Research

The Sciences Po team:

• Jan Rovny, CEE
• Caterina Froio, CEE
• Jean-Philippe Cointet, médialab
• Romain Lachat, CEVIPOF
• Elena Cossu, CEE

Partners:

• CEU Democracy Institute (Hungary), leader
• University of Oxford (UK)
• Charles University (Czechia)
• Scuola Normale Superiore (Italy)
• SWPS University (Poland)
• The Transatlantic Foundation (Belgium)
• University of Vienna (Austria)

Period: 2022-2026

The study aimed to capture the dynamics of neo-authoritarian and illiberal ideologies in the European Union as a whole, focusing in particular on seven countries: Austria, the Czech Republic, France, Hungary, Italy, Poland and the UK.

The Sciences Po team was responsible for the mapping of democratic and neo-authoritarian, illiberal ideologies in present-day Europe. This was based on the most extensive data-collection work in the project: party documents, speeches of public figures and social media activity of elites and engaged citizens. The team combined expert surveys with quantitative text analyses.

Other work packages included:

• identifying the historico-cultural context of these ideologies,
• identifying the factors leading to the support or rejection of these ideologies,
• better understanding the transnational exchanges between illiberal movements and the policies implemented when they come to power.

The overall aim of the project was to provide policy makers with a toolbox to counter the various neo-authoritarian ideologies and improve support for liberal democracy. In this vein, the project produced an interactive dashboard tool mapping illiberal attitudes across 7 European countries, which combines multiple data sources and measurement approaches in an accessible visual format.  

Recent publications:

• Cossu, E., & Froio, C. (2025). How Terror Attacks Shape Political Agendas on Multiculturalism in France. Politics and Governance, 13, Article 9743. https://doi.org/10.17645/pag.9743

Learn more about the project.

The Sciences Po team:

• Jean-Philippe Cointet, médialab
• Bernard Reber, CEVIPOF
• Manon Berriche, médialab
• Salim Hafid, médialab
• Théophile Pénigaud de Mourgues, CEVIPOF

Partners:

• Make.org
• Sorbonne Université (ISIR & CERES)
• CNRS

Period: 2024-2026

Make.org, Sciences Po and Sorbonne-CNRS have joined forces to set up the “Democratic Communs” research program, which aims to use the potential of generative AI to preserve and strengthen democracy.

Its primary objective is to develop and share a social science scientific framework for determining democratic principles applied to AI, a model for evaluating the biases of large language models (LLMs) against these principles, debiased LLMs, and citizen participation platforms that adhere to these principles. 

The project is built around a unique partnership between two research laboratories specializing in natural language processing and robotics (ISIR and CERES at Sorbonne University), two political science and sociology laboratories (CEVIPOF and médialab), and a civic tech start-up that designs citizen consultation tools (make.org).

The challenge is to define, within this multidisciplinary consortium, the conditions for the democratic integration of generative AI into such spaces. The work is initially being carried out at a theoretical and normative level to describe the democratic principles that are desirable in our societies.

We are also attentive to how citizens express certain expectations when they interact with non-human agents. How can these principles be characterized and measured? Finally, in collaboration with our IT partners, the project aims to develop solutions for training models that are likely to meet these expectations.

Learn more about this project.

The Sciences Po team:

• Beatriz Botero Arcila, Law School
• Pedro Ramaciotti Morales, médialab
• Rachel Griffin, Law School
• Claire Stravato Emes, Law School

Period: 2024-2026

The 2022 Digital Services Act (DSA) establishes a new comprehensive regulatory framework for online platforms in the EU. It reserves the most extensive obligations for ‘very large online platforms’ (VLOPs), defined as those with over 45 million users. These platforms are considered particularly important for the broader media system and for political and democratic discourse. VLOPs are obliged to assess and evaluate ‘systemic risks’ in specified areas (e.g. fundamental rights, polarisation, public health and security); reasonably and proportionately mitigate these risks; and report to the Commission on their mitigation measures.

Importantly, the DSA also introduces mechanisms for vetted researchers to access platforms’ internal data “for the sole purpose of conducting research that contributes to the detection, identification and understanding of systemic risks in the Union”. Policymakers envisage that independent scrutiny from academic researchers and civil society organisations will play an essential role in identifying and defining systemic risks associated with social media, as well as holding VLOPs accountable for effectively mitigating them.

This project will bring together a team of researchers from Sciences Po’s Law School and médialab to be a central part of that effort. The project aims to monitor and critically evaluate how the systemic risk framework is being implemented, and to position itself as a key resource for regulators implementing and enforcing the DSA. It will provide actionable guidance and resources for regulators, civil society and researchers on how to utilise this framework to strengthen the governance of online media platforms.

We will focus on two main lines of research: How are systemic risks and appropriate risk mitigation measures understood and defined by VLOPs and other stakeholders within the DSA framework? What data should researchers have access to in order to evaluate and monitor systemic risks within the DSA framework?

This project is financed by the Project Liberty Institute.

Learn more about the project.

The Sciences Po team:

• Emiliano Grossman, CDSP
• Jan Rovny, CEE
• Damian Peška, undergraduate intern, Sciences Po, CEE (January-May 2025)

Period: 2026-2029

What makes an issue politically contentious? How do political problems arise? How do they divide opinions? How do they go away? The “Dynamics of Issue Evolution” project uses modern computational social science techniques to answer these questions, by capturing the long-term evolution of key political issues. 

