Decentered ethics in the machine era and guidance for AI regulation
- PDF / 634,401 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 107 Downloads / 199 Views
OPEN FORUM
Decentered ethics in the machine era and guidance for AI regulation Christian Hugo Hoffmann1 · Benjamin Hahn2 Received: 24 June 2019 / Accepted: 8 October 2019 © Springer-Verlag London Ltd., part of Springer Nature 2019
Abstract Recent advancements in AI have prompted a large number of AI ethics guidelines published by governments and nonprofits. While many of these papers propose concrete or seemingly applicable ideas, few philosophically sound proposals are made. In particular, we observe that the line of questioning has often not been examined critically and underlying conceptual problems not always dealt with at the root. In this paper, we investigate the nature of ethical AI systems and what their moral status might be by first turning to the notions of moral agency and patience. We find that their explication would come at a too high cost which is why we, second, articulate a different approach that avoids vague and ambiguous concepts or the problem of other minds. Third, we explore the impact of our philosophical and conceptual analysis on the regulatory landscape, make this link explicit, and finally propose a set of promising policy steps. Keywords AI ethics · Decentralization · Moral status · Moral patiency · Regulations
1 Introduction The EU recently published guidelines on ethical Artificial Intelligence (AI) (Financial Times 2019). In the same vein, the Swiss government wishes commercial AI algorithms to obey ethical standards and is conducting studies (The Federal Council 2018) on this topic, but has not yet found satisfying answers on how to achieve these standards and what they might be. In particular, the government aims to minimize the risk that AI systems might be used as scapegoats to evade responsibility for discrimination, accidents or other damages. However, to proceed with this agenda, but without engaging in a critical preliminary is, as recognized by Kant, not only to grope blindly after often ill-conceived solutions to possibly misdiagnosed ailments, but to risk reproducing in a supposedly new and original solution the very problems that one hoped to repair in the first place. Therefore, we contribute to conceptual clarity by tackling * Christian Hugo Hoffmann [email protected] http://www.er.ethz.ch/ Benjamin Hahn [email protected] 1
Entrepreneurial Risks, ETH Risk Center, SEC F 7, Scheuchzerstrasse 7, 8092 Zurich, Switzerland
Autonomous Systems Lab, ETH Zurich, Zurich, Switzerland
2
the subsequent questions which raise foundational and methodological issues.
1.1 Guiding questions • What are ethical AI systems? What is the moral status of
AI?
• To what extent do we need to have machine ethics? • What are the implications on AI regulations?
Exactly because we cannot resolve those issues entirely through objective measurement and analysis (science), a critical role emerges for philosophy and ethics (Kurzweil 2005: 380). In the remainder of this paper, we first review the literature on AI governance and conclude that there are shortcomings on the conceptual and phil
Data Loading...
