Trust, Trustworthiness, and Assurance of AI and Autonomy
Abstract
The importance of trust in artificial intelligence and autonomous systems (AIAS) is widely touted, but not necessarily in productive ways. This paper argues that discourse about the importance of trust can obscure more than it illuminates, misleadingly suggesting similarities (and common solutions) across fundamentally unrelated challenges. In particular, it is important to distinguish whose trust is required, and whether that trust is subjective or objective. The paper also pushes back against the idea that trust can be built in as a property of the AIAS, and argues that assurance cases, as originally developed by the safety and security communities, provide a framework for understanding trustworthiness, disambiguating different concepts of trust, and enabling test, evaluation, verification, and validation (TEV and V) approaches to AIAS development and employment.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 2021
- Accession Number
- AD1150274
Entities
People
- David M. Tate
Organizations
- Institute for Defense Analyses