly exposed to international competition and automation.

No profession determines an ideology. But each professional world selects the problems that appear most concrete. Politics is also a sociology of attention.

GENERATIONS DO NOT GROW UP IN THE SAME WORLD

Age creates another political divide. But age effects must be distinguished from generational effects. People may change some attitudes as they grow older. Generations may also preserve part of the outlook shaped by the events that marked their entry into adulthood.

War, prosperity, unemployment, inflation, financial crises, housing access, pandemics and digital revolutions produce very different experiences.

Western generations that entered adulthood after the 2008 crisis have experienced housing, employment and economic stability differently from cohorts that benefited from decades of growth and substantial asset appreciation.

In the United States, Pew’s 2026 typology confirms that younger adults are more heavily represented in several categories on the left, while older adults are more present in several categories on the right.

Yet another development prevents easy generalization: new conservative, nationalist and libertarian communities have also emerged among young people, particularly within male-dominated digital subcultures. Youth is therefore not an ideology.

RELIGION AND SECULARIZATION

Religious geography also runs through societies. In several Western democracies, religious observance is statistically associated with certain conservative preferences. Yet the relationship differs across denominations, social groups and countries.

European Catholic social thought, for example, contributed to relatively interventionist economic policies. Some Latin American evangelical movements are culturally conservative but recruit among lower-income populations that may support redistributive policies. In Muslim societies, religiosity, support for democracy and economic preferences do not form a single bloc either. The relationship between religion and politics is therefore an interaction, not an equation.

GENDER: A REAL BUT NON-UNIVERSAL DIVIDE

Several contemporary democracies also display political differences between men and women, particularly among younger generations.

In some Western contexts, young women on average place themselves closer to progressive parties than young men. Yet the gaps vary greatly and can disappear or reverse depending on country, religion, class or issue.

Social media have also created political subcultures that no longer follow the same channels of socialization across genders. It would therefore be unwise to infer either a “male” or “female” ideology. The variable matters without determining outcomes.

THE INTERNET CREATED A NEW GEOGRAPHY WITHOUT TERRITORY

A digital geography has now been added to physical geography.

A person living in Casablanca may share more daily political references with a virtual community based in New York, London, Paris or Dubai than with some people living in the same city.

Algorithms have created transnational social spaces. Progressive, conservative, religious, libertarian, nationalist and ecological communities circulate across platforms and export their vocabulary. This circulation sometimes produces a peculiar importation of political conflicts.

Categories born in American debate are reused in societies with entirely different institutional histories. At the same time, local nationalisms and religious debates use the communication infrastructure of American platforms to globalize themselves.

Politics is therefore no longer only territorial. It is networked. And it is precisely in this digital space that large language models emerge.

SILICON VALLEY IS NOT A NEUTRAL TERRITORY

The companies building artificial intelligence systems are not located outside society. They have headquarters, recruit from specific universities, attract particular professional groups and operate under particular legal systems.

Silicon Valley is itself an extraordinarily distinctive social environment: concentrations of capital, high educational attainment, international immigration, engineering, entrepreneurship, elite universities, the technology industry and intense exposure to American cultural debates.

It would be surprising if this environment had absolutely no influence on the organizations born within it. This does not mean their models mechanically reproduce the average political views of San Francisco residents. The influence is much more indirect.

It runs through dataset selection, safety policies, evaluation methods, preferences of early users, legal judgments and commercial objectives.

Elon Musk’s claim that some models reflect the politics of the San Francisco Bay Area while Grok reflects “California plus Texas” should therefore not be treated as scientific proof.

It nevertheless raises a legitimate question: does the institutional geography of AI laboratories affect how they define what counts as a balanced answer?

THE GEOGRAPHY OF AI ITSELF IS HIGHLY CONCENTRATED

This question becomes more important once the concentration of the industry is considered. The Stanford AI Index 2026 shows that private industry produced more than 90% of the models classified as notable in its 2025 sample.

The United States remains the leading producer of notable models, while China dominates several dimensions of scientific research, publications, citations and patents. Stanford also reports that the performance gap between top US and Chinese models has narrowed substantially.

The production of frontier models therefore remains highly concentrated in only a few ecosystems. The United States and China dominate, while Europe, the Middle East, India, South Korea, Japan and other regions seek to expand sovereign capabilities.

