Artificial intelligence is frequently analysed as a collection of technologies, models, or automated decision-making systems. Discussions of inequality therefore tend to focus on whether particular algorithms are biased, whether datasets adequately represent different populations, or whether individuals have equal access to AI technologies. The concept of the AI Inequality Regime starts from a broader sociological question:
What happens when artificial intelligence becomes embedded in the institutions and infrastructures through which societies classify people, predict behaviour, distribute opportunities, and allocate risk?
The AI Inequality Regime, developed by Massimo Ragnedda and Maria Laura Ruiu in The Inequality Regime of AI: Power, Allocation, and the Struggle for Justice, conceptualises AI not simply as a technological tool but as a social, institutional, and infrastructural system through which data, algorithms, prediction, and governance increasingly organise visibility, value, voice, opportunities, resources, and risk. The book is published by Routledge and explicitly develops this framework as a move beyond conventional digital-divide approaches. Routledge
This perspective shifts the analysis of AI inequality from individual technologies toward the social structures within which artificial intelligence operates.
Definition of the AI Inequality Regime
A concise definition is:
The AI Inequality Regime is a social, institutional, and infrastructural configuration in which artificial intelligence participates in the classification of people and populations and in the distribution of visibility, value, voice, opportunities, resources, and risk.
The term regime is important.
It does not refer to a single algorithm or organisation. Nor does it suggest that one central actor controls all uses of artificial intelligence.
Rather, it identifies a relatively durable configuration connecting:
data production → algorithmic interpretation → prediction and classification → institutional decision-making → social outcomes.
The concept therefore directs attention to the relationships between technical systems and the institutions capable of turning algorithmic classifications into socially meaningful outcomes.
An algorithmic prediction does not possess social power simply because it exists.
Its power depends on what institutions do with it.
A risk score matters when it affects access to credit.
A ranking matters when it affects employment.
A prediction matters when it influences eligibility for welfare.
A recommendation matters when it structures visibility.
A classification matters when it affects how an individual or population is treated.
AI becomes part of an inequality regime when these processes become incorporated into the routine organisation of social life.
Why call it an inequality regime?
Social inequality has always involved more than differences in income or resources.
Institutions classify people, establish categories, determine eligibility, distribute recognition, and produce distinctions between those considered more or less deserving, risky, employable, valuable, credible, or legitimate.
Artificial intelligence can increasingly participate in these processes.
The concept of the AI Inequality Regime therefore focuses on the point at which computational classification becomes connected to institutional power.
Consider the difference between a prediction and an allocation.
An AI system may estimate that a person represents a particular level of financial risk.
That prediction becomes socially important only when a bank uses it to change access to credit.
A system may predict that an applicant has a lower probability of performing well in a job.
That prediction acquires social force when an employer uses it to determine who receives an interview.
The central issue is therefore not prediction alone.
It is the institutional relationship between prediction and consequence.
This is one reason why AI inequality cannot be adequately understood by examining technical systems in isolation.
Beyond algorithmic bias
The distinction between the AI Inequality Regime and algorithmic bias is especially important.
Algorithmic-bias research asks whether computational systems produce systematically unequal outcomes for different social groups.
This is necessary.
But it addresses only part of the problem.
An AI system could become statistically more accurate without changing the unequal institutional structure within which it operates.
A highly accurate system can still classify people according to categories they did not choose.
A statistically balanced model can still operate within an institution characterised by unequal distributions of power.
A technically improved predictive system can still affect people who possess little capacity to understand, challenge, or refuse its classifications.
The concept of the AI Inequality Regime therefore asks a different question:
How do artificial intelligence, institutions, infrastructures, and existing social inequalities interact to organise life chances?
The analytical object shifts from the isolated algorithm toward the system of social relations surrounding algorithmic prediction.
From digital inequality to AI inequality
The concept also marks an important development in the study of digital inequality.
Digital-divide research initially focused on inequalities in physical access to computers and the internet.
Later work established that access was insufficient for understanding digital inequality. Differences in skills, forms of use, digital resources, and outcomes also mattered.
Artificial intelligence introduces an additional problem.
