ABSTRACT
As Machine Learning (ML) systems are becoming increasingly used and more complex, different value systems are being coded into Artificial Intelligence (AI). In this paper, we delineate the value systems that many AI systems hold and suggest alternative value frameworks to combat inequities promoted by AI. Many contemporary AI value systems prioritize performance, generalizability, and efficiency which creates problems in the domain of equity. AI also often prioritizes an individualist framework over a collectivist one and carries the ‘ghost’ of homo economicus with it. This is a problem because economic rational choice models often do not consider community needs. We consider the frameworks of Multi-Agent Systems and Agent-Based Modeling, which prioritize a collectivist framework in AI as an alternative to other common systems such as deep learning. We argue that ML practitioners should prioritize and continuously evaluate the value systems of AI, and how their architecture choices inform those values, to protect and include marginalized groups.
Devgun, Ahan, Value, Homo Economicus, and Agent-Based Modeling: The Importance of Re-Evaluating the Value Systems of AI (February 4, 2023).
Leave a Reply