Why a Working Paper Series on Critical Search Now
Introduction to the Series
This essay introduces Working Papers in Critical Search, a GitHub-published, fast-turnaround venue for scholarship at the intersection of history, computation, and political economy. We argue that transformer-based AI has dissolved the old bottlenecks of digital history — transcription, structuring, coding — and shifted the discipline’s binding constraint to interpretation, design, and judgment. That shift lets historians re-enter longue durée debates about capitalism, society, and change on terms that preserve their social and ecological complexity rather than reducing them to quantitative trends. But it also exposes a publication ecosystem too slow for a research practice in which models and datasets evolve faster than peer review. The series responds with a “white-box” model of computational history in which method and interpretation are inseparable and evidentiary chains from source to claim remain visible, and treats critical search — attention to provenance, bias, and historical context in how knowledge is retrieved and assembled — as both a scholarly method and a public infrastructure.
Digital humanities has always moved faster than its formal venues. Journals, with their long review cycles and disciplinary gatekeeping, struggle to capture the iterative, collaborative, and experimental nature of work in labs where historians, computer scientists, and students co-produce knowledge. The year 2026 marks a hinge moment when rapid advances in transformer-based AI systems collide with universities’ struggle to govern, interpret, and audit them, making “critical search”—the practice of interrogating how knowledge is retrieved, ranked, and assembled, with attention to provenance, bias, and historical context—newly urgent as both a scholarly method and a public infrastructure.
The era of AI portends the transformation of the historical discipline. In previous generations, the promise of digital history was limited to a few expert scholars with metagrants capable of supporting lab-sized research; the founders of the journal have been among those practitioners. What has changed now with AI, as Cameron Blevins has pointed out, is that the old bottlenecks of digital history have largely collapsed. Tasks that once required teams, funding, and technical specialization—transcribing sources, structuring data, writing code—can now be handled by AI systems that combine vision, language, and reasoning.
The implication is profound: the constraint is no longer whether historical materials can be digitized or analyzed, but whether we can ask good questions and translate them into workflows. In that sense, we are all becoming computational historians, not because we all code, but because computation is now embedded in the basic act of working with sources. Any dataset can, in principle, be made machine-readable; any archive can be queried, structured, and modeled. The shift is not just technical but epistemological: the frontier of historical research is moving from access and processing to interpretation, design, and judgment. The emergence of generative AI has altered who can participate in scholarly production. Tools for drafting, translation, and stylistic transformation lower barriers for scholars who have historically been excluded from academic writing: dyslexic researchers, multilingual scholars working outside their first language, and students whose intellectual contributions exceed their fluency in disciplinary prose.
If the bottlenecks have shifted, from access and processing to interpretation and design, then our institutions need to catch up. We are producing more, faster, and in more varied forms than the traditional publication ecosystem was built to handle. The problem is no longer just how to do computational history; it is how to share, evaluate, and circulate it.
This working paper series formalizes that agenda. It provides a venue for rapid, serious, and citable contributions that reflect the pace and diversity of contemporary research. No single format—journal article, monograph, technical report, code or dataset—can accommodate the full range of outputs now emerging from digital humanities labs. A working paper series can. The series is published through GitHub, so that contributions can include not only prose but also code, datasets, and interactive demos. The medium reflects the argument: if digital humanities research is heterogeneous and collaborative, its publication infrastructure should be too.
A working paper series can serve scholars at different career stages and with different goals. For established researchers, it offers a venue to document and cite methods without always diverting the sustained attention that a peer-reviewed digital methods article requires. For early-career scholars the calculus is different: building a publication record in peer-reviewed venues rightly takes priority. For these scholars, a working paper might be a first draft of a piece being prepared for peer review as a way to circulate ideas, get feedback, and establish priority while the formal process runs its course. Or it might stand on its own as documentation of the methods that accompany a contributor’s core historical research. Most history journals have little appetite for extended methods sections, and authors are often asked to cut precisely the technical detail that makes computational and digital work reproducible. A working paper offers a citable home for that material, allowing the journal article to focus on historical argument while the methods remain accessible to readers who want to scrutinize, replicate, or build on them.
An Intellectual Agenda
A distinct body of work has emerged across several labs in recent years, one that demands its own intellectual space. This work sits at the intersection of artificial intelligence, political economy, global history, and the theory of history, and it is driven by a shared methodological problem: how to extract, at scale and with fidelity, information about changing ideas from historical text.
There is, right now, a consequential conversation about the long arc of human history unfolding largely outside the discipline of History. Figures like Steven Pinker, Peter Turchin, and Nassim Nicholas Taleb are advancing sweeping claims about violence, stability, and social change using large-scale quantitative datasets. These arguments are influential precisely because they speak in the language of generalization: they offer answers about what is increasing, declining, or recurring over centuries. And yet, for the most part, historians—especially social historians—have not been central participants in this debate.