A technology intended to serve billions of people is still being designed largely within a small number of economic and scientific centers.

AN ARTIFICIAL INTELLIGENCE DOES NOT HAVE AN IDEOLOGY IN THE HUMAN SENSE

A fundamental distinction is required here. An LLM is not a voter. It has no wealth, religion, family, social experience, economic interest or lived nationality. It does not believe a political proposition.

It computes distributions over possible responses from patterns learned during training and modified through post-training and the context in which it is used.

Calling a model “left-wing” or “right-wing” is therefore an anthropomorphic shortcut. What can be studied scientifically is not an internal political conviction. It is behavioral regularity.

When repeatedly exposed to political questions under defined experimental conditions, a model may produce distributions of responses resembling some human political families more than others. That is a real object of research. But it should be called what it is: the observed political orientation of responses under a particular experimental design.

HOW DOES AN LLM ACQUIRE THESE ORIENTATIONS?

The visible response of a model results from several successive layers. First come the pre-training data. Books, articles, websites, forums, technical documents and other materials already contain the conflicts, prejudices, norms and values of the societies that produced them.

Then comes selection. No model receives the entire digital world in perfectly representative proportions. Datasets are filtered, cleaned, deduplicated and weighted.

Then comes post-training. The system learns to favor responses judged more useful, safer, more accurate or more consistent with the objectives of the laboratory.

Human or synthetic preference data used to evaluate responses add another layer. Safety policies and legal constraints add another. System instructions used by the product can modify behavior further. Finally, the user prompt and conversational context influence the answer produced at that moment.

What people call the “opinion of the AI” is therefore the final output of a complex chain. Attributing that output simply to the personal ideology of a few engineers is almost always too simplistic.

THE POLITICAL COMPASS: A SEDUCTIVE IMAGE, A FRAGILE PROOF

Charts positioning different models on the Political Compass are visually striking because they transform a complex phenomenon into coordinates. The test arranges answers on an economic left-right axis and a social authoritarian-libertarian axis.

In the comparison that prompted Elon Musk’s reaction, many models appeared in an economically left and socially libertarian quadrant, while Grok 4.5 appeared closer to the economic center. It would be tempting to conclude that most AI systems are progressive and Grok is politically more centrist. Academic research demands considerably more caution.

A study presented at ACL 2025 by Mats Faulborn, Indira Sen, Max Pellert, Andreas Spitz and David Garcia explicitly notes that the Political Compass Test is not a scientifically validated instrument for measuring LLM political preferences. The authors developed a method more directly grounded in political science and tested eleven models across 88,110 responses.

Their result is particularly interesting: some left-leaning tendencies are indeed present among several instruction-tuned models, but the measures can be unstable and the Political Compass may sometimes exaggerate observed biases.

Another 2025 study performing a detailed factor analysis of the Political Compass found that prompt phrasing and fine-tuning can shift scores.

In other words, a model may not have one unique position on the map. The method used to question it partly constructs the point being displayed.

POLITICAL BIAS DOES NOT DISAPPEAR

Criticizing the Political Compass does not mean that bias is imaginary. Other methods also detect asymmetries. Bang, Chen, Lee and Fung proposed at ACL 2024 that researchers distinguish the political content of responses from lexical style in order to measure bias more explicitly.

Faulborn and co-authors subsequently showed in 2025 that some tendencies remain measurable using instruments better grounded in political theory.

A 2026 Scientific Reports study by Daniel Wang and co-authors examined how several models associated individual profiles with a range of characteristics, including political affiliation. Within the specific experimental framework, tested models frequently overrepresented liberal affiliations relative to reference distributions.

These results do not justify the statement that “LLMs are liberal.” They show that models can produce politically asymmetric distributions in certain tasks. The distinction matters.

LANGUAGE CAN MOVE THE POLITICS OF THE SAME MODEL

One of the most important findings from a global perspective concerns language. In 2025, Chadi Helwe, Oana Balalau and Davide Ceolin evaluated fifteen multilingual LLMs by varying both prompt language and the nationality assigned to the model, using a framework covering the fifty most populous countries.

Their central conclusion is highly relevant: the language used significantly affects measured political orientation. Another multilingual study published in 2025 by Daniil Gurgurov and co-authors also observed substantial variation across languages, model families and phrasings. It can therefore be misleading to ask: “What is this model’s ideology?”