The aim of this project is to map the evolution of key political issues, such as immigration, the environment, budgets and state administration, over time, in several European democracies. To this end, we will compile diverse corpora of political speeches, parliamentary debates, as well as media reporting, covering the period from the 19th to the 21st century.

This will help addressing questions such as:

• How and when do specific issues emerge in public discourse?
• How do political issues evolve in terms of framing, tone, and positional distribution?
• Who are the actors introducing these issues, and how do their framings, tones, and positions change over time?
• How do competitive environments or exogenous factors influence issue evolution?
• How have these dynamics shifted over time, considering institutional and technological developments?

Ultimately, the project will further our understanding of how political debates translate into policy decisions, focusing on the factors that push specific issues into the legislative arena.

The project is funded by OpenAI, as part of the international NextGenAI consortium, which brings together leading global universities and OpenAI.

Learn more about this project.

The Sciences Po team:

• Kevin Arceneaux, CEVIPOF
• Damien Bol, CEVIPOF
• Pierre-Henri Bono, CEVIPOF
• Raphaële Xenidis, Law School

Period: 2026-2029

Functioning democracies require open and inclusive fora for citizens to share, debate, and form opinions. There is widespread doubt about whether social media can play this essential role in its current form. A wealth of studies has demonstrated that polarization tends to create tensions in online discussions: toxicity is on the rise, including pervasive discrimination, misinformation is spreading, and opinions tend to radicalize and oppose more brutally while most moderate voices are sidelined and marginalized groups silenced and excluded. LLMs can play a role in alleviating this vicious circle, by allowing more diverse voices to express and by facilitating discussions. However their contributions must be empirically tested. 

In theory, conversational bots can promote increased participation from more diverse individuals engaged in facilitated discussions. We expect intelligent agents to encourage participation by summarizing lengthy conversations, finding common ground and diverging lines among the mass of expressed opinions. Additionally, they may lessen toxicity by rephrasing posts in a more civil way in order to increase the quality of conversation even when ideologically-opposed individuals are debating. However, such applications call for solid and legitimate public scrutiny and regulation regarding the potential biases introduced by using LLMs. How faithful are language models to the opinions expressed by social media users? How can we measure this fidelity and guarantee that LLMs are not reducing the pluralism of the views when summarizing debates? How do we ensure that LLMs do not amplify already pervasive discriminatory prejudices and stereotypes?

Complementing this investigation into the democratic potential of LLMs in online debate, we will also take a step back to explore the legal and epistemological implications of algorithmic governance more broadly to answer the question:  How are intelligent systems interacting with democratic norms, legal regimes, and social structures?

The project is funded by OpenAI, as part of the international NextGenAI consortium, which brings together leading global universities and OpenAI.

The Sciences Po team:

• Jean-Philippe Cointet, médialab 
• Carola Kloeck, CERI
• Bernard Reber, CEVIPOF

Partners:

• Sorbonne Université
• Nukk.AI

Period: 2024-2029

The aim of this project is to explore how post-generative AI systems can improve deliberation and collective decision-making in order to promote fair and collectively acceptable policies. The aim is to design AI mechanisms and tools that can be used in deliberative democratic processes, and to experimentally evaluate and assess their real value at different scales and in different contexts (e.g. civic assemblies, online deliberative platforms, international or local negotiations, etc.). To achieve this goal, we envisage a combination of:

• Descriptive approaches, based either on large-scale quantitative approaches or on localised studies. The aim is to analyse and visualise how people actually deliberate and issue early warnings about potential risks of manipulation or coercion (mapping controversies, coalition dynamics, diagnosing influence between individuals, etc.);
• Descriptive-normative approaches based on multi-agent simulations, enabling complex systems based on empirically validated behaviours to be run in silico and counterfactual scenarios to be tested;
• Normative approaches based on idealised models, which explore the properties of deliberation and collective decision-making mechanisms. More specifically, computational social choice, formal argumentation, multi-criteria decision support and preference learning offer a range of new techniques for designing mechanisms whose desirable properties are theoretically guaranteed. We intend to experiment with these AI-augmented deliberative frameworks in a variety of real-world situations.

This project is part of the PostGenAI@Paris project and is supported by State Aid managed by the French National Research Agency, under France 2023, with the reference ANR-23-IACL-0007.

Learn more about this project.

The Sciences Po team: 

• Alina Danciu
• Jieun Jeong
• Lucie Marie

Partners: 

• HUN-REN Centre for Social Sciences (coordinator)
• Research Documentation Centre
• Tampere University, Finish Social Science Data Archive 

Period: 2024-2026

The ONTOLISST project aims to redefine the way archives construct realities by assigning concepts to social science research data. Currently, there is a lack of multilingual tools to support concept attribution for this type of data, which hinders their exploration and accessibility.

The project team’s analysis of existing ontologies will provide a basis for interpreting thematic metadata systems, which are often taken for granted—even though they are shaped by specific languages and, more significantly, by historically formed practices rooted in a single discipline.

