Ethics or Power? Alternative Frameworks for Regulating AI

• In Event: Algorithms and Democracy

Abstract

No longer the realm of science fiction, Artificial Intelligence or AI, that reasonably approximates or surpasses human action or capacity become common. In fact, with some creators of AI actively pushing “scaling maximalism” as a way to make bigger and better models to do bigger and better things with computers, there is a very real sense that the possibility of Artificial General Intelligence, or AGI, may be upon us soon, perhaps in this century.

It's therefore no surprise that there have been a proliferation of calls for AI ethics. Yet, many extant forms of AI ethics are stuck in the past, concerning themselves with futuristic problems, like sentience, or AI that are indistinguishable from humans. Other forms of AI ethics seeks to engage with the problems of the now, and attempt to provide guidelines for the usage of AI in our modern society. But as AI is a continually mutating and advancing set of technologies, in order to understand the future of AI ethics and governance, we need to first recognize how power permeates the technology space. In short, we need to understand the creators of the technology and their values and interests.

Our paper is therefore a call to use the logic of power to explain the creation of AI ethics. First, AI has power to affect society and political relations. That AI causes all kinds of harms have become well-documented events. It’s clear these harms are not isolated and they are not random. They arise because of the design of AI systems. But not all AI systems cause the same kinds of harm. It has become clear that the AI for which ethical systems are designed is unclear – is it a recommender system? Facial recognition algorithms? Natural language processing? – because AI has become a catch-all category. This leads to the second point: how power permeates these algorithmic systems. The relationship between power and knowledge that Michel Foucault describes is a key one for datafication and AI, and is especially relevant as those with access to technical information, in terms of understanding the precise makeup and functioning of these technical systems, and those who provide the sorting and analytical purposes to algorithms. "Big Tech" companies have a power advantage over governmental regulatory bodies, who may lag in technical knowledge or because of the policymaking process. They also have advantages over most of the public who are subjected to the products of these technological developments. We argue that developing a full understanding of this power dynamic is crucial for any attempt to regulate or constrain the harms of AI.

Authors

• Stefan Kehlenbach, University at Albany

• Wendy H. Wong, University of British Columbia