By Francesco Bromo, University of Oxford, and Paolo Gambacciani, University of Bologna
In recent years, artificial intelligence (AI) has begun to reshape nearly every aspect of modern life, including the very fabric of representative democracy. From machine learning tools that classify legislative documents to natural language processing systems capable of summarizing plenary debates, AI is increasingly becoming embedded in parliamentary life. For political scientists, this transformation is not merely a matter of technological innovation or organizational change; it potentially represents an impending shift in how we study politics.
The COVID-19 pandemic accelerated the adoption of digital tools in parliaments worldwide. In this context, AI has found numerous applications. According to the Inter- Parliamentary Union (IPU), in 2024, 12 legislative chambers in eight countries (plus the European Union) reported some use of AI, for a total of 71 AI tools. Yet their applications vary widely. “AI-powered parliaments” use these tools/for: classification systems, bill drafting and amendments, transcription and translation, chatbots and user support, public engagement and open parliament, cybersecurity, and application development.
While many of these tools are currently primarily designed for administrative efficiency, allowing for the automation of routine tasks, some have the potential to reshape the way parliaments function. To mention a few examples: Italy developed a tool that computes similarity scores for legislative amendments to help the speaker determine the order in which these will be debated. Bahrain developed an AI chat-based system that allows legislators to engage interactively with parliamentary proceedings. Brazil produced a tool that summarizes public views on specific legislative proposals.
These examples illustrate the point that, while the motivations for adopting AI may differ, the net effect is a growing corpus of structured, machine-readable data related to politics. For researchers, this is an unprecedented opportunity. One of AI’s biggest potential contributions to the study of political science indeed lies in its ability to process, encode, and organize massive amounts of information relatively quickly and inexpensively. For legislative studies, this can be invaluable. It enables the systematic analysis of speeches, amendments, initiatives, and other parliamentary activities with precision and scale previously unattainable, cutting hours of labor-intensive hand-coding and improving small-scale sampling. Political scientists can thus repurpose these same tools to refine their analyses. Consider these areas:
- Parliamentary debates: AI can cluster, summarize, and classify speeches (or legislative documents) to extract relevant information on speakers, party positions, policy issues, etc. This is already being employed in the parliamentary context, to some extent. For instance, in Chile, an AI tool enables users to retrieve information on the content of debates, arguments made by different speakers, the outcome of a given discussion, and more. In addition to making legislative data more readily accessible, this could also aid researchers in tracking discursive patterns, rhetorical strategies, and issue salience over time, among other benefits.
- Constituency service: AI tools for public engagement enable citizens, among other things, to monitor their MP’s actions regarding a specific issue, thereby bridging the gap between formal legislative activity and political representation. Platforms like DepuChat (currently in development in Italy) could, therefore, supply valuable data, facilitating the study of MPs’ constituency work or thematic specialization.
- Legislative-executive politics: AI tools designed for legislative drafting or assisting clerks enable quick access and summarize relevant information on output and procedures (e.g., government vs. parliamentary proposals). For example, the NORMA project supports the Italian lower chamber in producing reports on legislation. For scholars, such structured data could provide a foundation for new comparative studies of policymaking complexity and executive-legislative dynamics.
Looking ahead, the integration of multimodal AI, capable of analyzing voice tone and facial expressions, may even enable sentiment analysis of legislators during debates. However, such applications raise significant ethical and regulatory concerns.
Despite the optimism, political scientists must remain attentive to the limitations of AI adoption in legislatures. According to the IPU, the development of AI tools depends heavily on a parliament’s prior investment in digital infrastructure, open and machine-readable data, and data ownership. “Wealthier” parliaments (often in Europe or the Americas) are far better positioned than their counterparts in developing countries, where two-thirds of legislatures lack the digital maturity to experiment with AI. This poses the risk of creating an epistemic inequality: researchers will have abundant, fine-grained data on parliaments in more highly developed countries, while less-institutionalized legislatures may remain understudied. Moreover, for now, AI in parliaments is rarely designed with researchers in mind. Its primary aim is administrative efficiency, not scholarly utility. While indirect uses may benefit academics, certain functions are unlikely to be fully developed unless they also serve the needs of parliamentary administrations.
AI’s promise for parliaments (as well as political science research) is tempered by the need for safeguards and guidelines to mitigate the risk of misinformation, bias, and opacity that arise from the way AI tools operate and produce output within the limitations of training data. The IPU and other scholars stress key principles vis-à-vis the use of AI:
- Transparency: AI output must clearly indicate when and how they were generated.
- Traceability: Systems must be able to trace sources and methods, enabling validation and replication.
- Accountability: Decisions must remain imputable to human actors, preserving democratic responsibility (what the EU refers to as “human-centric” approach to AI).Non-discrimination: Training data must avoid reinforcing stereotypes or marginalizing minorities.
- Privacy and copyright: AI output must be based on content whose use is free and fair, where user data is protected and intellectual property respected.
AI has the potential to transform the study of politics. Many traditional methods involving manual coding and/or survey-based estimates of parliamentary behavior and activity may soon be complemented with (and, to some extent, supplanted by) more precise, data-rich AI-supported analysis. Still, the benefits are not automatic. They require robust regulatory frameworks, ethical safeguards, and equitable access to data across legislatures worldwide. For political scientists, the challenge will be to critically integrate these new tools while remaining attentive to their limitations and biases. If successful, AI may not only change how parliaments work but also how we understand democracy itself.