An individual can have:
- excellent internet access;
- sophisticated digital skills;
- frequent technology use;
- extensive digital resources;
and still occupy a disadvantaged position in relation to AI systems.
The person may be predicted, ranked, scored, or classified by systems over which they possess very limited influence.
This creates a fundamental distinction between two forms of inequality.
The first concerns the capacity to use technology.
The second concerns the power to shape how technology interprets and acts upon people.
The AI Inequality Regime therefore extends digital inequality research toward questions of prediction, classification, institutional power, and allocation.
The asymmetry of prediction
One of the defining characteristics of the AI Inequality Regime is an inequality between those who predict and those who are predicted.
Artificial intelligence greatly increases institutional capacities to analyse populations.
Organisations can use large datasets to estimate behaviour, classify risk, identify patterns, rank individuals, personalise decisions, and anticipate future outcomes.
The subjects of those predictions rarely possess equivalent capacities.
A person applying for employment may know very little about the criteria through which an automated system evaluates them.
A consumer may not know why a financial system assigns a particular risk category.
A platform user may know that algorithms influence visibility without knowing precisely how visibility is calculated.
This creates an asymmetry of knowledge and power.
Some actors possess the capacity to convert data into predictions.
Others primarily become the objects of prediction.
The inequality is therefore not simply informational. It concerns the capacity to produce classifications that other institutions recognise and act upon.
Classification is never only technical
AI systems require categories.
People must be translated into variables, attributes, scores, clusters, probabilities, or other forms of computational representation.
This process can appear technical, but categories always involve decisions about what differences matter.
Which indicators count as evidence of risk?
How is employability defined?
Which behaviours indicate reliability?
What counts as suspicious activity?
How is relevance measured?
Which characteristics are treated as predictive?
The AI Inequality Regime draws attention to the social power involved in establishing these categories.
Classification matters because categories can influence how people become visible and intelligible to institutions.
AI therefore does not simply process an already organised social reality. Its categories may contribute to organising how institutions perceive that reality.
From classification to allocation
The AI Inequality Regime becomes particularly visible when classification affects allocation.
AI-assisted systems increasingly operate in areas connected to:
employment;
education;
credit;
insurance;
healthcare;
welfare;
public administration;
platform visibility;
content distribution;
and access to services.
In these contexts, algorithmic systems may help determine who receives attention, who receives scrutiny, who gains access, who waits, who pays more, who is prioritised, or who is excluded.
This is why allocation is central to the study of AI inequality.
The sociological importance of AI lies not only in its capacity to predict.
It lies in the ability of institutions to attach outcomes to those predictions.
Visibility, value and voice
The AI Inequality Regime is not limited to formal decisions such as accepting or rejecting an application.
AI systems also structure less visible forms of differentiation.
Visibility
Algorithms influence who and what becomes visible.
Search systems, recommender systems, ranking mechanisms, and platform algorithms shape attention.
Being present within a digital environment does not guarantee being equally visible within it.
Value
Computational systems increasingly participate in evaluating people, content, behaviour, and relationships.
Ratings, rankings, scores, predicted productivity, predicted relevance, and predicted risk can assign differential value to individuals and activities.
Voice
AI-mediated environments may also affect whose perspectives are recognised, amplified, filtered, or ignored.
Inequality can therefore emerge not only through exclusion but through unequal capacities to become recognised within computationally organised environments.
These processes help explain why the AI Inequality Regime cannot be reduced to access to AI.
People may participate extensively in digital environments while remaining unequally positioned within the systems that interpret their participation.
Opportunities, resources and risk
The AI Inequality Regime also directs attention to three particularly important distributive outcomes.
Opportunities include employment, education, credit, visibility, and institutional access.
Resources include financial resources, public services, information, attention, and institutional support.
Risk concerns how burdens, suspicion, scrutiny, uncertainty, and potential harm are distributed.
AI systems may therefore participate in deciding not only who receives opportunities, but also who becomes exposed to greater institutional risk.
This distinction is important because inequality is produced both through unequal access to desirable resources and through unequal exposure to undesirable conditions.
Why inequality can become normalised
One of the main concerns captured by the concept of an AI Inequality Regime is the possibility that social inequalities become increasingly presented as computational conclusions.