To address the above challenges, the team will develop the Light Social Science Thesaurus (LiSST), a controlled vocabulary of approximately 100 terms organised into two hierarchical levels. This thesaurus will be used to create a reference corpus for the (semi-)automated assignment of concepts within research data repositories, using annotated samples of social science (meta)data in English, Finnish, French, and Hungarian.

In addition, the project will explore how Natural Language Processing (NLP) tools can support efficient concept annotation across all languages.

Within this project, the CDSP is leading Work Package 1, which focuses on gathering metadata and thesauri from various European and international archives.

This project is funded by the European Union, as part of the first European OSCARS call.

Recent publications:

• Babolcsay, B., Hertrich, C., & Danciu, A. (2026, janvier 12). The ONTOLISST project. Using NLP for automated topic assignment for SSH research data. Zenodo, DOI

Learn more about the project.

The Sciences Po team: 

• Alina Danciu
• Jieun Jeong
• Lucie Marie
• William Feraux
• Malaury Lemaître-Salmon

Partners: 

• CLOSER UK
• The Finish Data Archive (FSD) 
• L'Institut national de la santé et de la recherche médicale (INSERM, Cohorte Constances) 
• Research Documentation Centre, HUN-REN Centre for Social Sciences

Period: 2024-2027

In recent years, many efforts have been made to implement the FAIR principles. While the first three, Findable, Accessible, and Interoperable, have been widely explored, the fourth, Reusable, remains under-investigated.

Existing FAIR assessment tools tend to focus primarily on metadata, rather than the data itself. This project aims to address that gap.

The first objective is to empirically validate formal criteria related to the "Reusable" aspect of FAIR principles.

The second goal is to develop online tutorials on the Data Documentation Initiative (DDI) — a standard that has proven its effectiveness in enhancing reusability.

The third objective is to design a metadata curation protocol using Artificial Intelligence (AI). The harmonized metadata will be made available to the research community through a question bank built by the CDSP team.

Each objective corresponds to one of the project's three work packages.

The project is funded by the French National Research Agency (ANR) as part of the Réso 2024 call for proposals.

The Sciences Po team: Frederico Varese

Period: 2021-2026

We all know that organised crime is harmful; however, some important aspects remain unknown. How do groups engaged in criminal activities (production, trade or governance) differ from each other? Do they each have a different organisational structure, members with different ‘professional’ profiles and skills? To what extent is there overlap between them? Under which conditions would one group specialising in production or trade evolve into a governance type organised crime group? The EU-funded CRIMGOV project will answer these questions. The focus will be on local cybercrime production hubs in Europe, the international trade of drugs from Colombia to Europe, and the emergence of criminal governance inside and outside prisons.

Recent publications:

• Where do cyber threats come from?
• Unpacking Organised Crime

Learn more about the project.

The Sciences Po team:

• Laura Morales
• Dimitrios Rafail Tservenis
• Yuma Ando
• Ekaterina Iakovleva 

Period: 2024-2026

The project will build on existing collaborations (for example the ETHMIGSURVEYDATA network) and cross-fertilising initiatives of Open Science with the aim of generating European-wide infrastructures that make research and data focusing on Ethnic and Migrant Minorities and on Migration findable, accessible, interoperable and reusable (FAIR). 

OPENMIN will consolidate and expand some existing FAIR tools: 

• the Ethnic and Migrant Minorities (EMM) Survey Registry, the EMM Question Data Bank, and the EMM Post-Harmonized Survey Data Bank, set up as part of the ETHMIGSURVEYDATA project;
• the nccr – on the move Migration-Mobility Survey data analysis and reuse tools; 
• the IMISCOE Migration Research Hub. 

It will also allow for cross-national learning and tool development through the conception and/or generation of new Open Science and Open Data infrastructure tools and resources: 

• a prototype for a new Ethnic and Migrant Minorities Qualitative Study Registry, 
• a self-depositing Ethnic and Migrant Minorities Open Data Repository, 
• a metadata collection on surveys conducted with Ukrainian migrants and refugees,
• the prototype for an Ethnic and Migrant Minorities Survey Data Playground. 

In so doing, OPENMIN will contribute to the Open Science and Open Data strategic agenda for the field of studies on Ethnic and Migrant Minorities, and on Migration more generally. In addition, it is designed to find synergies and opportunities for contribution to the EOSC - EU Node, by mobilizing already existing collaborations with EOSC projects and with the RDA.

Learn more about the project.

The Sciences Po team:

• Patrick Le Galès
• Eleonora Russo
• Luca Venga

Period: 2024-2028

The LAC-EU doctoral network research objective is to better understand LAC and its relationship with the EU, fostering respectful, non-Eurocentric research and engagement. It will enable a better understanding of current challenges in Latin American governance and support sustainable, equitable and successful relations between the EU and LAC.

LAC-EU is providing world-class, interdisciplinary research and training in social and political sciences, law, and the humanities, as well as research exchanges and policy-relevant secondments in the EU and LAC. Research themes include the rise of populism, democracy, gender rights, trade, migration, health, global cities, sustainability and green finance, and cultural studies.

The LAC-EU doctoral programme will thus provide PhD candidates a broad range of research-focused and transferable skills, and provide the EU with the vital inputs needed to design coherent, sustainable, equitable and comprehensive policies towards LAC.

