At our last EUROSFAT forum in Bucharest, industry leaders and artificial intelligence (AI) experts gathered for a thought-provoking discussion on the EU’s proposed AI Act (AIA). A comprehensive framework for AI governance is planned to be established by the EU through its groundbreaking legislative endeavour. Its main goal is to classify AI applications based on the degree of risks involved, addressing the growing concerns over privacy, openness, and the moral use of AI.
A number of important revelations came to light during our lengthy conversation, underscoring the difficulties and future directions in this field. The crucial necessity of striking a balance between fostering research and guaranteeing strict regulation was a recurring subject. There was a real concern that excessively stringent restrictions might impede the development of technology, particularly in industries where AI plays a major role in growth. The discussion also highlighted the possible function of regulatory sandboxes. These regulated spaces, which provide a secure area for innovation within a controlled framework, may prove to be extremely beneficial for startups and small-to-medium-sized businesses.
The panellists’ agreement on the divergent effects of AIA in various geopolitical contexts was another important area of debate. This realisation highlighted the need for adaptive and flexible regulatory strategies to meet the various needs and contexts found in different regions. Furthermore, everyone agreed on the significance of awareness and education for the future of AI. Initiatives to educate the public and industry about the advantages and hazards of this paradigm shifting technology were strongly urged. It is clear that managing AI in the future will involve more than just passing laws; it will also require cross-sectoral collaboration between academia, industry, civil society, regulators, in-depth knowledge, and a dedication to forming a society in which AI is applied morally and responsibly.
Following the analysis of the discussion at EUROSFAT, our fundamental stance supports a risk-based, nuanced approach to AI regulation that seeks to protect fundamental rights while fostering innovation. The classification and regulation of High Impact Foundational Models (HIFMs) and General Purpose AI (GPAI) are of particular interest to us. We unequivocally reject the proposed multi-tiered framework on the grounds that it may result in the unjust exploitation of particular vendors and models. It is our contention that the delineation of HIFMs ought to be predicated on tangible risk as opposed to random quantity or numerical thresholds. It is of utmost importance that any criteria for these models be applied prudently and solely subsequent to a model being suitably classified as an HIFM.
Furthermore, our viewpoint is thoroughly opposed to compulsory external assessment of AI models prior to and subsequent to their commercial release. This is deemed impractical and potentially hazardous to sensitive information. On the contrary, we suggest the adoption of a more focused strategy that evaluates risk with greater contextual specificity. At the same time, we discourage
EUROPULS – Conclusions from the EUROSFAT AI debate
Excessively broad prohibitions in AI applications, such as a blanket ban on biometric identification in general. It is our suggestion that a more discerning approach be taken, with a specific emphasis on systems that manifestly endanger both individuals and collectives. This nuanced viewpoint permits the advantageous applications of AI while safeguarding individual liberties.
Regarding governance, we suggest the formation of a scientific advisory council tasked with supervising the regulatory framework overseeing foundation models. By integrating scientific and industry knowledge, this council would guarantee that the regulation is well-informed and applicable in the real world. We advise against the decentralisation of the authority that begins investigations and propose that, in order to preserve transparency and circumvent legal complexities, a centralised approach be adopted.
Lastly, we convey our apprehensions regarding the expansion of high-risk categories within the AIA in the absence of comprehensive impact assessments. These expansions have the potential to disrupt the digital objectives of Europe. We underline the significance of upholding equilibrium within the AIA by simultaneously facilitating the progress and integration of AI throughout the union and ensuring the implementation of essential safety protocols.
EUROSFAT Forum is the project of the Europuls – Centre of European Expertise and the national winner of Charlemagne Youth Prize in 2022 for the best project that enhances European Values.
About the Author
Corneliu Coteț is a doctoral researcher at the Centre for Information Management at Loughborough University. He researches green mobility, infrastructure resilience, and disruptive technologies, but his major focus is establishing an automated decision-making system for train infrastructure management during harsh weather. Corneliu worked in academia and public policy. He completed the European Parliament DG-EPRS Schuman Trainee Programme in Transport Policy. He was Associate Tutor for Transport and Logistics, Lecturer in Strategy and Leadership, and Postgraduate Researcher in Transport Infrastructure Resilience. Corneliu has a Master of Research in Geospatial Data Science from Newcastle University, a Master of Science in Transport Planning from Leeds Institute of Transport Studies, and a Bachelor of Science in Air Transport & Logistics Management from Huddersfield University.