That absence matters. Because the history of capitalism, political economy, and social life has never been reducible to counts of wars, prices, or populations alone. Scholars like Maxine Berg, Sven Beckert and Jason W. Moore have shown that large-scale transformations emerge from the interaction of institutions, labor regimes, ecological systems, and cultural meanings that evolve over time (Berg 2021; Beckert et al. 2021; Beckert 2025; Berg and Hudson 2023). Capitalism is not just growth rates; it is plantation economies, imperial networks, and the contested organization of labor and land. Moore, in particular, reminds us that capitalism is always already a way of organizing nature as well as society—binding together energy, environment, and exploitation. Climate change, in this light, is not simply an atmospheric trend but a historical outcome of these intertwined systems. These are social and ecological processes, not just numerical trends.
Purely quantitative approaches, for all their reach, risk a kind of abstraction that flattens these dynamics—relying on incomplete or uneven datasets and mistaking what can be counted for what must be explained. In that sense, they engage in a form of cherry-picking of their own: privileging the measurable while overlooking the structures and experiences that give those measures meaning.
What is needed is not a rejection of scale, but a different way of achieving it.
This is where digital history—especially when augmented by AI—offers a path forward. By combining large-scale datasets with the analysis of language, narrative, and representation, historians can engage these long-run debates on their own terms. We can track patterns across centuries while still attending to how institutions form, how ideas circulate, and how people experience and interpret change. In doing so, we do not simply add nuance to existing claims. We reshape the questions themselves—bringing the full social and historical complexity of the past back into conversations that urgently need it.
Addressing this problem requires a longue durée perspective—decades, even centuries—and a global archival base. The sources are correspondingly expansive: imperial records such as the British Colonial Office Lists and Parliamentary Papers, the archives of the Dutch East India Company, and the administrative records of the Spanish Empire or the full run of the Tropical Agriculturalist and the Der deutsche Kulturpionier; for the modern period, the papers of international organizations, philanthropic foundations, and corporations; for the present, global datasets such as those produced by the IPCC and the UNFCCC. What unites these archives is not an interest in institutional elites alone, but the fact that tracing how societies have understood their relationship to labour, land, and resources, across space and over time, requires reading at a scale that exceeds any individual researcher. Only by assembling and analyzing such large-scale, multilingual archives can we connect the pattern-seeking ambitions of longue durée quantitative analysis to the historically grounded questions posed by Sven Beckert, Maxine Berg, David Harvey, and Jason W. Moore—linking abstract trends to the evolving social, institutional, and ecological processes that produce them.
Exploring New Methods Within a Framework of Accountability
Alongside these research agendas, a parallel body of pedagogical and methodological expertise has developed within digital humanities labs. Scholars are grappling with questions that rarely find a home in traditional publications: what constitutes “good enough” data for historical analysis; how to evaluate clustering and classification methods in ways that are historically meaningful; how to design workflows that integrate humanistic interpretation with machine-assisted analysis. Methods like GraphRAG are reshaping how large-scale text analysis works. At its best, generative AI can support the articulation of ideas without substituting for them. At its worst, it introduces new risks, most notably the fabrication or distortion of facts.
Getting the facts right is only half the challenge. Which tool to use for a historical analysis is a question that is never merely technical. Epistemological and ontological problems bear directly on which facts are selected, which model is trusted, which visualization is offered as proof of incontestability or meaning. These methods questions—of which tactics we choose and why—are almost never engaged in the journals, either historical or digital humanities. As a result, important conversations about how we know what we know remain underpublished, circulating informally in syllabi, lab meetings, and code repositories. This series aims to make that knowledge visible, citable, and cumulative.
Articulated in Guldi’s The Dangerous Art of Text Mining, the theory of “critical search” insists on attention to context, the provenance of data, and the interpretability of results (Guldi 2022, 2024). This series aims to leverage critical search as an orientation for social scientists and humanists who care not only about capitalism but also about the fit of methods to facts, theories, and archives.
As a working papers series, this project aims to publish pedagogy, student papers, and experimental analyses that make their evidentiary chains explicit: readers must be able to see where sources originate, how they have been transformed, how algorithms have been tested for bias, how the dataset’s own limitations have been interrogated, and which specific passages substantiate a given interpretation. In this sense, it advances a “white-box” model of digital history—one in which the analytical steps between raw source and published claim remain visible and open to scrutiny, so that results derived from large-scale data stay accountable to primary sources and legible to humanistic interpretation.