Sciences Po will lead one of the 12 research projects, centred around the use of big data, algorithms, and AI by private and public organisations in the process of governments in LAC and EU national governments, states and major cities. 

Learn more about the project.

The Sciences Po team:

• Caterina Froio
• Timothée Dennebouy, Research Intern (January 2026)

Partners:

• Centre d'Etude de la Vie Politique (CEVIPOL), ULB
• University of Zurich

Period: 2026-2028

Although far-right collective actors are engaging more frequently in protests, and their street actions are becoming more visible and widely attended, we still do not know whether far-right protest mobilisation matters to society, or to what extent. Does street activism help to entrench far-right campaigns and actors in society? Do far-right protests contribute to ideological divides among citizens?

IMPACT employs a multi-method, comparative approach combining protest event analysis, text-as-data methods and survey experiments. This provides comprehensive insights into the political impact of far-right protests on citizens' views in 12 countries between 2008 and 2024. The analysis examines both EU and non-EU members in Eastern and Western Europe, which are characterised by diverse institutional frameworks.

IMPACT is organised into two work packages:

• WP1: Mapping far-right protests in Europe through large language models (LLMs). WP1 will produce a new dataset of protest events, enabling qualitative and quantitative mapping of far-right mobilisation.
• WP2: Far-right protests and citizens' attitudes. WP2 will combine large-scale, multinational survey data from the HUMAN Surveys Project with protest event-level data from the database in WP1, enabling comparison between respondents interviewed just before and after such events.

A survey experiment will investigate the causal impact of the issue focus, tactics, and media coverage of protest events on perceptions of far-right protests.

Learn more about this project.

The Sciences Po team:

• Cyril Benoît, CNRS Researcher, PI of the team
• Sebastian Thieme, Postdoctoral Researcher

Period: 2024-2029

How exactly does political corruption work in today's digital societies? How pervasive is its negative impact on democracy? And how can anti-corruption efforts rebuild people's support for democracy? RESPOND will address these key questions by bringing together researchers who specialise in the detection, measurement and evaluation of corruption, as well as specialists in social movements and political participation who are interested in the effects of corruption on citizens' trust in the political system. 

Based on a mixed-methods research design, RESPOND examines 27 EU countries and 11 neighbouring countries in order to understand the mechanisms of corruption in digital societies, its impact on democracy, and the responses needed to promote people’s commitment to integrity. It does so by:

1. analysing four contemporary and relevant forms of political influence: political finance, lobbying, revolving doors/personal ties, and media capture; 
2. evaluating how political corruption is understood by political elites and citizens and is socially constructed through media and education; 
3. exploring how digital technologies entangle with political corruption and how they improve anti-corruption and pro-integrity strategies; 
4. engaging with relevant stakeholders to co-create practices and tools (including new risk indicators), to increase civic monitoring and integrity in democracies. 

The Sciences Po team will work closely with the CEU team and they will be responsible for the most extensive empirical data-collection work package in the project. They will assemble large-scale micro-level datasets on legislative and regulatory processes across various countries, and design a series of corruption risk indicators in order to measure corruption and favouritism in policymaking in Europe, and its effects at the level of companies or economic sectors. 

Learn more about the project.

The Sciences Po team: Romain Lachat

Period: 2024-2026

The goal of our project was to investigate how users of social media assess misinformation and how it influences their social media behaviour, their engagement with different types of posts, and how it may have downstream effects on political attitudes. In this regard, we were particularly interested in evaluating different ways to help social media users assess the credibility or reliability of social media accounts.

We approached the effects of ID verification or account credibility schemes with the help of experiments. In the studies we conducted, we relied on a simulated social media platform, which allowed us to manipulate the content to which participants were exposed and to monitor their behaviour on the mock social media (time spent on different posts, positive and negative reactions to messages such as “liking” or “flagging” certain posts). This simulated social media platform was integrated in a larger online survey, with pre-treatment and post-treatment attitudinal questions.

This project is financed by the Project Liberty Institute.

The Sciences Po team: Jen Schradie

Period: 2023-2027

What shapes how people respond to disinformation? Fake news, online deception, and computational propaganda are all terms to describe a phenomenon that has generated a broad array of studies, particularly over the past five years, yet one puzzle remains. Dramatically more people believe disinformation than those who report consuming it (Allen et al. 2020; Fletcher and Nielsen 2018; Grinberg et al. 2019; Tsfati et al. 2020). Understanding the mechanisms of this disconnect is the primary objective of my ACTIVEINFO research project. Most of this extant research has focused on top-down factors of disinformation diffusion in the digital era (3D), including Big Tech platform companies, the role of government regulation, and powerful political and financial actors. All of these approaches suggest that the audience for disinformation is passive, rather than active participants in the generation and recirculation of news and information, going against decades of communication research (Katz and Lazarsfeld 1955). The project will advance the 3D state of the art by avoiding the bias of selecting on the dependent variable of digitally visible disinformation and, instead, examine a broader array of bottom-up and everyday media practices. Furthermore, the project will uncover not just what people are sharing, but what they are not sharing and why. In ACTIVEINFO the project will address this selection bias in the literature with a two-country comparison (France and the United States) via qualitative-rich fieldwork and mixed methods, often absent from online data collection procedures.

