Related Experiment Videos
Beyond the algorithm: rethinking the network account of trustworthy ai through lexical threshold-based
Fei Song1, Julian Savulescu1,2, Michael Dunn1
1Centre for Biomedical Ethics, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore.
AI & Society
|July 13, 2026
Summary
This paper introduces a new framework for trustworthy artificial intelligence (AI) called the network account. It proposes Lexical Threshold-Based Multidimensional Utility Theory (LTMU) to manage conflicting AI trust attributes effectively.
Area of Science:
- Artificial Intelligence
- Computer Science
- Ethics
Background:
- Existing conceptual frameworks for trustworthy AI are reviewed.
- The network account is proposed as a superior alternative for AI trustworthiness.
- The importance of defining nodes, attributes, and thresholds in AI networks is highlighted.
Purpose of the Study:
- To introduce and detail the novel Lexical Threshold-Based Multidimensional Utility Theory (LTMU) framework.
- To establish a structured method for assessing AI trustworthiness.
- To demonstrate the practical application of LTMU in resolving attribute conflicts.
Main Methods:
- Identification and illustration of nodes within an AI network.
- Specification of key attributes and their sufficient thresholds for each node.
- Development of the LTMU framework with lexical ranking, minimum thresholds, and utility scales.
- Analysis of exceptional cases involving conflicting trust-relevant attributes.
Main Results:
- The LTMU framework provides a systematic approach to AI trustworthiness.
- Application of LTMU to conflict scenarios yields reasonable and defensible outcomes.
- The network account, underpinned by LTMU, offers a robust model for trustworthy AI.
Conclusions:
- The network account and LTMU framework offer a significant advancement in trustworthy AI research.
- LTMU effectively addresses complex challenges in AI trust assessment.
- The proposed framework has practical implications for developing reliable AI systems.