Broad Exchange on the Published Guidelines on the Introduction and Use of Artificial Intelligence in the Parliamentary Workspace Where do we go from here?
0 likes90 views
The document summarizes a research workshop on artificial intelligence in parliaments. It discusses the motivation for developing guidelines on introducing and using AI in parliamentary workspaces. It provides an overview of the state of play in using AI in legislative chambers in different countries. It then outlines the regulatory framework developed by an ad hoc working group, including sections on ethical principles, artificial general intelligence, privacy and security, governance and oversight, system design and operation, and capacity building and education. Next steps discussed include refining version 2.0 of the guidelines and extending the working group.
Broad Exchange on the Published Guidelines on the Introduction and Use of Artificial Intelligence in the Parliamentary Workspace Where do we go from here?
- 1. Research Workshop - AI in Parliaments Broad Exchange on the Published Guidelines on the Introduction and Use of Artificial Intelligence in the Parliamentary Workspace – Where do we go from here? 3-7-2023 Dr. Fotis Fitsilis Scientific Service, Hellenic Parliament
- 2. Structure ● Motivation ● State-of-play ● Regulatory framework (making of) ● Conditions and limits ● Next steps 2 DALL·E 2, June 12, 2023, „rotating gears“
- 3. Motivation ● AI-based tools & services gain momentum ● Emergence of ChatGPT & related LLM’s ● Introduction in public sector institutions: Parliaments ● Broad call for regulation ● Determination of regulatory parameters ● What about parliaments? 3 DALL·E 2, June 12, 2023 „motivated person“
- 4. State-of-play ● Use of AI in the parliamentary workspace ● Where?* ○ 39 AI-based solutions ○ 10 legislative chambers in 8 countries ● Guideline development (V1.0, April 2023) 4 *As of late 2022; source: Fitsilis & de Almeida, forthcoming DALL·E 2, June 12, 2023 „Awkward situation“
- 5. Ad hoc working group 5 DALL·E 2, June 12, 2023 „Artificial Intelligence in parliaments“ 1. Fotios Fitsilis, Hellenic Parliament (coordinator) 2. Jörn von Lucke, Zeppelin University 3. George Mikros, Hamad Bin Khalifa University 4. Jonathan Ruckert, NovaWorks 5. João Alberto de Oliveira Lima, Brazilian Federal Senate 6. Ari Hershowitz, Govable.ai 7. Bruce Philip Todd, expert on Justice and Home Affairs 8. Sotiris Leventis, Hypernetica
- 6. 6 • Open access (EN): https://doi.org/10.6084/m9.figshare.22687414 • Also available in EL, DE, ES (forthcoming) • Eight original working group members • Inter-sectoral & multi-disciplinary • Work in progress The guidelines Expert distribution in sectors
- 7. Geographical distribution 7 US BR UK DE GR x2 QA AU Geographical distribution of experts
- 8. Regulatory framework ● Introduction & use of AI-based ParlTech ● Establish ethical & operational guidelines ● Ensure accountability, transparency, human autonomy & promote SDGs ● Uphold the rights enshrined in the Universal Declaration of Human Rights ● Guideline development in sections 8
- 9. Dissecting the guidelines ● Starting point: ○ Regulatory guidelines (scholars and practitioners) ○ Corporate AI principles (IBM, Google, Microsoft, …) ○ Organizational frameworks (EU, UNESCO, OECD, …) ● Six sections: ○ Ethical principles ○ Artificial General Intelligence ○ Privacy & Security ○ Governance & Oversight ○ System Design & Operation ○ Capacity Building & Education 9
- 10. a. Ethical principles I ● Ensure accountability and transparency in the use and deployment of parliamentary AI systems; ● Respect human dignity and privacy in all aspects of AI development and implementation in the parliamentary workspace; ● Apply principles of fairness, equity and non-discrimination in the use and deployment of parliamentary AI systems; ● Apply principles to understand the input training data to avoid potential biases in the underlying data and algorithms employed; ● Avoid using training data that breaches intellectual property rights; 10
- 11. a. Ethical principles II ● Preserve human values and cultural diversity in parliamentary AI design and implementation; ● Evaluate and mitigate unintended consequences or collateral damage from the introduction and use of parliamentary AI systems, with assessments and recommendations from the evaluation of existing systems where necessary; ● Encourage public participation and engagement in developing, implementing, and overseeing parliamentary AI systems to ensure inclusiveness and representation; ● Respect the rule of law and democratic values in the development and use of parliamentary AI; ● Promote the UN’s sustainable development goals through the use of parliamentary AI. 11
- 12. b. AGI and Human Autonomy ● Promote human autonomy and decision-making, including ensuring that parliamentary AI is used to supplement human capabilities rather than replace them; ● Assess designers and developers of parliamentary AI systems for their ethical responsibility to consider the potential impact of such systems for institutions and society at large. Security vetting of developers should be required; ● Promote understanding that the evolution of AI is progressive and that AGI should be considered as a real prospect. 12
