Minds and Machines Special Issue: Ethics and Artificial Intelligence

Paula Boddington, Peter Millican, Michael J. WooldridgeMinds Mach 2017 · DOI · OpenReview

Abstract

This paper serves as the editorial introduction to a special issue of Minds and Machines, synthesizing four key perspectives on the intersection of ethics, law, and artificial intelligence. It argues for a shift away from speculative "magical" thinking about superintelligence toward a focus on "sociotechnical ensembles"—the practical systems where humans and machines interact in fields like finance and law. By framing AI ethics through interdisciplinary lenses, the authors provide a roadmap for navigating the "control problem" and the ethical vacuum created by rapid technological advancement.

Introduction

The rapid advancement of AI has created a significant "control problem" and an ethical vacuum. Our existing moral and legal frameworks are primarily tailored for human agents with biological constraints, yet AI operates at electronic speeds with vast data processing capabilities. This paper addresses the fundamental gap in defining "intention" and "responsibility" for machines, which is essential for regulating autonomous decisions in sensitive areas like criminal law and global financial markets. If we cannot define these concepts for machines, we cannot effectively trust or regulate them.

Interdisciplinary Perspectives on AI Ethics

The editorial employs conceptual synthesis and interdisciplinary review to categorize research into four distinct approaches:

1. The Sociotechnical Ensemble

Authors Verdicchio and Johnson argue that the term "autonomy" often triggers public fear because it implies a machine acting in a vacuum. In reality, AI is "embedded" in human-designed systems. We suffer from "sociotechnical blindness" when we imagine a superintelligence acting without human cooperation. Understanding AI as part of a larger system of human behavior and social arrangements is crucial for ethical analysis.

2. Modeling Moral Uncertainty

Given that humans cannot agree on a single "correct" moral theory, Kyle Bogosian suggests that we should not attempt to program one into AI. Instead, AI should be programmed to recognize moral uncertainty and use a "voting" mechanism among different ethical frameworks (e.g., Utilitarianism vs. Deontology) based on assigned "credence" levels.

3. The Intentionality Gap in Finance

Financial markets serve as a "canary in the coal mine" for AI ethics. Wellman and Rajan point out that current laws rely on proving "intent" to distinguish between legitimate innovation and market manipulation. This concept does not easily translate to an algorithm performing arbitrage at millisecond speeds, creating a regulatory gap.

Traditional rule-based ethics (Deontology) may be too rigid for the unpredictability of robotics. Ugo Pagallo proposes that "Virtue Ethics"—which focuses on flexibility and character—might be a better fit for "robotic testing zones" where legal norms are still evolving. While civil law (contracts/torts) can handle AI through "strict liability," criminal law faces a crisis because it cannot yet attribute "guilt" or "free will" to a machine.

Results and Future Directions

The synthesis provides a qualitative roadmap for the future of AI ethics, emphasizing that autonomy should be viewed as a matter of degree based on the level of human intervention required. Financial trading is highlighted as a useful model for AI ethics because it is a tightly regulated landscape where AI is already pushing systems to their limits.

However, several limitations remain. There is a need for more empirical research into public attitudes toward AI. Furthermore, the "moral uncertainty" approach faces the problem of an infinite regress: if we cannot agree on moral theories, how do we agree on the "weights" or "credences" assigned to them? Finally, while ensuring a "human-in-the-loop" is a viable alternative to fully autonomous machines, it becomes increasingly difficult as AI speed outpaces human cognition.

Summary updated

BibTeX
@article{boddington2017minds,
  author = {Paula Boddington and Peter Millican and Michael J. Wooldridge},
  title = {{Minds and Machines Special Issue: Ethics and Artificial Intelligence}},
  journal = {Minds Mach.},
  year = {2017},
  doi = {10.1007/s11023-017-9449-y},
}
Figure