10  Fault Lines: Inequality and the Future of Quantitative Social Science in the Age of Artificial Intelligence

Alicia Eads
Department of Sociology and Centre for Industrial Relations and Human Resources, University of Toronto

Fedor A. Dokshin
Department of Sociology, University of Toronto

Abstract: We focus on inequality especially three key faultlines: 1) Equalizer Paradox; 2) Democratization Paradox; 3) Output Paradox. We argue that the social sciences are uniquely positioned among fields being transformed by AI to understand its distributional effects. We highlight the potential for decisive action to ameliorate its worst impacts.

AI usage statement: During the preparation of this chapter, the authors used Anthropic’s Claude (Opus 4.7, via claude.ai) at several stages of the writing process: to stress-test the paper’s argument and framework, and for editorial feedback on organization, flow, and clarity. The authors undertook all substantive conceptual development and made all argumentative and interpretive decisions. The authors take full responsibility for the content of this publication, including its claims, citations, and conclusions.

We thank Sharla Alegria for reading an earlier draft and providing valuable feedback that improved this paper.

10.1 Introduction

The current moment in AI deployment within the quantitative social sciences reflects science fiction writer William Gibson’s observation that “the future is already here — it’s just not very evenly distributed.” Large language models (LLMs) and related tools lower the cost of tasks that once required substantial specialized training or labor, including data cleaning, exploratory analysis, and writing and debugging code (Messing and Tucker 2026). Enthusiastic early adopters report dramatic transformations to their workflows, productivity gains, and high expectations for future research opportunities (Van Noorden and Perkel 2023). One reading of Gibson’s quote is that the future these early adopters are living is just around the corner for everyone else. The promise on offer is not just faster research but a more equal field. If AI can substitute for resources that the best-funded researchers have and others lack, such as capable research assistants, computing infrastructure, and methodological collaborators, then its broad deployment will narrow the gap between the well-resourced and the rest. This optimism about a future with a more productive and more broadly accessible research ecosystem is common in discussions of AI deployment in science, but it misses the deep sociological insight in Gibson’s quote. Positive and negative effects of frontier technologies are always experienced unevenly. In other words, the uneven distribution of the future is structural, not transitional. Equipped with structural theories of stratifying processes, the social sciences are uniquely positioned among fields being transformed by AI to understand its distributional effects and, if we act decisively, to ameliorate its worst impacts.

Here, we apply a sociological lens, developed through decades of research on earlier technological transformations and their impacts, to develop expectations about ways that AI will affect inequality in quantitative social science. Although AI is new, inequality is reproduced through familiar sociological mechanisms: the status-structured acceptance and evaluation of AI use (DiMaggio and Hargittai 2001; van Deursen and Helsper 2015), the deskilling and polarization of methodological labor (Braverman 1974; Autor et al. 2003), and cumulative (dis)advantage (Merton 1968). Reading these inequalities as structural rather than individual aligns with a sociology of AI that locates AI’s inequalities in institutions, and calls on the field to study them as such (Alegria and Yeh 2023; Joyce et al. 2021).

We organize our discussion around a three-level typology of inequality elaborated by researchers of the digital divide, consisting of inequality in: (1) access to the technology, (2) use of the technology, and (3) returns to the use of the technology (DiMaggio and Hargittai 2001; Hargittai 2002; van Deursen and Helsper 2015). First-level inequality concerns access to the technology itself: who can obtain it, who can afford to maintain it, and whether their access is autonomous and uninterrupted rather than shared, monitored, or constrained. Advantaged groups tend to be overrepresented among early adopters (Rogers [1962] 2003), and although access disparities can persist (Attewell 2001; van Deursen and van Dijk 2019), they tend to narrow as a technology becomes more affordable and widely accepted (O’Shaughnessy 2021). Yet even as first-level gaps close, second- and third-level inequalities can still reproduce pre-existing class, educational, and geographic stratification (Attewell 2001; van Deursen and van Dijk 2019).

