Abstrakt
As artificial intelligence (AI) will continue getting integrated into important places in our society and getting more capable, the associated risks may start becoming too large and start coming too quickly for reactive regulation to be effective. At the same time, humans' ability to foresee future harms is inevitably limited, especially at the level of specificity usually required for detailed legal regulation.
If these two considerations limit our ability to get specific in the rules we are laying down, a path forward is to lean onto more general principles, but give them teeth by making them legally binding. We think negative human rights could go a long way in protecting us against unforeseen harms from AI, but we need to go beyond how they are understood today in human rights law.