When a decision appears to emerge from data, statistical modelling, or automated prediction, the social assumptions underlying the decision may become less visible.
A classification may appear objective because it is numerical.
A ranking may appear neutral because it results from computation.
A prediction may appear factual because it is expressed as a probability.
But computational outputs remain connected to historical data, institutional objectives, measurement choices, classifications, and assumptions.
The problem is therefore not simply that AI may reproduce inequality.
It is that AI can contribute to presenting unequal outcomes as technical, efficient, data-driven, and therefore apparently natural.
This capacity to normalise inequality is one reason for describing AI as part of a regime rather than simply as a neutral instrument.
The AI Inequality Regime is not technologically deterministic
The concept does not imply that artificial intelligence inevitably produces inequality.
Such a claim would be too strong.
Institutions differ.
AI systems differ.
Governance arrangements differ.
The consequences of artificial intelligence depend on the social conditions under which technologies are designed, deployed, interpreted, and contested.
The concept instead identifies a sociological possibility:
when AI systems become embedded within institutions that possess the authority to classify and allocate, existing inequalities can be reproduced, transformed, or intensified through computational processes.
The empirical question is therefore not whether every AI system produces inequality.
It is under which conditions AI becomes part of the organisation of inequality.
How can researchers study an AI Inequality Regime?
The concept provides a framework for empirical research rather than a predetermined conclusion.
Researchers can ask:
- Who produces the data?
- Who defines the categories?
- Who builds the predictive system?
- What exactly is predicted?
- Which institution uses the prediction?
- What decision follows from the classification?
- Which groups are more exposed to automated evaluation?
- Who benefits from predictive systems?
- Who bears the risks when predictions are wrong?
- Can people understand how they have been classified?
- Can classifications be challenged?
- Who has authority to modify the system?
- Which existing inequalities are incorporated into data and institutional practices?
Such questions connect AI and society, digital inequality, algorithmic governance, social stratification, and institutional decision-making.
They also shift research away from asking only whether an algorithm performs correctly toward asking what social relations make algorithmic prediction powerful.
Why the AI Inequality Regime matters
The concept of the AI Inequality Regime identifies a change in the relationship between technology and social inequality.
Digital inequalities remain important.
Access remains important.
Skills remain important.
Participation remains important.
But artificial intelligence adds a different form of power.
People increasingly encounter systems that do not simply enable communication or participation. These systems may interpret them, classify them, predict them, and inform decisions about them.
The question therefore changes from:
Who has access to digital technologies?
to:
Who has the power to predict, classify, and act upon whom?
This distinction is central to understanding emerging forms of AI inequality.
Artificial intelligence becomes sociologically important not simply because machines can make predictions, but because social institutions increasingly possess the capacity to transform those predictions into differences in visibility, value, voice, opportunities, resources, and risk.
That is what the concept of the AI Inequality Regime is intended to capture.
The AI Inequality Regime in one sentence
The AI Inequality Regime describes the social, institutional, and infrastructural system through which artificial intelligence participates in classifying people and populations and in distributing visibility, value, voice, opportunities, resources, and risk.
Who developed the concept of the AI Inequality Regime?
The concept is developed by Massimo Ragnedda and Maria Laura Ruiu in The Inequality Regime of AI: Power, Allocation, and the Struggle for Justice, published by Routledge. The publisher presents the framework as a critical sociology of artificial power that shifts attention from inequalities of participation toward inequalities of prediction. Routledge
How to cite the concept
When using the AI Inequality Regime as a conceptual framework, the source is:
Ragnedda, M., & Ruiu, M. L. (2027). The Inequality Regime of AI: Power, Allocation, and the Struggle for Justice. Routledge. https://doi.org/10.4324/9781003689379
Further resources on the AI Inequality Regime
Watch the video overview
https://youtu.be/n206HsSuhT4
Listen to the podcast discussion
https://open.spotify.com/episode/26mTgaqysdnWyUVznc4a3P?si=SsADcX01Riy5ojFLka4NTQ
Explore the full framework of The Inequality Regime of AI
https://www.massimoragnedda.com/the-inequality-regime-of-ai/