Scientists release new ‘four as’ framework to tackle key ethical challenges governing AI

Kirti Prakash and Ian van der Linde of Anglia Ruskin University's Advanced Computing Research Centre propose the 'Four As' framework in the journal AI and Ethics, arguing that the common 'slow down versus accelerate' AI debate is too blunt. The framework separates AI governance into four distinct ethical problems: Action (whether humans retain effective control over what a system does), Access (who may obtain or use a dangerous capability), Authority (which judgements may be delegated to a system), and Allocation (how benefits, burdens and power are distributed). The authors apply an 'independence test,' arguing that solving one problem does not resolve the others. They identify two failure modes: governance mismatch, where an intervention addresses a different problem than the one at issue, and cross-locus displacement, where reducing concern in one category intensifies it in another—for example, restricting access reduces misuse risk while concentrating power. The authors note that a global slowdown is an incomplete objective, since capability is no longer confined to single organisations and politics and commercial incentives complicate restraint. Drawing on imperfect precedents such as the 1975 Asilomar conference on recombinant DNA and nuclear arms control, they propose an 'AI stability framework' with a minimum ethical floor, category-specific constraints, identifiable enforcers and scheduled reassessment. They acknowledge the framework is conceptual rather than tested, does not rank risks by severity, and is not a complete inventory of AI harms.

(Source：The Brighter Side of News)