Second-level inequality concerns what users do with the technology once they have it, and the socially structured conditions that shape that use. These conditions include the autonomy of a person’s use and the social support they can draw on, each unequally distributed (DiMaggio and Hargittai 2001). In the internet case, skill was the decisive condition. Growing access did not ameliorate digital inequality, because the internet skills that structure the benefits of access were themselves unequally distributed (Hargittai 2002; Büchi et al. 2015). For AI, as we discuss in the first fault line, a key condition is whether a researcher’s use is socially accepted rather than penalized. Access and use are thus necessary for deriving benefits from a technology, but not sufficient.

Third-level inequality concerns the differential returns that users translate access and use into. Even when access and skill profiles are broadly comparable, the same activity or output yields different recognition and material benefits for differently positioned people. The inequality in returns stems partly from how the same work is evaluated, since identical output is not always judged identically, and partly from how social position converts a given activity into advantage. For instance, even where internet access was near-universal and usage broadly similar, higher-status users extracted greater offline returns, including better jobs, cheaper purchases, and more useful institutional information, from the same online activity (van Deursen and Helsper 2015). Keeping the levels analytically distinct is essential for understanding why the diffusion of AI will not, on its own, produce more equal outcomes.

Leveraging the analytic distinction between these three levels of inequality, we advance our argument through three fault lines facing the field of quantitative social science (Table 10.1). The first concerns access, adoption, and evaluation of AI tools. Here, unequal resources, reputational penalties, and double standards leave the existing hierarchy intact, inequality is reproduced. The second fault line concerns methodological stratification within the field. Here, AI is expected to drive labor market polarization that has been consistently observed during technological transformations in other industries (Braverman 1974; Kellogg et al. 2020). The change tends to deskill the overwhelming share of the workers, while cementing a small upper tier of specialists with valued expertise. In quantitative social science this may look like a democratization of many applied methods with lower expectation of technical knowledge, alongside an emergence of a smaller, upper tier of technical experts who effectively command AI (Lin and Sohail 2026). The third fault line concerns cumulative advantage in the context of rising output, where productivity gains translate unequally into recognition and resources. Here, small initial differences compound over a career, and existing inequality is amplified. However, none of this is determined by the technology itself (DiMaggio and Garip 2012; Joyce et al. 2021). Each fault line reflects choices the field is now making, often implicitly, about what counts as scholarship and who gets to do it. We conclude by naming two key choices the field now faces.

Table 10.1: AI operating at the fault lines.
Fault line Core dynamic Levels Mechanisms Inequality outcome
Fault Line 1 — Equalizer Paradox: Status-Structured Access, Use, and Evaluation Existing status hierarchy redeploys onto AI; the same researchers stay ahead Access, use, & returns Flexible resources & status-structured use and reward Reproduced
Fault Line 2 — The Democratization Paradox: Field Stratification Field splits into an upper-level, directing-AI layer and a lower-level serving-AI layer; new division of methodological labor Use Deskilling & polarization Restructured
Fault Line 3 — The Output Paradox: Cumulative Advantage in an Era of Rising Output Small initial differences compound into large ones; gaps widen over time Returns Cumulative advantage Amplified

10.2 Equalizer Paradox: Status-Structured Access, Use, and Evaluation

In an optimistic view, AI is an equalizer. Broad adoption of AI, in this view, will narrow the gap between those who currently have advantages and those who do not. One advantage is resources for research, which include funding to hire capable research assistants, computing infrastructure, and programming and methodological expertise. AI can substitute for many of these resources or significantly reduce their costs, so that the resource gap between elite and non-elite researchers, senior faculty and graduate students, and the Global North and Global South could decrease (Messing and Tucker 2026).

But there are reasons to be suspicious of this optimistic view. Although a limited-use monthly AI subscription is inexpensive, the kinds of AI use researchers are moving toward, including running AI models in premium tiers and sustaining agentic workflows, would already strain the budgets of many social scientists. The costs may also rise as AI companies consolidate their business models and the early stage of nearly unlimited support from investors ends (Jin 2025).