The more accurate question is: “What political behavior does this model produce, in which language, under which context, with which prompt and according to which measurement instrument?”

The same system may not occupy exactly the same position when responding in English, French, Arabic, Spanish or Chinese.

TRANSLATION IS NOT POLITICALLY EMPTY

The issue becomes even deeper because political concepts do not have perfect equivalents across languages. The American term “liberal” is not equivalent to the French “libéral.” French laïcité has no exact conceptual equivalent in American law or political vocabulary.

“Secularism” can refer to very different arrangements between religion and the state. The concepts of community, nation, sovereignty, family and authority also carry different histories.

A 2026 study in Humanities and Social Sciences Communications examined ideological differences between human translations and ChatGPT-4o translations of Chinese political speeches delivered at the United Nations General Assembly.

The issue therefore extends beyond lexical accuracy. Translation can implicitly transport a different way of framing a political concept. A multilingual model is not merely confronted with several vocabularies. It is confronted with several intellectual histories.

MODELS MAY ALSO REFLECT GEOPOLITICAL GEOGRAPHIES

Comparing models developed in different countries introduces a second dimension. A study published in March 2026 in Humanities and Social Sciences Communications compared GPT-4o and DeepSeek-R1 on questions concerning international affairs and different conflicts.

The authors observed differences in geopolitical framing between the two systems. Again, no single study should be turned into a universal rule. But the result is consistent with the idea that models are not produced in an institutional vacuum. Available data, regulation, company policy and national information environments can all influence answers.

As American, Chinese, European, Indian, Arab and other ecosystems build their own systems, normative plurality will probably become more visible.

LLMs MAY ALSO ADAPT TO THE USER

Another phenomenon complicates the idea of a fixed ideological position even further. A Scientific Reports study published in May 2026 in the Brazilian political context described certain LLMs as “ideological chameleons.”

Researchers studied their ability to adapt responses to the political positions expressed by users and the risk that such adaptation could contribute to personalized echo chambers. The problem then becomes almost the opposite of fixed bias. An AI may influence a user by imposing a frame. But it may also reinforce a user’s convictions by adapting excessively to that frame.

The goal of a balanced model is therefore not simply to sit in the middle. It must also avoid becoming a flattering mirror of every interlocutor.

WHAT DOES NEUTRALITY MEAN FOR ARTIFICIAL INTELLIGENCE?

We now reach the real problem. Suppose a laboratory decided tomorrow to build a perfectly neutral model. Where should it be positioned?

At the center of American politics?
At the center of French politics?
At the center of Moroccan politics?
At the center of Indian politics?
At the average position of the European Union?
At the average position of eight billion human beings?
And how would such an average be calculated?
One human, one vote?
One internet user, one vote?
Each language weighted equally?
Only democratic societies?
United Nations norms?
The law of the user’s country?

Once asked concretely, the idea of a universal neutral point collapses.

FREEDOM OF SPEECH DOES NOT HAVE THE SAME CENTER EVERYWHERE

Freedom of expression provides an obvious example. The United States maintains an especially expansive constitutional conception of protected speech.

European democracies generally accept more legal restrictions on certain forms of expression. France adds a distinct historical and legal tradition surrounding laïcité.

Other societies place greater weight on the protection of religion, public order or collective cohesion. When a model explains where the “normal” balance lies between speech and restriction, it does not discover a natural equilibrium. It necessarily chooses a frame.

THE ROLE OF THE STATE HAS NO ZERO POINT EITHER

Consider healthcare. A model treating the United States as the default could describe certain forms of public financing as highly interventionist. A model using Western Europe as the baseline might describe them as ordinary.

The same divergence appears in pensions, higher education, labor law and unemployment insurance. Ideology may therefore appear simply because a local institution has been mistaken for universal normality.

RELIGION, FAMILY AND NATION ARE EVEN HARDER TO UNIVERSALIZE

The place of religion in public life differs enormously between France, the United States, Morocco, India and Japan. Family norms vary as well.

The conception of nationhood is not identical in a society historically shaped by immigration, in an older European nation-state, in a monarchy with religious legitimacy or in a civilization-state.

A response that appears perfectly reasonable in San Francisco may therefore seem culturally tilted in Casablanca or New Delhi. The reverse is equally true. The challenge is not to determine which society is “right.” It is to separate facts from norms.

