Wednesday will be a fateful day in Brussels, a faraway city of which post-Brexit Britain knows little and cares less. It’s the day on which the EU’s AI proposals enter the final stages of a tortuous lawmaking process. The bill is a landmark (first in the world) attempt to seriously regulate artificial intelligence (AI) based on its capacity to cause harm and will soon be in the final phase of the legislative process – so-called “trilogues” – where the EU parliament, commission and council decide what should be in the bill, and therefore become part of EU law. Big day, high stakes, in other words.
However, the bill is now hanging in the balance because of internal disagreement about some key aspects of the proposed legislation, especially those concerned with regulation of “foundation” AI models that are trained on massive datasets. In EU-speak these are “general-purpose AI” (GPAI) systems – ones capable of a range of general tasks (text synthesis, image manipulation, audio generation and so on) – such as GPT-4, Claude, Llama etc. These systems are astonishingly expensive to train and build: salaries for the geeks who work on them start at Premier League striker level and go stratospheric (with added stock options); a single 80GB Nvidia Hopper H100 board – a key component of machine-learning hardware – costs £26,000, and you need thousands of them to build a respectable system. Not surprisingly, therefore, there are only about 20 firms globally that can afford to play this game. And they have money to burn.
