Research
A growing set of papers and a monograph are currently under way. They combine global analyses with within-country case studies of how authoritarian rule shapes what scientists publish. Working titles and abstracts are provisional; drafts are shared on request.
Tore Wig, Martin Søyland (University of Stavanger), Ole Magnus Galaas Hansen (University of Oslo), and Jens Jungblut (University of Oslo)
Are the contents and conclusions of the social sciences different in autocracies and democracies? This paper studies how autocratic regimes shape the content of the social sciences, using three million social science abstracts from across political regimes.
Tore Wig
How do autocratic regimes shape what scientists write? The assumption that scientists self-censor yields a wide range of different implications. This paper explores these implications using an "agent orchestra": a structured pipeline in which many independent AI analyst teams investigate the same overarching question, each with a pre-specified hypothesis, and their results are synthesised and reviewed under common standards.
Tore Wig
Does authoritarian rule shape how historians write about their own nation? History is central to how regimes legitimate themselves, which makes historiography a revealing test case for the politics of knowledge. This paper asks whether autocratic conditions shape the topics historians choose and the way national history is framed, distinguishing celebratory legitimation from the quiet avoidance of politically difficult periods and topics. It provides the first large-scale cross-national study of the content of academic historiography under authoritarian rule.
Code on GitHub ↗ (private during development)
Tore Wig and Felix Haass (Humboldt University of Berlin)
What does authoritarian rule do to social science? Divided Germany offers a rare natural comparison: East and West shared a language and a pre-1945 scientific tradition, but lived under sharply different political regimes for more than forty years before reunifying in 1990. This paper compares social-science and humanities research from East and West German institutions, before and after reunification, to isolate the imprint of the authoritarian period and its aftermath on what researchers studied and how visible their work became internationally.
Code on GitHub ↗ (private during development)
Tore Wig and Oguzhan Turkoglu (Trinity College Dublin)
This paper studies the impact of growing autocratic consolidation in Turkey on academic output at the level of researchers, PhD candidates, and graduate students. We draw on a new dataset of 500,000 graduate theses, 500,000 native-language publications, and 200,000 Web of Science articles by Turkey-based authors. We look at how topic choice, political framing, and sentiment change with broader regime trends and events, such as the 2016 coup attempt and the 2020 university reform, and how they vary within and across universities.
Tore Wig and Nils B. Weidmann (University of Konstanz)
Does the credibility of scientists, and of science itself, depend on politics? This paper asks whether people judge a scientific finding differently when the scientist is politically engaged, as a critic or a supporter of the government, and when the scientist works under an authoritarian rather than a democratic regime. It also asks whether the two constraints interact, so that political engagement carries a different cost for a scientist's credibility in autocracies than in democracies. The questions are studied in a multi-country survey experiment spanning democracies and autocracies, complemented by a sample of scientists.
Tore Wig and Daniel Goldstein (University of Oslo)
Much current debate concerns how to align AI models with human values. This paper asks a different question: what happens when AI models are aligned with political regimes? We theorise the costs and benefits of aligning models with democracies and with autocracies. For an autocrat, alignment can reduce the flow of regime-threatening information and the potential for model-assisted challenges to the regime, but it may at the same time constrain the productivity gains from AI adoption and model-driven technological growth. The paper develops this trade-off and draws out its implications for how regimes of different types will shape, and be shaped by, the AI systems they adopt.
Frederik Hjorth (University of Copenhagen), Tore Wig, Eli Baltzersen (University of Oslo), and Martin Søyland (University of Stavanger)
Where does critical social science come from? The interface between social science theories and politics is rarely studied systematically, yet political actors and movements have repeatedly shaped the social sciences. This paper studies the emergence of critical social science as a political phenomenon and investigates its political origins. It charts the rise of critical social science over time and space and asks whether the radical student protest movements of the late 1960s and 1970s left a lasting imprint on the social science later produced at the universities where they took place. The contribution highlights the importance of political history for understanding the trajectory of the social sciences, and shows how this can be studied rigorously at scale.
Working paper; draft available on request.
Tore Wig and Sirianne Dahlum (University of Oslo)
Can we predict which institutions and departments are hit by academic-freedom attacks by looking at their academic output? This paper draws on and presents a new dataset on attacks on academic institutions around the world, 2000–2026, and links these data to university research production. We investigate whether specific content dimensions, such as regime-critical social science, predict attacks on academic institutions.
Mitchell Bosley (Pontificia Universidad Católica de Chile) and Tore Wig
The most prominent political science theories are often general in scope, yielding long lists of empirical implications across a range of outcomes and empirical domains. These "grand theories" are found in studies of diverse political phenomena, such as war, great power competition, democratisation, state formation, legislation, and elections. Since they never yield one definitive empirical test, it is hard to evaluate their evidentiary support. We remedy this by proposing a new approach to evaluating the evidence strength of grand theories, in an agentic-AI architecture custom-made for this purpose. The architecture deconstructs theories into their canonical empirical implications and visualises them in theory graphs. It then assigns agents to scan a pre-set corpus of research papers for relevant evidence claims traced to concrete citations. These claims are evaluated by a "theory guardian" layer that decides which pieces of evidence are admitted as relevant to the theory's implications, before an evidence-aggregation step updates the Bayesian posterior of each theory implication and, in turn, of the overall theory. We present this in an interactive interface where researchers can pre-set their meta-theoretical and philosophical commitments as constraints on the workflow and visually inspect how the evidentiary status of a theory evolves in confrontation with different streams of research.
Tore Wig
The question of how autocracy and democracy affect scientific work invites a long list of more or less plausible theories. How do we decide which ones to test? This paper develops an "automated theory lab" approach to identify the most plausible candidate theories, in which an AI pipeline is constructed to sharpen theories and their predictions. Rival theoretical lenses — from political economy and legitimation theory to field autonomy, repression, and selection — are developed into competing theories, each with concrete, testable implications, and stress-tested against one another in a structured tournament with independent refereeing and adversarial red-teaming. The main payoff is a register of discriminating tests: the places where rival theories predict opposite things, which empirical papers can adjudicate.
Tore Wig
A book-length synthesis of the project's findings, drawing the empirical papers together into a single theoretical and empirical account of how authoritarian rule shapes the production of scientific knowledge. Expected in the later phase of the project.
Working titles and author lists are provisional. Papers and code are shared as they reach the working-paper stage; drafts are available on request.
Research integrity
Artificial intelligence is both an object of study in AutoKnow and a tool in its research process. Language models serve as measurement instruments in several papers, and AI assistants support coding, literature searches, and project administration. All of this happens under a written project policy, adopted in 2026 and binding for every paper. Its principles:
AI tools are never authors or co-authors. The named researchers are accountable for every claim, interpretation, and conclusion, and they review and take ownership of anything an AI tool has contributed.
Each paper keeps a running log of substantive AI use: what was done, with which tool and version, and when. Where AI has generated text, code, or figures, or has been used to analyse data, this is declared in the published paper, following the rules of the journal that publishes it.
Where language models are the measurement instrument, the pipeline, the model versions, and the validation against human coding are described so that results can be reproduced and scrutinised.
Every citation proposed by an AI tool is checked against the real source before it enters a manuscript. References that cannot be verified are flagged and removed.
Analysis code written with AI assistance is reviewed by a member of the team before any result it produces is reported.
AI does not lower the bar. Analyses follow pre-specified plans, all estimated models are reported, null results are reported as null results, and exploratory work is labelled as exploratory.