On October 2nd, 2023, the Institute of Public Policy – Lisbon (IPP) hosted a PromethEUs Network workshop on the topic of “Artificial Intelligence: Opportunities, Risk and Regulation” with a specific focus on policy developments at the EU level.
Professor Paulo Trigo Pereira, Director of the IPP and Professor at ISEG, and João Cadete de Matos, Chairman of ANACOM, the Portuguese communications regulator, opened the workshop. Cadete de Matos described AI as a crucial topic in their work with BEREC, the group of EU communications regulators, and other international fora of communication regulators, as well as with the UN. AI is considered globally to have the potential of reducing asymmetries and inequality among countries. While this is one of many hoped-for outcomes, AI can also be a powerful tool used against the consolidation of democracies, peace, and the respect of human rights. All these dimensions are at the centre of the concerns of Portugal, the EU and the UN. Cadete de Matos highlighted the double goal of minimizing the risks and maximizing the benefits of AI through regulation, which will thus be indispensable for its future development. Yet, the introduction of rules for AI will be complicated: As the attempts to police the internet showed, the regulation of AI will require fast and agile cooperation between countries in order to be effective. He concluded by calling for this cooperation and for the EU to attribute to regulators the powers, competences, and legal instruments to ensure the correct development of AI.
The workshop consisted of two panel sessions. The first panel consisted of presentations by the four PromethEUs think tanks of their ongoing work for a joint policy paper on the EU AI Act. For IPP, Steffen Hoernig discussed the regulatory aspects of the EU AI Act. He provided a brief description of the European and international backgrounds, the regulatory approach of the AI Act, some important open issues currently under discussion in trilogue, topics concerning the implementation of the Act and its future. Stefano da Empoli from ICOM discussed the impact of generative AI, providing a brief definition and outlining the difference with analytical/traditional AI. He then provided forecasts of its economic potential and value in global markets and some thoughts on the absence of EU countries at the current frontier developments in this area. Aggelos Tsakanikas from IOBE explored the Southern EU perspective, focusing on current policy regimes, AI readiness, and economic potential in Italy, Spain, Portugal, and Greece. Finally, Raquel Ricart of Elcano Royal Institute discussed the geopolitics of generative AI, concerning security, economy, rights, and international governance, and commented on the current EU approach on these topics.
The second panel included Gabriel Osório de Barros, Director of Services for Economic Analysis at the Office for Strategy and Studies of the Ministry of Economics of Portugal, Father Paolo Benanti, Professor of Digital Ethics, Pontifical Gregorian University of Rome, and Panos Louridas, Director of R&D at GRNET S.A. and Associate Professor, Department of Management Science & Technology, Athens University of Economics and Business.
Gabriel Osório de Barros first signalled the positive side of AI, which could serve as a powerful policy tool for current political compromises such as sustainability goals, as it crosses sectors and has an interconnected nature. He then mentioned three studies that position Portugal globally with respect to important measures for AI. Portugal takes 17th place in the Global Innovation Index published by the World Intellectual Property Organization (WIPO), 14th place in the McKinsey Global Survey on AI Readiness, and 2nd place for SMEs’ usage of at least one AI technology in the EU (according to Eurostat). Osório de Barros then pointed out several threats and concerns, such as its impact on social biases and jobs. AI can perpetuate biases if systems and models are fed with biased datasets, if these exclude communities, languages or any other identity source in its data. Finally, he identified three potential issues to address when regulating AI: 1) overlaps with existing regulation (example of the GDPR); 2) the current ever-evolving nature of AI development may quickly make any existing regulation obsolete; and 3) the possibility of overregulation, i.e., that red tape may impede growth and innovation in AI – especially relevant for the current EU preparations of the AI Act.
Father Benanti started his contribution by stating that every technology has the power of misplacement and exclusion. One question with AI then is the same as with previous technologies: How can ethics and technology fit together? AI brings many potential benefit and problems. For example, AI will be able to save lives by predicting what is wrong with a turbine in a space station orbiting the Earth. But AI can be applied not only to mechanical but also to biological systems. Or take the case of many online shopping AI systems suggesting highly appellative products based on previous behaviour and preferences. Now, these AI systems are not only predicting the output, but also producing the output, in this case, a specific behaviour. Father Benanti pointed out that AI is acquiring the force to produce results or outputs similar to Law, without sharing Law’s fundamental characteristics: being knowledgeable, universal and general. AI also brings with it important social and political problems. For instance, AI’s ability to predict may replace not only manufacturing but also many white-collar jobs, precisely because AI can learn from the know-how and experience of current workers from the data their actions produce. This means that it is possible that the best-paid jobs are the next ones to be replaced. How will societies react to this? How will they be transformed? AI’s labour-replacing possibility needs not to be a problem. Father Benanti mentioned that for parts of Southern Europe (or any country facing a demographic winter) it can be a salvation. In several Italian regions, annually there are more people retiring than entering the job market, hurt this region’s productivity. By helping to maintain economic activity, AI may assume a fundamental role in developing the distribution of richness that allows democracy to work and keep all regions integrated.
Panos Louridas focused on the increasing interaction between AI providers and the public sector. To start with, AI models exist thanks to data, including data that is collected, stored, and made available by public authorities. This raises new issues in the areas of intellectual property rights and authorship of data. For example, when an AI system instead of using the text or a line of an identifiable academic work, uses fragments of text or words, can the AI system even know its authorship? The public sector may also acquire an active role in improving the quality of AI systems. For example, models are trained with data in specific widely spoken languages (primarily English). Communities or cultures with less visibility in the data (“resource-lean languages”) will not be fairly represented. Louridas pointed out that public sector data creators can work with data providers to help enrich AI models in the case of resource-lean languages. Finally, developing AI models may be way too expensive for smaller actors to participate in the AI revolution. By pooling resources, public authorities could provide access not only to data but also to AI models themselves.
Summing up, both panels not only highlighted various risks associated with AI, but also pointed out numerous opportunities and constructive recommendations for the future of AI governance. We are excited to announce that on November 14th, the PromethEUs network will host a follow-up workshop in Brussels, where we will delve deeper into these critical topics. Additionally, we will present the finished policy paper on the AI Act, incorporating the insights and discussions from the Lisbon event. We look forward to seeing you there!