Cost, however, is not the only thing that stratifies access. Because these tools are owned and controlled by private firms, access is conditional and revocable. Models can be retired, altered, or moved behind higher-priced tiers at a provider’s discretion, much as the platforms that computational social science once relied on restricted their data interfaces and produced what Freelon (2018) describes as a post-API age of scarce and uneven data access. What Wagner (2023) terms independence by permission, in which a researcher’s access depends on continuing corporate approval, increasingly characterizes the model layer as well. Research on AI itself has already concentrated among large technology firms and elite universities (Ahmed and Wahed 2020; Bail 2024; Besiroglu et al. 2024) and some of the most capable models are highly restricted. Anthropic, for instance, limited access to its most advanced model, Claude Mythos Preview, on security grounds to a small group of large technology firms and critical-infrastructure organizations rather than the research community at large (Anthropic 2026).

Even the models all researchers can reach are not equally capable for everyone. Large language models perform better in high-resource languages that are well represented in the training data, such as English or Spanish, than in low-resource languages where training data is scarce, such as Swahili or Tamil, even when those languages have comparable or larger speaker populations (Wang et al. 2025). A researcher working in Swahili or Tamil is using a weaker instrument than one working in English, even with identical access.

These are all first-level inequalities, concerning who can afford the technology, who is granted access to it, and how well it performs for the researcher using it. Public policy initiatives aim to address access disparities by providing shared infrastructure that is open to a wider community, including the U.S. National AI Research Resource (Parashar et al. 2023), the European Union’s EuroHPC, and Canada’s Sovereign Compute Infrastructure Program. These initiatives are laudable, but they address only affordability, lowering the cost of access without changing who controls the models or how well they serve different researchers. Even if every first-level inequality were resolved, deeper problems would remain.

In the second use-of-technology level, what divides AI users is whether a researcher’s use of it is accepted, and that acceptance is status-structured.1 There is a reputational penalty for AI use, and it is distributed unevenly across the field. Researchers in many fields now face a social penalty simply for using AI (Reif et al. 2025), but it is steeper for those whose competence is already viewed with suspicion (Gai et al. 2025). Such suspicions are well documented in quantitative and technical domains, where women, scholars from underrepresented racial and ethnic minorities, and others outside the field’s demographic core are presumed less competent even when their records are comparable (Correll and Ridgeway 2006; Moss-Racusin et al. 2012; Eaton et al. 2020; Tang et al. 2025). In a field study of software engineers, the competence penalty for AI use was roughly twice as large for women as for men, and older workers were penalized more than younger ones (Gai et al. 2025). The same AI use can be read as a productivity tool in the hands of a rational, time-constrained expert or as a crutch masking the incompetence of a dilettante. Broader cultural beliefs and stereotypes structure the attribution of meaning that others give to observed AI use and, consequently, who decides to use and to disclose their use of AI. Those who anticipate harsher judgment will be more reluctant to adopt it (Correll and Ridgeway 2006; Gai et al. 2025).

Emerging research on gendered AI adoption is an illustrative case. Gender gaps in academic productivity, as measured by publications and citations, persist (Stockemer et al. 2026), which AI adoption could in principle narrow. However, women are about 25% less likely than men to use AI across regions, sectors, and occupations (Cranney et al. 2026; Humlum and Vestergaard 2025). Among academic researchers, male productivity on SSRN grew 6.4% more than female productivity after ChatGPT’s release (Tang et al. 2025). Women in these studies report concerns about being judged as “cheating” or as lacking expertise, concerns that men do not report at comparable rates. The same presumption that has long made women’s quantitative competence suspect (Correll 2001, 2004; Moss-Racusin et al. 2012; Reuben et al. 2014), makes their AI use suspect as well, and makes them more reluctant to take advantage of these powerful tools. The result is a second-level inequality, in which equal access to AI does not produce equal use, because the acceptance of that use is status-structured (DiMaggio and Hargittai 2001; Hargittai 2002).

The third level concerns returns, the recognition and reward that AI-assisted output receives when others evaluate it. This is where AI looks most promising as an equalizer. Non-native English speakers, for example, are penalized in publishing their scholarship: they spend more time on manuscript preparation, they face higher rejection rates, and they receive lower quality ratings for comparable work (Politzer-Ahles et al. 2020; Amano et al. 2023). If this penalty were simply a matter of language, a technology that polishes prose should equalize their evaluation. But it has not (Khan et al. 2025). This will come as no surprise to inequality scholars. The benefit of the doubt in evaluation is not evenly extended (Brewer et al. 2020). It is withheld from the same groups that are judged more harshly, including women and scholars from underrepresented minorities, who are presumed less competent even when their records are comparable (Correll and Ridgeway 2006; Moss-Racusin et al. 2012).