Learn more about the project.

The Sciences Po team: 

• Sylvain Brunier
• Jean-Noël Jouzel

Partners:

• INRAE
• INRIA
• Ecole Supérieure d'Agriculture d'Angers
• Institut Polytechnique UniLaSalle

Period: 2023-2028

The Coevolution of Equipment, Digital Technologies and Agroecological models project, funded by the ANR's priority research program and equipment, is managed by Pierre Labarthe, Director of Research at INRAE. It brings together 16 research units and around 50 permanent researchers and academics in the social sciences, including Sylvain Brunier and Jean-Noël Jouzel for the CSO, alongside AGIR, DATASPHERE, INNOVATIONS, ITAP, LEREPS, TERRITOIRES and TETIS.

The CoEDiTAg project (Coevolution of Equipment, Digital Technologies and Agroecological models) aims to understand the co-evolution between 1) the trajectories of development of equipment and digital technologies, and 2) the transformations of the structures, arrangements and organisations that frame agroecological transitions. This project posits that the direction of this coevolution is not predetermined and depends, to a large extent, on the strategies of actors, their socio-economic interactions and public policies. Indeed, Equipment and Digital Technologies (EDiTs) can reinforce the industrialisation of agriculture or be a lever for a stronger ecologisation of agriculture. CoEDiTAg gathers research in different social sciences to understand the mechanisms of this co-evolution.

Learn more about the project.

The Sciences Po team: Eduardo Perez

Period: 2021-2026

Informational environments are largely endogenous. They can be, and often are, chosen or designed by individuals or organizations with specific objectives in mind. As recognized by a large literature in economics, information plays a crucial role in shaping the outcome of downstream decisions by strategic players (i.e. at the receiving end of informative signals). However, the structure of information also impacts decisions by strategic agents upstream of the generation of signals, as agents mould the underlying reality differently depending on how other players will eventually be informed about it. Finally, designed information production systems are susceptible to manipulations by third party agents pursuing their own interests.

I will seek to further our understanding of socially or privately optimal information designs, how they shape upstream and downstream decisions, how they can be manipulated by private interests, and how to best anticipate and counter such manipulations. I will rely on the analysis of a largely unexplored designer-agent-receiver class of games, in which the designer picks an information generation system, the agent takes an upstream decision affecting the states of the world, or manipulates the production of information, and receivers choose downstream actions based on realized signals.

The project is organized around the different technologies available to the agent. I will consider fake news production, which is the fabrication of signals that pass as informative but are in fact independent of the truth; state falsification, which consists in falsifying the state of the world, or feeding the information production process with falsified data; pure agency, which is the possibility for the agent to secretly deviate to a different but undistinguishable information generation technology; and state shifting, which is the upstream effort an agent can exert to actually transform the probability distribution of states of the world.

This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 101001694).

Learn more about the project.

The Sciences Po team: Julia Cagé

Period: 2021-2026

The goal of PARTICIPATE is to develop a comprehensive approach to the study of campaign finance, information and influence, using new individual-level data and investigating recent changes in participation behaviors. What are the motives behind small campaign contributions? Does tax policy affect political giving? Why are so few politicians from the working class and can this change? What is the ability of the media to induce citizens to make electoral decisions they would not make if reporting were unbiased? While there is evidence at the macro level on the flow of money in elections, news consumption, or the extent of charitable giving, relatively little is known at the micro level, e.g. on individual-level behaviours such as the motivations of small donors, the tax-price elasticity of political donations, or the exposure to competing information flows. PARTICIPATE will help fill this gap. By providing groundbreaking evidence on new forms of participation, it will lead to the reassessment of influential theories of special interest groups and policy formation. One distinguishing feature of the proposal is to combine comprehensive individual-level datasets and the use of computer sciences tools (such as natural language processing techniques and machine learning).

PARTICIPATE will advance the existing research in three steps. First, I will propose a unified analysis of political contributions. This will include a groundbreaking assessment of the importance of small campaign donations, and a combined study of charitable giving and political contributions investigating the impact of tax policy on donations. Fundraising success can lead to the emergence of new candidates. I will then consider citizens’ decision to run for elections, and investigate the role played by network in political selection. Finally, given the importance of media organizations in shaping participation, I will study the changing patterns of information propagation and political influence.

This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 948516).

Learn more about the project.

The Sciences Po team: Alfred Galichon

Period: 2020-2026

This project seeks to build an innovative economic toolbox (ranging from modelling, computation, inference, and empirical applications) for the study of equilibrium models with gross substitutes, with applications to models of matching with or without transfers, trade flows on networks, multinomial choice models, as well as hedonic and dynamic pricing models. While under-emphasized in general equilibrium theory, equilibrium models with gross substitutes are very relevant to these problems as each of these problems can be recast as such.

Thus far, almost any tractable empirical model of these problems typically required making the strong assumption of quasilinear utilities, leading to a predominance of models with transferable utility in applied work. The current project seeks to develop a new paradigm to move beyond the transferable utility framework to the imperfectly transferable utility one, where the agent’s utilities are no longer quasi-linear.

The mathematical structure of gross substitutes will replace the structure of convexity underlying in models with transferable utility.