FACTS, INTERPRETATIONS AND VALUES

This is probably where a more robust definition of AI neutrality can be found. Neutrality should not mean systematically producing an opinion located exactly halfway between two adversaries. The midpoint between a true statement and a false statement is not more true. Neutrality does not mean permanent equidistance.

A rigorous system should distinguish at least three levels. When a question concerns a well-established fact, it should be able to state it clearly. When evidence is uncertain or contradictory, it should represent that uncertainty. When disagreement is fundamentally normative, it should identify the values in conflict and present the principal competing positions honestly.

On an economic issue, for example, it could distinguish the likely empirical effects of a tax from the normative question of how much inequality a society should accept.

On a religious question, it could distinguish what the law says from what different doctrines consider desirable.

On a geopolitical question, it could distinguish verifiable facts, strategic interpretations and narratives produced by the parties involved.

This form of neutrality is much more demanding than producing a score of 0.0 on a political map.

THE PARADOX OF ALIGNMENT

AI laboratories nevertheless face a contradiction. A model entirely devoid of values would be almost unusable. We generally expect it not to endorse arbitrary violence, discrimination, fraud or fabrication. We expect it to respect certain safety norms.

In other words, we want it aligned with principles. At the same time, we ask it not to impose an ideology. The boundary is not always simple. Equality before the law may be treated as a broadly universalizable liberal principle. But the level of redistribution needed to produce a “just” society is a genuine ideological disagreement.

Protection from physical harm is a widely shared norm. The point at which security justifies restrictions on liberty is politically contested.

Alignment therefore has to avoid two extremes: complete absence of norms and the elevation of one particular political philosophy into an invisible universal truth.

THE RISK IS NOT ONLY PROPAGANDA

The political significance of LLMs does not necessarily come from their ability to generate slogans. It comes from their role as intermediaries. A traditional search engine presents multiple links for the user to consult. An assistant can select information, summarize it, order it, contextualize controversy and provide a conclusion directly. It therefore possesses framing power.

A question can be presented as economic or moral, as a problem of security or liberty, as an individual choice or collective responsibility. Framing influences how a problem is understood before solutions are even debated.

If hundreds of millions of people use conversational assistants daily to understand politics, economics, history or international relations, the framing choices of these systems become a form of cognitive infrastructure. They do not need to be openly partisan to become politically significant.

WHO WILL DEFINE NORMALITY FOR MACHINES?

During the first phase of the internet, this problem mainly concerned media organizations, search engines and platforms. Generative AI makes it more profound. Google could rank sources. An LLM can synthesize the world. The difference is substantial.

Power no longer lies only in which information is made visible, but in how the explanation itself is formulated. Companies developing these models therefore become, intentionally or not, normative institutions.

They must decide what it means to be balanced, safe, respectful, factual or offensive. Within a pluralist democracy, these decisions can be publicly challenged and subject to competition among companies.

At the global level, the problem is harder. The norms of an American company do not necessarily match the expectations of an Indian, Moroccan, Brazilian or Japanese population.

Conversely, adapting model behavior completely to each country could lead systems to reproduce every local norm uncritically, including norms that conflict with certain fundamental principles. Global AI will therefore live with a permanent tension between universality and pluralism.

TOWARD A PLURALITY OF ARTIFICIAL INTELLIGENCES

The future may not be one global intelligence carrying a single normative culture. The rise of Chinese, American, European, Indian and regional models may produce genuine plurality.

The Stanford AI Index 2026 already shows that American leadership in notable models coexists with very substantial Chinese scientific strength and rapidly converging performance.

Gulf states are investing in infrastructure and models. Europe is attempting to build technological capacity governed by its own legal norms. India has the linguistic, demographic and technological scale that makes the development of systems tailored more closely to its realities almost inevitable. African and Latin American initiatives are also seeking to reduce dependence.

This diversification may introduce new risks of fragmentation. It may also make more visible a reality that was always present: no universal intelligence can be culturally located nowhere.

A WORLD WITHOUT A POLITICAL CENTER

Human politics is geographic because societies have different histories. It is social because entrepreneurs, farmers, engineers, civil servants and economically insecure workers do not observe the same parts of the system.