Even among groups that have overcome or ignored the reputational penalty and are adopting AI as much or more than advantaged groups, their output continues to be evaluated by a double standard. Authors in the Global East and non-native English speakers generally have adopted AI at rates equal to or higher than English-speaking Western authors, yet Western authors continue to receive disproportionate benefits (Khan et al. 2025; Kusumegi et al. 2025). The same pattern holds beyond academic publishing. In a study of college admissions essays, lower-SES applicants used AI at higher rates than higher-SES applicants. Yet AI use carried a steeper cost for them: at the same level of detectable use, and with the same credentials, lower-SES applicants faced about twice the reduction in admission odds (Lee et al. 2026). Because lower-SES applicants adopted more and were penalized more, the socioeconomic gap in admissions widened. Measured qualities of the essays explained little of this gap, which points to how the writers were judged rather than to the writing itself. In each case, the disadvantaged adopt more and are rewarded less.

The equalizer promise assumed that what separates advantaged from disadvantaged researchers is skill and resources, so that supplying both through AI would close the gap. It assumed, in other words, that a disadvantaged group could access the technology on the same terms, adopt it on the same terms, and have its output evaluated on the same terms. The reality is that access is conditional and unequally distributed, adoption is depressed by the reputational cost of being seen to use AI, and output is discounted by a double standard in how it is judged. At every level, AI reproduces the inequality it promised to erase.

10.3 The Democratization Paradox: Field Stratification

AI is widely expected to democratize quantitative social science. By lowering the cost of methods that once required substantial technical training, it puts sophisticated analysis within reach of researchers who could not previously afford the expertise those methods demanded. But the same shift is likely to stratify the field along a different axis. As AI absorbs the execution of methods, the premium moves to a narrower capacity, the analytical and methodological judgment needed to choose among methods, to recognize when an output is wrong, and to build or modify the underlying tools. The likely outcome is a polarization between researchers who exercise that judgment and produce methodological innovation and those positioned only to execute pre-specified procedures (Lin and Sohail 2026). What presents itself as democratization is therefore also a stratification of the field, spreading technique widely while concentrating methodological power narrowly.

This is not the first time a technology has reshaped a field’s skill structure. New technologies tend to split previously integrated domains of skilled practice, separating conception from execution, so that a small group retains control over how the technology works and what it can do while a much larger group operates it without the training or opportunity to understand it (Braverman 1974; Kellogg et al. 2020). The 19th-century machinist, exercising craft judgment over how to cut a piece of metal, became the 20th-century CNC operator running a program written by someone else. The same pattern has been documented in the printing industry, where skilled typesetters were progressively replaced by keyboard operators running computerized systems designed by a much smaller elite (Wallace and Kalleberg 1982).

Subsequent work has generalized the lesson of these historical cases. Technology does not deskill uniformly. It polarizes, hollowing out the middle of the skill distribution while the frontier pulls further away and a broad lower-skill layer expands (Autor et al. 2003; Acemoglu and Restrepo 2022). Quantitative social science has already lived through a milder version of this transition. Successive generations of statistical software packaged complex procedures into menus and commands, black-boxing many assumptions and analytical choices that once required explication, and shifting statistical pedagogy from interpretive judgment toward technique execution (Uprichard et al. 2008).

AI tools push this pattern further along the same axis. On one hand, generative AI dramatically reduces the cost of methods that once required substantial technical investment. An AI-assisted researcher can now complete tasks like text analysis, complex data cleaning, web scraping, and sophisticated regression analyses in an afternoon, without the specialized training or a methodological collaborator that would have been required in the past. On the other hand, the same forces raise the methodological frontier. If everyone can train a word embedding model, the marginal contribution of doing so is minimal, and the premium flows to researchers who can build, modify, or critically evaluate the underlying models. The likely outcome for the quantitative social sciences is a version of the polarization documented in earlier domains. The skills required to execute methods become widely accessible, so the layer of researchers who can run sophisticated analyses expands. At the same time, the analytical judgment required to direct those methods becomes the scarce, valued capacity, and a small frontier pulls away. The upshot is democratization at one end and concentration at the other.

