Foresight for ethical AI

Area of Science:

• Artificial Intelligence
• Human-Computer Interaction
• Sociotechnical Systems

Background:

• Growing expectation for AI developers to foresee and mitigate harms from AI creations.
• Difficulty in mitigating harms due to emergent behaviors in complex sociotechnical systems.
• Need for robust methods to ensure AI trustworthiness and resilience.

Purpose of the Study:

• To propose Naturalistic Decision Making (NDM) principles, models, and tools as a solution for mitigating AI-related harms.
• To demonstrate the applicability of NDM tools in identifying and preventing unintended consequences of AI.
• To use predictive policing algorithms as a use case to highlight the effectiveness of NDM.

Main Methods:

• Application of Naturalistic Decision Making (NDM) principles and tools.
• Utilizing NDM tools, such as the premortem, to uncover potential risks.
• Analysis of predictive policing algorithms to identify factors leading to ethical harms.

Main Results:

• NDM tools are well-suited for tackling the challenge of emergent behaviors in AI.
• NDM tools can uncover a reasonable set of risks leading to ethical harms.
• NDM has been used to develop more trustworthy and resilient AI.

Conclusions:

• NDM offers a viable framework for AI developers to proactively identify and mitigate potential harms.
• Implementing NDM tools can lead to AI systems that avoid unintended negative consequences.
• The use of NDM is crucial for responsible AI development, particularly in high-consequence domains.