General Purpose Artificial Intelligence Systems and the European Commission's Proposed Regulation
Dalloz IP/IT, February 2022
4 Pages Posted: 8 Aug 2022
Date Written: 2022
Until the last thirty years, when one thought about artificial intelligence (AI), they would think about a general intelligence, comparable to ours, capable of performing various tasks. However, to date, there is no artificial intelligence as was envisaged at the time.
Instead, single-task artificial intelligence systems have been created that are capable of outperforming humans at specific tasks such as calculus or chess. Moreover, whenever an AI becomes capable of a task that was thought to be very complex, such as driving, many people conclude that this task did not require so much intelligence after all.
Therefore, today, when one hears "artificial intelligence", they think of a restricted and mono-task intelligence. However, the latest progress in AI is proving us wrong again, in that it leads to increasingly general and multi-functional systems. The European Commission has fallen into this trap with its legislative proposal on artificial intelligence. The text is innovative and ambitious because it is the first in the world to propose a general regulation on AI with a broad geographical scope. Nevertheless, it is not adapted to the evolution of AI because it imposes ex-ante safety requirements based on the field of application of the AI system. AI systems are becoming more and more multitask and general purpose, dooming the classification of the proposed regulation to become obsolete.
Keywords: AI Act, General-purpose AI, safety requirements, GPAIS
Suggested Citation: Suggested Citation