To investigate this class of models, one builds a general framework embedding all the models described above, the “equilibrium flow problem.” The gross substitute property is properly generalized and properties (existence of an equilibrium, uniqueness, lattice structure) are derived. Computational algorithms that rely on gross substitutability are designed and implemented. The econometrics of the problem is addressed (estimation, inference, model selection). Applications to various fields such as labor economics, family economics, international trade, urban economics, industrial organization, etc. are investigated.

The project touches upon other disciplines. It will propose new ideas in applied mathematics, offer new algorithms of interest in computer science and machine learning, and provide new methods in other social sciences (like sociology, demography and geography).

This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 866274).

Learn more about the project.

The Sciences Po team: Justine Knebelmann

Period: 2025-2028

This project aims to investigate the potential of digitalization to strengthen government capacity in low-income countries, both on the taxation and service delivery sides. We will work with two government partners in Senegal and rely on unique at-scale randomized experiments. 

Digitalization can have transformative effects on the work of administrations, but research is needed to document its impacts for governments and citizens. First, we will build on a long-lasting partnership with the national tax administration to investigate the effects of a digitized property tax census on the social contract. In Dakar, the capital region, 25,000 tax bills will be distributed in neighborhoods covered by the tax census, where 60% of properties have an assigned bill, against 15% of properties in comparable control neighborhoods. We will leverage the experimental design and a set of surveys with taxpayers and local actors to document whether this increased tax pressure has effects on citizens interactions with the government. We will also investigate taxpayers' understanding and acceptance of the use of an algorithm to establish the tax base. These results will generate novel insights into the political acceptability of digitally enhanced tax reforms. 

Second, we will evaluate the effects of switching payments of the national anti-poverty program from cash to digital (mobile money) payments. Digital payments and registrations of beneficiaries may reduce transaction costs, and help alleviate informational limitations for better targeting. On the other hand, adoption by vulnerable households may prove challenging and exclusion is a major risk. We will also assess whether the nudge to adopt mobile money via a government program has effects on beneficiaries' financial outcomes. 

Beyond the academic outputs, the project will have immediate policy impacts as a result of the close partnership with the administrations.

This project is financed by the ANR.

Learn more about this project.

The Sciences Po team:

• Raphaële Xenidis
• Steven Vethman

Partners:

• The Nederlandse Organisatie voor Toegepast Natuurwetenschappelijk Onderzoek (TNO) 
• CorTexter Technologies B.V.
• Eticas Foundation
• Turing College
• University College Dublin
• The Women4Cyber Foundation
• Women In AI

Period: 2023-2026

The development of the artificial intelligence (AI) sector internationally has created a range of training and education needs, two of which this project addresses. The first is overcoming the shortage of experts in algorithmic fairness in AI, who have the skills to identify and deal with bias and discrimination in AI. The second is the effective integration of an intersectional approach to understanding and addressing bias, equity, discrimination and equality in AI to prevent the emergence of invisible minority groups who would suffer systematic disadvantage due to AI. DiversiF-AI-R aims to develop a pioneering EU-wide approach to addressing intersectional equity in AI based on use cases of data and models that originate from or have been deployed in Europe, as well as through the use of innovative tools, structuring them in a practical way.

This project has received funding from the European Education and Culture Executive Agency (EACEA) in the framework of Erasmus+, EU solidarity Corps A.2 – Skills and Innovation under grant agreement 101107969.

Learn more about this project.
 

The Sciences Po team: 

• Dina Waked
• Teodora Groza

Partner: Stanford Center for Legal Informatics (CodeX)

Period: 2024-2026

The rapid rise of AI technologies in the past years has sparked widespread discussions amongst academics, policymakers, and the general public. AI-powered chatbots, image generators, and other customer-facing AI tools have become commonplace, prompting questions about their ethical, legal, and societal implications. While much of the conversation has focused on the risks and benefits of these tools, less attention has been given to the governance of AI companies themselves. “Track AI”, a joint collaboration between Sciences Po and the Stanford Center for Legal Informatics (CodeX), addresses specifically the critical but under-explored governance ecosystem of AI firms.

The emergence of AI technologies has long been heralded as a departure from the “Web2” era, dominated by companies like Amazon, Google, and Meta. These so-called Big Tech firms built their empires on proprietary software, user data collection, and network effects that encouraged centralization of market power. In contrast, AI startups were seen as potential disruptors which offered not only breakthrough technologies, but also new approaches to governance centered on putting the interests of society first and developing open-source models.

However, a closer look at the AI landscape reveals that many firms are increasingly intertwined with Web2 giants. From OpenAI to Anthropic and Mistral AI, it is hard to find an AI player that has yet to enter a deal with Big Tech players. Through partnerships with companies such as Microsoft and Amazon, AI firms get access to resources like secure cloud computing, easy access to sizable customer bases, and data. But these “AI partnerships” are not just technical collaborations—they represent a new kind of inter-firm cooperation with potential long-term implications for the entire technology ecosystem.

This project is financed by the Project Liberty Institute

Learn more about the project.

