Year: 2023
Type: conference-paper
Abstract: AI harms challenge conventional ideas about fault and causation. Rather than being clearly attributable to discrete negligent conduct (a bad line of code, a specific missing test), they emerge from the complex interactions of multiple inputs. They are hard to foresee and control, because AI systems operate autonomously, with a relatively open-ended range of possible outputs. And, after the fact, AI opacity makes proof of causation difficult. This paper argues that these difficulties need not prevent the attribution of responsibility, fault and ultimately liability, even under existing law. While regulatory interventions may be necessary to promote AI accountability, fault still has a role to play. The attribution of fault will easier to conceptualise (and prove) if the focus is on socio-technical, rather than technical failings. Inquiries into fault and liabilty for AI harms should focus on systemic risk management: what processes, organisational arrangements, allocations of responsibilities, and ways of designing and influencing the user experience and expectations, are in place. The question is not whether the system is error free, but whether these processes have created a safe (digital) ‘environment’.
Author: Fraser, Henry
Language: English
Subfield: Law
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7071136150