|
Abstract: Artificial intelligence (AI) is the most transformative technology of our era, affecting every industry sector and aspect of our lives. While AI promises enormous benefits, some of which are already manifesting, AI also has the potential to create many risks and problems, some of which are already starting to appear. Traditional command-and-control government regulation, referred to as “hard law,” barely exists for AI, and following the pattern of other technologies, is likely to be adopted incrementally in a trickle that will extend over future decades. Thus, for now, and for the immediate future, AI will be primarily governed by “soft law,” which consists of a variety of instruments creating substantive expectations that are not directly enforceable by governments. The primary problem with soft law is that because it is not enforceable, there are doubts about its effectiveness. This article provides the results of a two-year study on how to make AI soft law more effective and credible. It first summarizes lessons from decades of soft law governance of other technologies, including biotechnology, nanotechnology, information and communication technologies, and environmental technology. Next it identifies, analyzes, and draws observations and insights from over 600 existing AI soft law programs. Finally, building on the previous two tasks, it proposes a new Soft Law 2.0 model that consists of a toolbox of thirteen different mechanisms that can be used to ensure that soft law measures are implemented as intended, which should help make AI soft law more effective and credible.
|