The Sciences Po team:

• Séverine Dusollier
• Rebecca Mignot-Mahdavi
• Beatriz Botero Arcila
• Raphaële Xenidis
• Klaudia Kolnowska
• Nicolas Maleve
• Anamaria Munoz Rincon

Partners: 

• Sorbonne Université

Period: 2024-2029

This project, developed within the PostGenAI@Paris cluster, aims to explore the normative and distributive effects of AI systems. It will study and unpack the new modes of governance that emerge when such systems are integrated into existing law-making social practices like decision-making in government, socio-economic policymaking or even war. The starting point of the project is that in these contexts, law and AI are not separate objects, but rather they interact with each other in complex ways and are related nodes of norm-making networks, producing new modes of normativity, governance and legal subjectivities.

Our objectives are both to understand the consequences of the imbrication of AI systems and policy making for some of our societal premises - such as the guarantee of fundamental rights, democracy or market competition, while also to formulate policy recommendations on how to best regulate AI systems, design and adapt them and use, or not, AI within law and policy-making.

This inquiry will be supported by case studies in partnership with industry partners, such as:

• The use and governance of AI-enabled defense systems, in order to analyze the role of private entities in the development and governance of AI systems and the impact of regulation and other incentives on decision, design and implementation processes.
• The impact of AI and generative in cultural production and consumption, from algorithmic recommendations in music streaming platforms, to uses of images and texts in generative AI tools. In both cases, copyrighted works are not anymore seen in their singularity but as components of databases and cultural flows that radically changes the cultural practices and remuneration of authors and artists.
• The distributive and normative impacts of algorithmic decision-making systems used by private and public actors to allocate social benefits, insurance policies and jobs, in order to understand what forms of discrimination, incl. gender and social discrimination, are co-produced at the intersection between algorithmic bias and these systems’ operation within particular socio-economic policy contexts (for the public sector) and economic extraction models (for the private sector).

This work is conducted as part of the PostGenAI@Paris cluster and is supported by the French National Research Agency (ANR) under the France 2030 program, reference ANR-23-IACL-0007.

Learn more about this project.

The Sciences Po Team:

• Gustav Kalm
• Beatriz Botero Arcila

Partners:

• London School of Economics (LSE)
• Bocconi University 
• European University Institute (EUI)
• South African Institute of International Affairs (SAIIA)
• African Economic Research Consortium (AERC)
• UN Economic Commission for Africa (UNECA)
• Wits University

Period: 2026-2027

Africa’s digital transformation is moving fast. Mobile services, digital trade, data centres, and online public services are becoming an everyday part of life across the continent. This transformation is not just about technology, it is also political, economic, legal, and historical. Decisions about infrastructure, regulation, and global partnerships shape who controls value, power, and resources in Africa’s digital future, 

DigiAfrica: Infrastructure, Critical Minerals, and Transnational Governance in Africa’s Digital Transition (DigiAfrica) is one of the new Research Hubs launched under the first CIVICA call. Running from February 2026 to June 2027, the hub focuses on the interplay between digital infrastructure, critical minerals, and transnational governance in Africa. 

As part of the CIVICA European University Alliance, DigiAfrica combines expertise from the social sciences including law, political economy economics, history, and anthropology, to explore the broader societal and policy implications of digitalisation. The hub approaches Africa’s digital transition as complex, interconnected process, where European social sciences research can help understand and guide real-world policy choices. 

Learn more about this project.

The Sciences Po team: 

• Sylvain Parasie
• Pedro Ramaciotti Morales
• Jimena Royo-Letelier
• Tim Faverjon
• Antoine Vendeville
• Paul Bouchaud

Period: 2022-2026

The AI-Political Machines (AIPM) projects studies how algorithms mediating social media and AI conversational agents (possibly inadvertently) infer represent and leverage attributes of users, and in particular political attitudes. AIPM uses state-of-the-art inference of political stances and to develop AI explainability. The goal of the project is twofold. First, AIPM seeks to render implicit computation and processing of attributes explicit, blurring the line separating active and passive profiling, especially regarding sensitive attributes such as political opinions. Second, AIPM seeks to leverage these advances in the development of new algorithms and AI systems, capable of selectively disregard attributes of users and contents in the computation feeding into downstream tasks, such as recommendation is social media or conversations with AI agents.

Financed by the Project Liberty Foundation.

Recent publications:

• Vendeville, A., Royo-Letelier, J., Cassells, D., Cointet, J. P., Crépel, M., Faverjon, T., ... & Ramaciotti, P. (2025). Mapping the political landscape from data traces: multidimensional opinions of users, politicians and media outlets on X.

Learn more about the project.

The Sciences Po team:

• Donato Ricci
• Axel Meunier
• Dominique Cardon
• Valentine Crosset

Period: 2023-2026

Styles of Moderation is a research project in media studies and design that examines the role of social media users in content moderation, alongside the platforms that control the content they publish and the bodies responsible for governing freedom of expression.

Initially, the project sought to capture this participation by analyzing the results of participatory workshops held in 2022 and 2023 with groups of 3 or 4 people of diverse ages, genders, political views, geographic and social backgrounds, and levels of digital literacy. Participants were invited to “moderate” the same set of 30 real-life cases of disputes surrounding problematic messages posted on the social network X.

In a second phase, the project examines professional practices from the fields of law, computer science, design, communication, and politics that shape user engagement with problematic speech online.

Overall, the project aims to describe the ways in which social media users encounter problematic speech online: both as the development of a situated practice and as an experience resulting from a design process incorporating elements derived from the judgments of professionals across numerous disciplines.

