AI – PSA Parliaments

By Franklin De Vrieze.

As well as debating and adopting new legislation aimed at establishing a sustainable legal framework for the governance of AI, parliaments are also exploring and experimenting with the application of AI in their own operations. New AI guidelines by Westminster Foundation for Democracy (WFD) can inform their efforts.

According to the 2024 AI Index Report by Stanford University, the global legislative landscape has seen a significant increase in AI-related laws over recent years. From 2016 to 2023, parliaments in 127 countries passed a total of 123 bills mentioning AI in various contexts. These laws address a variety of issues, including educational reforms, non-discrimination in AI algorithms, and the establishment of AI training programs. This trend highlights the increasing recognition of the need for regulatory frameworks to manage the development and deployment of AI technologies responsibly.

The legislative measures aim to ensure that AI advancements benefit society while mitigating potential risks of AI being manipulated or misused. Therefore, parliaments need to ensure that the adoption of AI is guided by stringent policies, ethical testing, and comprehensive training, as highlighted in WFD’s recent policy brief, “A Democratic Approach to Global Artificial Intelligence (AI) Safety.”

Current applications of AI in parliament

Parliaments have started exploring and experimenting with the application of AI in their own operations. As Dr. Fitsilis from the Hellenic Parliament argues: “the rise of AI is expected to play a significant role in transforming legislatures from paper-based organisations into data-driven institutions”.

The Inter-Parliamentary Union (IPU) has shared insights on “Use cases for AI in parliaments”, highlighting the growing integration of AI in parliamentary functions.

Firstly, AI is increasingly used for transcription and translation, managing records of debates, and subtitling video content. For example, Estonia’s Parliament automates stenography, while Italy’s Senate leverages AI for translating documents. Finland’s Parliament uses AI to summarise documents and create podcast audio, and Brazil’s Chamber of Deputies uses AI to transcribe audio and video files. These tools save time, enhancing productivity and allowing staff to focus on more complex tasks.
Secondly, AI is introduced to assist in drafting legislation and amendments and analysing large volumes of text to identify key themes and insights. This can help in creating more comprehensive and well-informed legislative documents. For instance, the Italian Chamber of Deputies uses AI to receive, store and number amendments according to presentation time (voting order) and uses AI to compare amendments to identify similarities. Brazil’s Chamber applies AI to interpret and group amendments, while Italy’s Senate uses it to ensure compliance with drafting rules, ensuring consistency, accuracy and adherence to legal standards.
Thirdly, AI supports public engagement by analysing public submissions. Brazil’s Chamber of Deputies uses AI to categorise citizen comments on bills, while Italy’s Senate employs AI to facilitate natural language queries about bills, enhancing accessibility and user experience. This means AI systems are helping citizens interact with parliamentary activities, such as answering questions about legislative processes or providing information on parliamentary sessions.
Fourthly, AI is employed in classification systems to manage large data volumes. Italy’s Chamber of Deputies uses AI to categorise plenary session reports, while the European Parliament applies it for automatic text classification with predefined labels.
Finally, AI powers chatbots and user support, improving access to parliamentary processes. Italy’s Chamber of Deputies provides a chatbot for querying parliamentary proceedings, and Estonia’s Parliament uses AI to subtitle live broadcasts for the hearing impaired. AI also automates tasks like schedule management and responses to common queries, while enhancing cybersecurity.

Lessons learned

Based on the lived experience of AI in parliament, there are already a few lessons learned on the governance of AI in parliaments, in line with WFD’s new Guidelines for AI in Parliaments..

Parliaments need to adopt a cautious, step-by-step approach to integrating AI, particularly with generative AI technologies. Initial experimentation should occur in controlled environments to mitigate risks before wider implementation. Such approach is emphasised in WFD’s guidelines, highlighting the importance of pilot projects and controlled rollouts to manage potential risks effectively.

AI systems must be developed and used in accordance with ethical principles to prevent biases and ensure fair representation. WFD’s guidelines stress the importance of developing AI systems that adhere to ethical standards, preventing misinformation and stereotyping.

Establishing clear governance structures and transparency measures is crucial. This involves setting up auditing processes, maintaining transparency about AI usage, and ensuring that AI systems can be held accountable for their outputs. The WFD guidelines recommend comprehensive transparency measures and accountability frameworks to build trust and credibility in AI applications within parliamentary functions.

Parliaments benefit from sharing experiences and best practices regarding AI implementation. Collaborative efforts, such as those facilitated by the IPU’s Centre for Innovation in Parliament, help build a collective understanding of effective AI governance. The new Global Community of Practice on Post-Legislative Scrutiny helps facilitate dialogue on applying AI in legislative scrutiny processes, as argued by Dr Marci Harris from POPVOX Foundation.

