Document(s): 

ethics_for_ai_-_way_forward_or_false_dawn_january_2020.pdf

External link: 

https://www.researchgate.net/publication/338840378_Ethics_for_AI_-way_forward_or...

Author(s): 

Brian Williamson

Year of publication: 

2020

This paper considers the proposed application of regulation based on ethics for machine learning-based algorithms (referred to as Artificial Intelligence or AI) and contrasts this approach with a counterfactual whereby markets, subject to public policy designed to correct market imperfections such as externalities and information asymmetries and subject to political choice, largely determine technology and market choices. Both approaches have ethical underpinnings, and proposals for regulation based on ethics for AI per se should be assessed relative to the counterfactual. Indeed, it remains unclear why AI per se should be the focus of ethical debate and proposed new rules. Placing a higher regulatory hurdle for machine-based versus human-based approaches to prediction and decision making would result in foregone opportunities for productivity growth, and potentially foregone health and safety benefits. Operationalisation of ethics for AI may also centralise decisions over what technologies are good, rather than relying primarily on decentralised innovation by entrepreneurs coupled with selection pressures, and error correction by venture capitalists and consumers. An explicitly ethical or rights-based approach to regulation may reduce scope for consideration of policy trade-offs in terms of economic welfare and broader political choice. An alternative approach, given the promise of AI, would be to forebear from setting higher standards for machine than human algorithms (individual, institutional and coded software based) whilst focussing on identifying and removing unnecessary barriers to the adoption and use of AI. Proposed new rules should be subject to a cost-benefit test, be made on a politically accountable basis and applied to existing as well as machine-based algorithms where justified. Change is likely required, but regulation based on ethics for AI per se arguably represents a policy false dawn.