This project is funded by the Project Liberty Institute.

Learn more about this project.

The Sciences Po team: 

• Jean-Philippe Cointet
• Donato Ricci
• Maxime Crépel
• Cosma Cazé
• Tommaso Prinetti
• Imane Bayoub

Partners:

• CNRS
• MNHN
• IFREMER
• Réseau des universités marines : Université Littoral Côte d’Opale, Université de Nantes, Université de Bordeaux, Université de Montpellier et Aix-Marseille Université
• Office français de la biodiversité

Period: 2022-2028

The FUTURE-OBS project is a pioneering initiative to advance the understanding of coastal socio-ecosystems in the face of increasing anthropogenic pressures. The project combines traditional ocean observing methods with cutting-edge techniques such as environmental genomics and in situ biodiversity imaging. By using social media data to monitor human interactions with these vulnerable spaces - the work package led by the médialab - the project pushes the boundaries of what is possible to observe and understand these complex systems.

The project's trans-disciplinary work in high-risk areas such as marine façades, marine protected areas, and aquaculture areas represents an exciting opportunity for collaboration between scientists and operational stakeholders.

Funded by the ANR in the framework of the priority research program “Ocean and Climate”.

Learn more about the project.

The Sciences Po team: 

• Jean-Philippe Cointet
• Alex Kindel
• Célia Nouri
• Carlo Santagiustina

Partner: INRIA

Period: 2024-2028

The SaLM (Socially-Aware Language Models) project is an interdisciplinary project between Inria Paris and Sciences Po that aims to redefine current NLP, LLM-based algorithms by incorporating social contexts into their development and evaluation. It emphasizes the importance of understanding language as a reflection of cultural and social identities.

Recent publications:

• C. Nouri, J.-P. Cointet, and C. Clavel, “Graphically Speaking: Unmasking Abuse in Social Media with Conversation Insights,” in Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), W. Che, J. Nabende, E. Shutova, and M. T. Pilehvar, Eds., Vienna, Austria: Association for Computational Linguistics, July 2025, pp. 18271–18286. doi: 10.18653/v1/2025.acl-long.894. 

Learn more about the project.

The Sciences Po team: 

• Donato Ricci
• Gabriel Alacras
• Tommaso Prinetti
• Zoé de Vries 

Period: 2024-2028

The booming rise of large language models (LLMs) such as ChatGPT has sparked a rush to produce discourse about these technologies. The quick crystallisation of a shared outlook around a few key themes has narrowed the scope of potential interrogations. 

Public and scientific debates focus on technical issues: algorithmic bias, confabulation, and intellectual property violations. However, the problems and consequences associated with their actual use – for both their users and their professional contexts – remain largely unexplored. This asymmetry fuels a mechanical view of technological development and its effects, as if the technical analysis of these systems were enough to predict their social impact. 

Moreover, these discourses present AI as a monolithic and disruptive entity, dismissing the possibility that it may be aligned with existing practices and that its effects may vary depending on situations encountered in one's job. There is thus an urgent need to move beyond predictions about the future of work, to account for the professional contexts in which LLMs are used, and to identify current issues – not prospective ones. 

How do AI's well-known problems (bias, confabulation, etc.) manifest in established practices? What new, unexpected problems are surfacing? How do LLMs shape individual work practices? And in turn, how do professional environments shape LLMs and their use?

With support from Google.org.

Recent publications:

• Alcaras, Gabriel, and Donato Ricci. "Configuration Work: Four Consequences of LLMs-in-use." arXiv preprint arXiv:2512.19189 (2025).

Learn more about the project.

The Sciences Po team: Pedro Ramaciotti Morales

Partners: 

• CNRS
• INRIA
• Sorbonne Université
• Central European University
• Hertie School
• Bocconi University
• London School of Economics
• SNSPA

The European Polarisation Observatory (EPO) was created in 2021 with support from CIVICA - The European University of Social Sciences. During its existence, EPO has evolved from a research network into an incipient infrastructure for the collection, processing and sharing of social media and public opinion data, producing more than 20 research articles, and several Masters’ and doctoral theses, covering all four priority areas of CIVICA. EPO’s data has also supported courses for students and fostered collaborations within CIVICA. The data produced by EPO has been instrumental for important studies in political science, sociology, and computer science, covering all four of CIVICA’s thematic axes. The accuracy and the wide coverage of the data collected and processed in EPO have allowed us to make breakthroughs in the study of online politics, social media platforms and regulation, and the impact of algorithms in digital social ecosystems all across Europe. This new iteration seeks to expand EPO’s infrastructure to become 1) a durable research network connecting researchers in the project but also colleagues within and beyond CIVICA and students, and 2) a consolidated and sustainable infrastructure for data collection in a context of scare access for researchers worldwide, sharing and maintenance for the study of online social ecosystems, addressing all of CIVICA’s axes through a focus on social platforms and digital environments.

Recent publications:

• Vendeville, A., Royo-Letelier, J., Cassells, D., Cointet, J. P., Crépel, M., Faverjon, T., ... & Ramaciotti, P. (2025). Mapping the political landscape from data traces: multidimensional opinions of users, politicians and media outlets on X.

Learn more about the project.