It is generational because people do not grow up through the same crises. It is religious because not all societies separate the spiritual from the political in the same way. It is economic because property, income and material security reshape priorities. It is territorial because metropolitan centers and peripheral regions do not experience globalization in the same way. It is now digital because political communities cross borders.

The World Values Survey has shown for decades that modernization profoundly transforms values, but historical and cultural legacies remain powerful.

Humanity therefore does not occupy one political line. It inhabits several overlapping, evolving and sometimes conflicting spaces of values. Artificial intelligence does not erase this geography. It absorbs it. Grok is not Texan. ChatGPT, Claude and Gemini are not Californian. DeepSeek and Qwen are not Chinese in the way citizens possess national identities. No model believes in a doctrine and no model votes. But every system is built somewhere. From data produced somewhere. By institutions located somewhere. Under rules developed somewhere. In languages whose concepts come from somewhere. And these systems are now used everywhere.

The great political question of artificial intelligence is therefore probably not whether machines will become left-wing or right-wing. It is far more difficult.

How can systems intended to serve different societies be built without transforming the values of one society into a universal standard?

How can common principles be preserved without erasing cultural pluralism?

How can a model distinguish a fact from the interpretation surrounding it?

How can it understand several political traditions without becoming the simple mirror of each one?

And, above all, who will decide where the boundary lies between neutrality, pluralism and universal principles?

The geography of ideologies is not disappearing in the age of artificial intelligence.

For the first time, it is entering the machines.

MAIN SOURCES AND BIBLIOGRAPHICAL REFERENCES

The bibliography below distinguishes works of political theory, major comparative datasets on values and regimes, regional public-opinion research and recent studies devoted to political bias in large language models. The inclusion of these institutions does not imply that they share identical definitions of democracy, ideology or bias. On the contrary, their methodological differences are an important part of the subject.

POLITICAL THEORY AND THE HISTORY OF IDEOLOGIES

Norberto Bobbio — Left and Right: The Significance of a Political Distinction. University of Chicago Press. Bobbio examines the persistence of the left-right divide and places particular emphasis on attitudes toward equality. The work is used here to distinguish the conceptual persistence of left and right from their historically changing content.

Michael Freeden — Ideologies and Political Theory: A Conceptual Approach. Oxford University Press, 1996. Freeden develops a “morphological” approach to ideology in which political concepts derive meaning from their combination and hierarchy within a doctrine. This approach directly supports the central argument of this article that ideologies should be analyzed as configurations rather than simple positions on a line.

Andrew Heywood — Political Ideologies: An Introduction. Palgrave Macmillan. A standard reference for the comparative study of liberalism, conservatism, socialism, anarchism, nationalism, fascism, feminism, ecologism and religious ideologies.

Seymour Martin Lipset and Stein Rokkan — Party Systems and Voter Alignments: Cross-National Perspectives, 1967. A classic work on the way historical conflicts involving class, religion, center-periphery relations and church-state relations become crystallized in party systems. It remains essential for understanding the social geography of voting.

Benedict Anderson — Imagined Communities: Reflections on the Origin and Spread of Nationalism, 1983. A central reference on the historical construction of the nation as an imagined political community.

Ernest Gellner — Nations and Nationalism, 1983. An analysis of the relationship between industrialization, culture and the formation of modern nation-states.

Robert O. Paxton — The Anatomy of Fascism, 2004. Used to distinguish historical fascism from conservatism, ordinary nationalism and authoritarianism in the broader sense.

Roger Griffin — The Nature of Fascism, 1991. A major work in comparative scholarship on fascism, particularly regarding the concept of national regeneration.

Cas Mudde and Cristóbal Rovira Kaltwasser — Populism: A Very Short Introduction. Oxford University Press, 2017. Used for the definition of populism as a “thin-centered” ideology opposing the people and elites and capable of attaching itself to either left- or right-wing doctrines.

VALUES, CULTURES AND GLOBAL POLITICAL GEOGRAPHY

World Values Survey — Findings & Insights; Inglehart-Welzel Cultural Map; Wave 7 and historical series. The WVS is one of the main international databases for comparing attitudes toward religion, family, authority, autonomy, democracy, trust and social values. This article relies in particular on the dimensions “traditional versus secular-rational values” and “survival versus self-expression values.” The WVS emphasizes both changes associated with socioeconomic development and the persistence of distinct cultural traditions.

