Bryson warns EU AI Act may use the wrong definition of AI

The definitional fight over the EU AI Act is the kind of policy detail that looks dry on paper and matters enormously in practice, because "what counts as AI" is the same question as "what has to comply." In a guest essay for Wired, Joanna Bryson, Professor of Ethics and Technology at the Hertie School, argues that Europe is edging toward a definition that is too narrow, and that this is bad for both businesses and citizens.

Her specific concern is with the push to restrict the Act's scope to only "complex" machine learning or other reasoning "ordinarily understood" to be intelligent. Bryson's read is that this narrowing would leave a real amount of consequential software outside oversight, because plenty of decision-influencing systems in the wild do not look like anyone's mental image of AI, and yet still shape outcomes for citizens.

The counterweight, and this is where the piece is honest, is that the definition cannot be too broad either. Bryson's own concern with a sprawling scope is that it might "push companies and governments to use the second-best system" specifically to sit outside the rules. What she advocates instead is a simple, broad definition of artificial intelligence that stays clear and maintainable over time, rather than a boundary drawn around whichever techniques feel intelligent this year.

The honest caveat is that this is an op-ed rather than a leaked draft, and it does not tell you exactly which systems the current text would exempt, nor how the "human-defined objectives" language in the Act's annex plays out against generative models. But the underlying point is worth carrying into any conversation about AI regulation: the scoping clause is the whole game. Legal, compliance, and product teams working on lightweight statistical or rules-based systems should not assume they are automatically out of scope.