What will ultimately determine which side of the polarization a researcher ends up on is not only the distribution of resources, but also the foundational skills acquired before AI became capable of substituting for the activities that build them. Researchers who learned to clean data, write code, or estimate models without AI assistance can recognize when an output is suspicious and can ask the questions that would expose a default that should not have been used or a model that was misspecified. These skills are built precisely through the activities AI can now do for social scientists (Lin and Sohail 2026)2. Research shows that prior domain knowledge functions as a complement to AI rather than a substitute for it (Vendraminelli et al. 2025), and what looks like AI-driven equalization across users risks masking a growing dependence on AI among those without foundational knowledge and skill (Zhang 2026). The popular assumption that digital or AI “natives” should be better positioned to use these tools well is not supported by the evidence (Bennett et al. 2008; Kirschner and De Bruyckere 2017).

Relatedly, discussions of AI in academia have rightly identified skill atrophy as an important concern (Messing and Tucker 2026), but the debate too often frames this risk at the level of individual researchers. Rather than established scholars losing core competencies through AI automation, field-level deskilling is more likely to occur through cohort replacement (Ryder 1965). Established researchers, having acquired foundational skills before the advent of AI substitutes, are well positioned to delegate routine tasks to AI while retaining the discernment required to evaluate and contextualize its outputs. New cohorts, by contrast, if trained in environments where AI displaces rather than supports foundational learning, may never acquire the skills necessary to recognize what AI is doing, or what it is doing badly.

The historical lesson from calculators applies here. When handheld calculators entered American classrooms, parents and educators feared that students would lose fundamental arithmetic skills and the capacity to evaluate whether an answer was even plausible. Yet there was not a systematic decline in computational ability, as long as calculators were introduced after foundational fluency had been established and were integrated into instruction that continued to value conceptual reasoning. The original worries were only vindicated when the technology was allowed to displace rather than supplement the formative stage (Hembree and Dessart 1986; Ellington 2003). The technology supplements skill when foundational fluency is established before the tool arrives, but substitutes for skill when the technology is adopted first.

The result is a structural shift that the three-level inequality framework helps us understand. Skill is not simply unevenly distributed across researchers, it is being polarized. First-level inequalities in resources and access, together with the foundational training a researcher acquired before AI, determine who reaches the frontier, and third-level returns then concentrate among those who do. This is a distinctive second-level inequality of use, in which use itself splits into the exercise of judgment and the execution of pre-specified methods.

10.4 The Output Paradox: Cumulative Advantage in an Era of Rising Output

AI is poised to raise research output across the board. At first glance that looks like a leveling force, more work produced faster, with the largest increases going to those who produced least before. But output is not recognition. productivity is converted into citations, grants, students, and positions through institutions that already sort scholars hierarchically, and the same output yields more reward for some than for others. The paradox is that a rising tide of output does not lift scholars equally. Small initial advantages compound into large ones.

Early evidence confirms that output is indeed increasing. Analyzing 2.1 million preprints across arXiv, bioRxiv, and SSRN, Kusumegi et al. (2025) found that adopting AI accelerated manuscript output and lowered barriers for non-native English speakers (see also He and Bu 2026). Experimental evidence even suggests AI can compress the output distribution, helping lower-skilled writers increase their output most (Noy and Zhang 2023). But these are gains at the level of use and output. Whether they translate into recognition and reward is a third-level question, and the sociology of science gives reason to doubt equal returns. Merton (1968) observed that eminent scientists receive disproportionate credit for work of comparable quality, and that early recognition translates into tangible resources, such as grant funding and publication success, which in turn enable further recognition. Small initial differences amplify into large eventual inequalities (Dannefer 2003; DiPrete and Eirich 2006). Bol et al. (2018) showed the magnitude. Among Dutch early-career applicants, grant winners just above the funding threshold accumulated more than twice as much research funding over the next eight years as near-identical non-winners just below it. The same dynamic falls hardest on non-native English speakers. AI lowers the language barrier that ostensibly has disadvantaged them, and Kusumegi et al. note that the signals that have been used to judge quality, such as language complexity, are becoming unreliable indicators of merit. But as Fault Line 1 showed, removing the surface signal does not remove the disadvantage. Comparable work by disadvantaged groups is still rewarded less. AI lowers the barrier to producing the work without lowering the barrier to being rewarded for it.