As AI technology evolves rapidly, parliaments must remain flexible and continuously update their policies. This includes partnering with academic institutions and other stakeholders to stay ahead of technological advancements and their implications. The WFD guidelines highlight the importance of ongoing education and adaptation to ensure that both parliamentarians and parliamentary staff are equipped to handle the evolving AI landscape.

Conclusion

By building on their existing digital infrastructure, parliaments can harness the benefits of AI while ensuring accountability and protecting democratic values. The WFD Guidelines for AI in Parliament provide a comprehensive framework for this endeavour. Parliaments may take proactive steps to pilot these guidelines, document use cases, and share best practices globally. This will not only strengthen their role as guardians of accountability but also ensure that AI serves the public good, enhances governance, and upholds the principles of democracy.

Europe’s first political tech summit in Berlin The summit, taking place on Saturday, 25 January, will bring together the global political tech ecosystem—spanning countries and political parties—under one roof. Westminster Foundation for Democracy will lead a panel on “Democratic innovation through AI in parliaments”. Together with the German and Hellenic Parliaments and other tech experts, the panel will explore the potential of AI in fostering innovation and resilience in parliaments worldwide, balancing technological opportunities with ethical and cultural complexities. Info and registration: https://www.politicaltech.eu/

About the authors

Franklin De Vrieze is Head of Practice Accountability at Westminster Foundation for Democracy (WFD).

By Alberto Mencarelli.

In the landmark paper “Attention Is All You Need” (Vaswani et al., 2017), the concept of transformers fundamentally changed the landscape of AI, particularly in how machines understand and generate human language through mechanisms like Generative Pre-trained Transformers (GPT) and Retrieval Augmented Generation (RAG). These breakthroughs not only represent significant technological strides but also strikingly parallel the intricate processes in parliaments. Just as transformers learn and adapt by absorbing and synthesizing vast amounts of information, so do parliamentary systems refine legislative measures by assimilating diverse data and viewpoints, underscoring a shared theme of progressive enhancement and complex data integration in both realms.

The essence of this analogy lies in the conceptual similarity between the AI ‘attention’ mechanism and the parliamentary ‘refinement’ process, where both prioritize and contextualize information to produce nuanced outcomes. The refinement process in parliaments evolves through stages: from committee to plenary, and across chambers in bicameral systems – each procedural step meticulously crafted to enhance output. In parliamentary terms, refinement involves a multifaceted iterative process that ensures broad deliberative bases, adversarial but inclusive consideration of diverse views, and the scrutiny of legislative provisions based on data-driven assessments. The similarity between AI systems and parliamentary processes shows that both share a fundamental approach: just as AI uses attention and retrieval mechanisms to process information, parliaments use iterative, knowledge-based methods to develop legislation. Thus, the dynamics of generative AI not only mirror, but also offer new perspectives on traditional methodologies of parliamentary activity, suggesting a unique intersection between technology and governance.

In the digital age, turning data into meaningful action is the cornerstone of progress. Generative AI, with its sophisticated attention mechanism, offers a new lens through which to examine the complexities of parliamentary discussions, identifying connections, trends, and implications that may elude human analysis. This does not mean replacing the human process, but rather enriching it, providing political actors with advanced tools to address contemporary challenges with a deeper understanding.

Linguistic Foundations: Bridging Parliaments and Large Language Models

At the core of both parliamentary procedure and Large Language Models (LLMs) lies a deep appreciation for the nuance and power of language, serving as the foundation for creating, disseminating, and negotiating knowledge.

The parallels between the linguistic dynamics of parliaments and those of LLMs are striking and suggest a natural synergy. Both domains achieve their goals through the meticulous refinement of language, optimizing their outputs to meet pre-defined objectives. This common linguistic underpinning presents a compelling case for the integration of LLMs into parliamentary processes. Such a union promises to enhance the clarity, precision, and accessibility of communication – a cornerstone upon which effective governance is built.

Leveraging AI can streamline legislative workflows and inspire explorations at the intersection of technology and parliamentary practices. By integrating technology and linguistics, we unlock innovative opportunities that could dramatically transform how parliaments and legislatures operate.

Ensuring Diversity, Inclusivity and Depth in AI-Augmented Parliamentary Processes

In recent years, pioneering experiments in some parliamentary settings have laid the groundwork for a more integrated approach to the use of AI tools (see this Popvox Foundation report for an updated overview of current international experiences). Real-time captioning of parliamentary proceedings, automated classification and transcription of policy documents, and the early stages of conversational document search are at the forefront of this integration. The leading parliaments include the U.S. Congress, whose House Subcommittee on Modernization issues regular flash reports on AI strategy and implementation (most recently in December 2023), the Brazilian Chamber of Deputies, which has developed the Ulysses artificial intelligence suite, and the Estonian Parliament. But other parliaments have launched or plan to launch pilots in the coming months, as in the case of the Italian Chamber of Deputies, which recently adopted a report.