- 13. c. Privacy and Security I ● Embed safety and robust security features into parliamentary AI systems, with a focus on preventing harm to individuals, the intranet, or the institution as a whole; ● Include cybersecurity and existing privacy protections in the design and deployment of parliamentary AI systems; ● Understanding that AI systems may process Personal Identifiable Information (PII), ensure that this information is secured and that appropriate internal and external safeguards are in place; ● If outsourcing AI/AGI, understand what is stored, processed and captured in the parliamentary AI system, with particular concern for privacy; ● Understand and agree to any data sovereignty issues when processing data; 13
- 14. c. Privacy and Security II ● Recognize that AI should not replace original source material, its goal should be to supplement in the parliamentary space; ● Recognize that an overreliance on AI may introduce a risk where the Parliament may become too reliant on these systems, leading to a false sense of security; ● AI systems are still vulnerable to cyber security attacks; it is important to secure the training data of any parliamentary AI system from adversaries that may attempt to retrain a system to interact in a specific way; ● Human oversight is still important, security decisions should still be escalated to be made by human operators. 14
- 15. d. Governance and Oversight I ● Embed the design and implementation of AI systems into a broader digital parliamentary strategy; ● Utilise efficient data governance and management protocols that ensure the accuracy completeness, and security of data used in AI systems; ● Create and empower a parliamentary ethical oversight body or add the task to an existing oversight committee that reviews parliamentary AI systems and applications; 15
- 16. d. Governance and Oversight II ● Assess the effects of parliamentary AI on a wide range of issues, including intellectual property, liability and accountability, employment and labour, socio-economic issues, privacy and data protection, bias and discrimination, national security and defence, ethical governance and oversight, and environmental matters; ● Ensure transparent, yet secure access to and control over the data used in parliamentary AI systems; ● Strive to cooperate with stakeholders from various sectors, including other parliaments (interparliamentary cooperation), academia, civil society, industry and others, to develop resilient policies and regulations that foster innovation while protecting human rights. 16
- 17. e. System Design and Operation I ● Implement standardised data schemes and processes to ensure interoperability and compatibility across different platforms and applications; ● Emphasise AI algorithms' explainability, ensuring that the reasoning behind AI-driven decisions and recommendations is clear, understandable, and accessible to relevant stakeholders, fostering trust and enabling informed decision-making; ● Build robust and reliable parliamentary AI systems that include the ability to detect and correct errors and failures; ● Regulate the use and deployment of parliamentary AI systems, including risk assessment, licensing requirements and safety standards; 17
- 18. e. System Design and Operation II ● When utilising a Software as a service (SaaS) product or cloud based implementation that contains AI features, perform a risk assessment to ensure that ethical and other protections are ensured; ● Regularly monitor and evaluate the operation and output of parliamentary AI systems to assess their impact on society and the environment; ● Accuracy of AI is dependent on the application and the intended use of the implemented system; a minimum level of accuracy should be agreed with all relevant stakeholders before acceptance of an AI implementation. 18
- 19. f. Capacity Building and Education ● Build and expand an expert team to keep up with technology innovations in the field of AI and beyond; ● Organise frequent training programs on AI for parliamentary officials and administrators for the development of critical AI literacy skills and the promotion of promote its responsible and ethical use; ● Support knowledge exchange with external stakeholders and the participation in bi and multilateral cooperation schemes; ● Document the steps toward and the results of AI-related activities to build up institutional memory and disseminate knowledge to internal actors; ● Inform the public about the use and limits of AI in Parliament in an understandable manner. 19
- 20. Next steps ● Working towards version v.2.0 ● Timeline (late 2023) ● Sectoral approach (preliminary; six distinct sub-groups) ● Further refinement/widening of sectors possible ● Coordination by the Hellenic OCR Team ● Center of weight in operational sectors ● Spark interest through presentation & networking ● Extend working group & community building ● Translation of the guidelines in further languages 20
- 21. Contact & disclaimer www.fitsilis.gr fitsilisf@parliament.gr 21 The information and views set out in this presentation are only those of the author and do not necessarily reflect the official opinion of the Hellenic Parliament DALL·E 2, June 12, 2023 „people linked to each other“