A 'human-centric and lifecycle approach' to lega responsibility for AI

摘要

Discussion of legal responsibility has negative connotations with compulsion and punitive sanctions provoking some Al proponents to claim undue hampering of technological progress. But law exists to mediate between competing interests through rights, responsibilities, and remedies for harms suffered and wrongs committed. As demonstrated, sanctions are only one aspect of the function of responsibility, which also includes representation of victim interests and societal values. Beyond punitive obligations to rectify past misconduct, interaction between sanctions, victim interests, and societal values helps conceptualise prospective obligations that encourage responsible conduct and action to protect against and prevent harm. We can perceive Al as having an overall aggregate harm-causing potential to society, far greater than individual harm, which justifies utilising the precautionary and strict liability models. Where fundamental societal interests are at stake these models are justifiable as an effective means to minimise harm and encourage people to seek other means to avoid harm. The 'human-centric and lifecycle approach' combines precautionary and strict liability models to allocate responsibility across a spectrum of human agents. These agents are involved at different stages in the Al lifecycle and have specific protective and preventive obligations relating to user awareness, human rights, Al limitations, and non-deterministic influences on decision-making. This approach is predicated on maintaining responsible human design, development, and deployment of Al in order to avoid harm to individuals and society, thereby enabling public confidence in Al systems. Al systems wield a transformative social and economic capability with huge potential for individual and societal harm. Because of this potential it is entirely appropriate to adopt a 'human-centric and lifecycle approach' to legal responsibility.