Parliaments are on the cusp of a transformative journey, blending deep-rooted traditions with cutting-edge AI technology. While still early days, a promising path is unfolding. As AI rapidly evolves, it’s crucial for parliaments worldwide to share best practices and develop common approaches. This collaboration is key to tackling the challenges of this significant shift.

AI’s advanced language models can revolutionize how laws are made. These models are not only adept at producing well-crafted text but also at ensuring laws reflect a broader range of societal objectives and values. Such tools could make legislation more inclusive, weaving minority perspectives into the legislative narrative. By enhancing the drafting phase, AI could lead to laws that represent a more diverse range of opinions, softening the impact of decisions made by slim majorities.

The implications of generative AI in parliamentary contexts are profound, offering numerous possibilities for its application, such as:

Predictive Analysis: AI can forecast voting outcomes based on historical data, aiding in strategic negotiation planning.
Deliberative Refinement: Through quantitative analysis of debate intensities and MP motivations, AI can foster a more inclusive deliberative process. Projects like Inclusive.AI and the latest research highlights AI’s role in enriching democratic deliberation by capturing a broad range of opinions and value preferences. In an effective parliamentary AI system, the technology should not just simplify the aggregation of viewpoints but also ensure that the depth and variation of these viewpoints are preserved.
Moderation and Persuasion: AI moderation tools can enhance the quality and inclusiveness of debates, especially in committees or remote settings. According to empirical evidence, LLMs have the potential to enhance conversational persuasiveness and provide real-time, evidence-based recommendations that can improve participants’ perception of feeling understood in conversations. This, in turn, has the potential to improve the quality of conversations and reduce political divisiveness, without altering the content of the conversation (Argyle et al., 2023).
Data Analysis and Policy Simulation: Generative AI has the capability to analyze vast amounts of data from diverse sources and formats to identify pertinent variables, causal relationships, trends, and anomalies. This capability can be utilized to simulate alternative policy scenarios and predict the potential effects of public policies, while considering various factors and uncertainties.
Participatory Democracy: AI can gauge public opinion on legislative proposals, capturing not just the majority view but the intensity of support or opposition.
Hybrid Proceedings Facilitation: With the shift towards hybrid parliamentary models, AI offers tools for virtual participation, document sharing, real-time translation, and even emotional tone recognition, enhancing session efficiency and accessibility of hybrid sittings by addressing the challenges of virtual engagement.

Integrating AI into parliamentary processes could mark a major shift towards more efficient and inclusive governance.. By leveraging AI’s capabilities, parliaments can navigate the complexities of modern governance, ensuring that legislative processes are both reflective of and responsive to a broader range of societal needs and perspectives.

AI in Parliaments: Opportunities, Risks, and the Democratic Balance

As we contemplate the integration of AI into parliamentary processes, it becomes apparent that this endeavor is both promising and fraught with complexities. The ethical and the procedural implications of these tools for parliaments can shape entirely new functions and attributes. The potential of AI to enhance efficiency and decision-making is counterbalanced by significant ethical and operational challenges. Concerns regarding privacy, data protection, and the transparency of AI-driven decisions highlight the need for rigorous scrutiny. Furthermore, the importance of maintaining accountability in the face of AI’s potential errors or biases cannot be overstated. The task of customizing AI to complement political processes, while ensuring that its decisions remain accessible and understandable to the general public, underscores the critical balance between innovation and the foundational values of democracy. A notable challenge is the risk of AI being manipulated or misused, particularly in an age characterized by advanced cyber threats. In this regard, it seems crucial that parliaments strive to ensure that the introduction of AI tools in their processes is accompanied by maintaining full control over the AI technologies deployed in order to establish sufficient levels of technological sovereignty.

Parliaments are thus confronted with a pivotal challenge: to adopt AI through the development of stringent policies, the conduct of ethical testing, and the provision of comprehensive training. The stakes are significant, as failure to effectively integrate AI could result in falling behind in the ongoing technological revolution, potentially compromising the resilience and adaptability of parliamentary ecosystems.

Embracing AI in parliaments thoughtfully yet boldly requires striking a balance between innovation and democratic integrity, safeguarding the core values of these vital institutions.

About the author

Dr Alberto Mencarelli is a parliamentary official in the Italian Chamber of Deputies. He gained his PhD in constitutional law from Sapienza University of Rome.