On the Brussels-Washington Consensus About the Legal Definition of Artificial Intelligence

Abstract

The global debate on regulating Artificial Intelligence (AI) has led to a tentative consensus between Brussels and Washington about the legal definition of AI, yet differences persist. The EU's AI Act, still under discussion at the time of writing, includes an explicit reference to content as one of the outputs of AI systems. Notably, President Biden's Executive Order overlooks this significant output, which is pivotal in areas like education, media, business, and politics, raising concerns about coherence in the documents' applicability. Both documents stress human accountability, attributing AI misuse to human decisions rather than AI itself. This is in line with ISO’s definition. Meanwhile, the OECD's revised definition removes the 'human-defined objectives' clause, potentially allowing for speculative AI autonomy. Within the EU's AI Act, subsequent drafts and modifications lack clarity, complicating consensus building. A suggested unified definition frames AI as “an engineered system that can, for a given set of human-defined objectives, generate outputs – such as content, predictions, recommendations, or decisions – learn from historical data, improve its own behaviour, and influence people and environments”. This proposed amalgamation aims to bridge the gaps between existing definitions, aiding in establishing a cohesive regulatory framework across the Atlantic.

Keywords: AI Act, Definition of Artificial Intelligence, Executive Order, ISO, OECD