Ronald Inglehart and Christian Welzel — Modernization, Cultural Change, and Democracy: The Human Development Sequence. Cambridge University Press, 2005. A major theoretical work connecting material development, existential security, changes in values and the demand for autonomy.

Pew Research Center — Religious Nationalism Around the World, 2025. A 36-country survey used to document the international diversity of relationships among religion, nationhood, political leadership and legislation.

Pew Research Center — Right-Wing Populism in the Decade Since Brexit, 2026. The study compares several European right-wing populist parties and documents differences in support according to education, as well as certain patterns related to gender and religiosity.

Pew Research Center — Beyond Red vs. Blue: The 2026 Political Typology, 2026. A survey of more than 10,000 US adults using thirty questions to construct nine distinct political groups. It is used here to illustrate the internal diversity of both Democratic and Republican coalitions and the importance of generational differences.

OECD — Trends Shaping Education 2025 and work on Trust and Democracy. These studies help place polarization, institutional trust and support for political extremes within a comparative framework.

DEMOCRACY AND POLITICAL REGIMES

V-Dem Institute — Democracy Report 2026, 10th Edition. The report applies the Regimes of the World methodology and distinguishes, among other categories, liberal democracies, electoral democracies, electoral autocracies and closed autocracies. Under this method, 92 autocracies and 87 democracies were identified at the end of 2025, with approximately 74% of the world population living in regimes classified as autocratic. These figures are used with explicit recognition that they are methodological outputs rather than ideological classifications.

International IDEA — The Global State of Democracy 2025: Democracy on the Move. The report analyzes 174 countries according to several dimensions — representation, rights, rule of law and participation — rather than reducing democracy to a single index. It reported that 94 countries had experienced significant decline in at least one dimension between 2019 and 2024.

International IDEA — Global State of Democracy Indices Methodology: Conceptualization and Measurement Framework. This documentation is important for understanding that democracy indices rest on explicit conceptual and methodological choices and that different organizations may produce different classifications without one being automatically reducible to an absolute truth.

MOROCCO, THE ARAB WORLD AND NORTH AFRICA

Arab Barometer — Morocco: Public Opinion Report 2024, Wave VIII. Survey conducted between December 11, 2023 and January 30, 2024 among more than 2,400 Moroccans. It covers the economy, institutions, freedoms, democracy, international relations, gender and the environment. The study particularly highlights differences in attitudes associated with socioeconomic status.

Arab Barometer — Wave VIII Morocco Report. The report indicates, among other findings, that 73% of respondents agreed that democracy may have problems but remains preferable to other systems, while 68% considered a multiparty parliamentary system appropriate or very appropriate for Morocco. Other questions simultaneously reveal openness to stronger concepts of authority or a political role for religion, confirming the coexistence of several normative frameworks.

Arab Barometer — Political System Preferences in the Middle East and North Africa 2023–2024, published in 2025. Comparative analysis across several regional countries distinguishing abstract support for democracy, preference for parliamentary institutions, demand for governmental effectiveness and possible acceptance of more concentrated forms of authority.

Afrobarometer — Morocco Democracy Scorecard, 2024. A complementary source on Moroccan attitudes toward democracy, institutions, freedoms and corruption.

THE GLOBAL GEOGRAPHY OF ARTIFICIAL INTELLIGENCE

Stanford Institute for Human-Centered Artificial Intelligence — AI Index Report 2026. A key reference for the industrial and scientific geography of AI. Stanford reports that private industry produced more than 90% of the models classified as notable in its 2025 sample; that the United States remains the leading producer of notable models while China dominates several research indicators, including publications, citations and patents; and that the performance gap between top American and Chinese models has narrowed considerably.

The report also documents the geographic concentration of infrastructure, funding and frontier-model development, which is central to the question raised in this article: a global technology remains produced within a relatively limited number of institutional ecosystems.

POLITICAL AND IDEOLOGICAL BIAS IN LLMS

Mats Faulborn, Indira Sen, Max Pellert, Andreas Spitz and David Garcia — “Only a Little to the Left: A Theory-grounded Measure of Political Bias in Large Language Models,” Proceedings of ACL 2025. DOI: 10.18653/v1/2025.acl-long.1529. A major study for this article. The authors criticize the Political Compass Test as an unvalidated instrument for measuring LLM political preferences, develop a method grounded more directly in political theory and analyze 88,110 responses from eleven models. They observe that several instruction-tuned models display somewhat leftward tendencies, but that measurements can be unstable and that the Political Compass may exaggerate bias.