A second mechanism compounds the first. As average output rises, the bar against which output is judged rises with it, and it applies to everyone uniformly. Because the bar is set by the most productive, it is hardest on those who produce less for reasons that have nothing to do with the quality of their work, such as caregiving or heavy teaching loads. The same rising standard that rewards the unconstrained penalizes the constrained, even when their work is equally good. This is third-level inequality. Even where access and use converge, the returns do not.

10.5 The Choices Facing the Field

A pre-existing hierarchy shapes who can use AI, how their use is read, and what their output is worth, and the same hierarchy is poised to restructure the field’s division of methodological labor. The distributional consequences of a new technology are shaped by the institutions surrounding it, not determined by the technology itself (Acemoglu and Restrepo 2022; Joyce et al. 2021; Scheidel 2024).

The direction of AI’s effects on inequality depends on choices we are making now. Journals have moved quickly on AI-use disclosure in principle, but implementation lags. He and Bu (2026) found that roughly 70 percent of journals have adopted AI policies, mostly requiring disclosure. Yet, only about one paper in forty that shows evidence of AI use actually disclosed using it. Part of the reason is the penalty described in Fault Line 1. He and Bu attribute much of the non-disclosure to authors’ fear that admitting AI use will invite added scrutiny and reputational harm. No parallel infrastructure has developed for assessing how disclosed AI use is judged. Without it, stronger disclosure enforcement risks amplifying rather than reducing evaluation inequality, giving reviewers more information to act on without changing the standards by which they act.

The field faces a choice about evaluation standards. Publication expectations have been rising for decades, and the rise has fallen hardest on scholars with caregiving responsibilities, heavy teaching loads, and limited collaborative resources (Antecol et al. 2018; Myers et al. 2020). AI accelerates this, locking in the cumulative-advantage dynamic described above. Countering this requires evaluation practices that recognize productivity as conditional on the resources available to produce it, not as a property of the individual scholar. This principle underlies existing reform efforts in research assessment, including the San Francisco Declaration on Research Assessment and the narrative-CV formats now used by several major funders (Hicks et al. 2015; DORA 2012). These decisions, about disclosure infrastructure and evaluation standards, have distributional consequences, and they are being made right now, but too often by default.

Beyond these choices, the inequalities AI produces are themselves a research object the field is unusually equipped to study. We can measure how access to premium AI tools and computing infrastructure is distributed across institutions, regions, and career stages and test whether AI use produces equivalent citation and publication outcomes for different populations. A sociology of AI focused on inequality and structural change already exists (Joyce et al. 2021; Alegria and Yeh 2023), though its empirical attention has so far been primarily on ML’s effects outside the academy. We must extend that attention inward, to how AI is reshaping quantitative social science itself. Few fields being reshaped by AI have the methodological tools to study their own reshaping. Ours does.

  1. This is a departure from the internet’s second-level inequality, which is based on skill differentials. Unequal online skill structured who benefited from access (Hargittai 2002). AI may yet reach that phase, where differences in skillful use structure who benefits, but only once its use is more widely accepted. For now, accepted use is uneven and contested, so the operative second-level constraint is reputational rather than skill-based.↩︎

  2. AI use may pose a social dilemma in which individual outputs improve while the collective diversity of outputs narrows. Doshi and Hauser (2024) found that writers given access to GPT-4 produced stories rated more novel, with the largest gains for less inherently creative writers, but stories produced with AI assistance were measurably more similar to each other. The mechanism is convergence on patterns already optimized in the model’s training data. Researchers without foundational skills and training to recognize and push beyond these patterns may be especially likely to produce work that conforms to the AI standard (Lin and Sohail 2026).↩︎