A complementary purpose is to develop and disseminate short-form benchmarks for historically grounded artificial intelligence. These contributions test how well AI systems extract, summarize, and reason about historical materials, with particular attention to questions of evidence, disagreement, and context. By pairing transparent workflows with evaluative criteria, the series aims to establish practical standards for what it means for AI to produce historically credible knowledge.
Historians cannot afford to leave the development of the tools entirely to computer scientists. Source criticism, reading between the lines, attending to what archives omit as much as what they contain; these are historical skills, and they need to be built into the systems that process historical text, not bolted on afterward. That requires active, collaborative engagement: training students who can move between historical reasoning and computational practice, building benchmarks grounded in the evidentiary standards of the discipline, and contributing directly to the design of retrieval and reasoning systems. This series is one venue for that work.
The Problem of Speed
The problem of speed is especially acute for work involving large language models. A paper submitted to a traditional journal today may describe a model that is already twelve months obsolete by the time it clears two rounds of peer review. Reviewers may request engagement with tools or benchmarks that did not exist when the paper was drafted. The result is a publication system structurally mismatched to the pace of AI development, a mismatch that discourages precisely the kind of iterative, applied experimentation this field needs. Working papers, by contrast, can circulate while the methods they describe are still current, and can be revised openly as the technology evolves.
The practitioners of digital humanities and social science have always improvised: blogs, Substacks, lab notebooks, and informal publications have long been central sites of intellectual exchange. Today, Ted Underwood’s blog is widely cited and taught; Jessica Marie Johnson’s lab circulates readings and reflections through an active Substack; Mark Humphries shared his success with handwriting transcription on his blog and then provided key updates to his journal article when Gemini 3 launched (Humphries et al. 2025). These auxiliary forms are not peripheral; they are often the most faithful record of what digital humanities actually is: heterogeneous, collaborative, and in motion.
But they also reveal a structural problem. The pace of discovery now exceeds the speed of formal publication. Methods evolve between submission and print; datasets change; models improve; interpretations are revised in real time. Blogs and Substacks fill the gap, but they lack the stability, citability, and collective visibility of formal scholarship. What is needed is not a replacement for these forms, but an infrastructure that matches their speed while preserving scholarly standards. A working paper series does precisely that: it captures work in motion without sacrificing rigor, making it possible to publish quickly, revise openly, and cite reliably in a field where waiting two years is no longer viable.
There is also a personal motivation. For any scholar, there are unpublished ideas that are not yet ready for formal publication but that might nevertheless benefit the author and reader from circulation. The overflow principle is all the more true when it comes to the intersection of fields. Clifford edits Historical Methods because he believes it is an important journal, and cares deeply about the computational methods we are developing to study the past. But he wants to spend most of his limited deep writing time on the historical questions that drive his research: environmental history, global history, and the history of capitalism. Guldi directs the Center for the Future of Trust at Emory, an interdisciplinary lab of data scientists, computer scientists, digital humanists, and historians. But the lab’s published output only documents a handful of the conversations about pedagogy, design, interdisciplinarity, and method under development at any given time. Working Papers in Critical Search is designed to capture conversations that transcend any individual journal—offering a space where ideas about political economy, design practice, and interdisciplinary method can be shared in formation, tested across fields, and refined in dialogue before they harden into more conventional forms of publication.
The conventions of academic publishing are themselves under pressure from generative AI, and the early signs of disruption are already visible. Journal editors are beginning to see how AI-assisted writing destabilizes the old calculus of what counts as a publication, how long it takes to produce one, and how reviewers evaluate originality. These shifts will only accelerate. We need to build new practices that provide a bulwark against the tyranny of publish-or-perish and the smallest publishable unit culture that dominates much of the academy in the twenty-first century. We need room where big ideas can be explored in company, even when they require later refinement. In a moment when AI is compressing the time between idea, execution, and dissemination, working papers offer a way to match that speed with rigor—creating a space where innovation can be shared, tested, and improved in real time rather than delayed into obsolescence.
A New Working Papers Series
Working Papers in Critical Search is designed to formalize and accelerate this emerging field. It provides a venue for work that is simultaneously conceptual and experimental: early-stage arguments, methodological reflections, datasets, code, and interpretive essays that operate at the intersection of history, computation, and political economy. Its primary purpose is to create a venue for work that enacts and extends the principles of critical search. Its distinctive contribution is to treat method and interpretation as inseparable—to publish not only findings, but the reasoning, design choices, and iterative processes that produce them.
In doing so, the series aims to bring historians back into long-run debates about society and change—equipped not only with richer archives, but with methods that allow them to engage scale without surrendering interpretation.
Edited by Jim Clifford and Jo Guldi, this series is equally committed to pedagogy: training students and researchers to produce computational work whose “proofs” are not only technically valid but historically interpretable.