Sadia Kamal et al. — “A Detailed Factor Analysis for the Political Compass Test: Navigating Ideologies of Large Language Models,” 2025. The authors statistically examine the Political Compass and show, among other findings, that prompt design and fine-tuning can significantly influence measured positions. The study is used to emphasize that an LLM political score is not independent of the experimental protocol.

Yejin Bang, Delong Chen, Nayeon Lee and Pascale Fung — “Measuring Political Bias in Large Language Models: What Is Said and How It Is Said,” ACL 2024. DOI: 10.18653/v1/2024.acl-long.600. The authors propose a method distinguishing ideological content from lexical style and argue for fine-grained, explainable evaluation rather than simple left-right labels.

Daniel Wang et al. — “Measuring stereotype and deviation biases in large language models,” Scientific Reports, 2026. The study examines several forms of bias in profiles generated by different LLMs. Within its experimental design, liberal political affiliations are frequently overrepresented relative to the reference distributions used. This article does not generalize that observation to all model behavior but uses it as evidence that politically asymmetric distributions can appear.

LANGUAGES, CULTURES AND IDEOLOGICAL VARIATION IN LLMS

Chadi Helwe, Oana Balalau and Davide Ceolin — “Navigating the Political Compass: Evaluating Multilingual LLMs across Languages and Nationalities,” Findings of ACL 2025. The authors evaluate fifteen multilingual models while varying language and nationality in a framework covering the fifty most populous countries. Their principal finding relevant here is that language strongly influences measured political orientation.

Daniil Gurgurov, Katharina Trinley, Ivan Vykopal, Josef van Genabith, Simon Ostermann and Roberto Zamparelli — “Multilingual Political Views of Large Language Models: Identification and Steering,” 2025. A study of several model families across fourteen languages, demonstrating substantial differences across languages, models and paraphrases, while also showing that measured orientations can be experimentally steered.

Julia Watson, Sophia S. Lee, Barend Beekhuizen and Suzanne Stevenson — “Do language models practice what they preach? Examining language ideologies about gendered language reform encoded in LLMs,” COLING 2025. The study shows that linguistic ideologies can emerge in model lexical choices even when the context is not explicitly framed as political.

Fei Gao and Binhua Wang — “A tale of two ‘contexts’: ideological differences in the translations of UN political speeches by human interpreters and by ChatGPT4o,” Humanities and Social Sciences Communications, 2026. This research is used to illustrate how translation, framing and ideology can interact when a model transfers political discourse across languages.

GEOPOLITICS AND NATIONAL MODEL ECOSYSTEMS

Andre G. C. Pacheco, Athus Cavalini and Giovanni Comarela — “Echoes of power: investigating geopolitical bias in US and China large language models,” Humanities and Social Sciences Communications, 2026. DOI: 10.1057/s41599-026-06577-6. The study compares GPT-4o and DeepSeek-R1 on several international affairs topics and observes differences in geopolitical framing. It is used cautiously as evidence that models developed in different national environments may behave differently, not as proof that all American or Chinese models automatically reproduce the positions of their country of origin.

ADAPTATION TO USER PREFERENCES

Anderson Luis Bento Soares et al. — “LLMs are ideological chameleons: personalized echo chambers in the Brazilian political context,” Scientific Reports, 2026. The study examines how LLMs can adapt their responses to the political profile expressed by users and explores the risk of personalized reinforcement of existing beliefs. It complements the question of fixed bias: a system may not only display its own orientation under certain experimental conditions but also alter its behavior according to the user.

GENERAL METHODOLOGICAL CAUTION

None of the studies on political bias in LLMs allows researchers to attribute ideology to models in the psychological or sociological sense. Models possess neither beliefs nor political interests. Research measures distributions of responses obtained under specific conditions.

Results depend on the model, model version, post-training, language, prompt, context, temperature, output format, classification method and date of testing.

A political map of an LLM should therefore be treated as an experimental snapshot of its behavior under a specific protocol, not as a measurement of a stable conviction.

That limitation is precisely what makes the phenomenon interesting.

The ideology of machines may not exist as consciousness.

The political geography of their answers already does.