[Theme] Moderators of AI Trustworthiness

Theme Name

Moderators of AI Trustworthiness

Proposer(s)

Michael Habash, Esq., Managing Director, State Street

Summary

Trust in AI can be viewed as the willingness of people to accept AI and believe in the suggestions, decisions made by the system, share tasks, contribute information, and provide support to such technology. (Keng Siau, Weiyu Wang, "Trust is Hard to Come By: Building Trust in Artificial Intelligence, Machine Learning, and Robotics," (2018) Cutter Business Technology Journal Vol. 31, No. 2, pg. 47-53).

Large Language Models (LLM) are trained an enormous amount of human writing. This source material contains lies, deception, fraud, bias, and criminal behavior. Not surprisingly, the LLMs provide false information and produce unreasonable, illegal and destructive results at times. LLMs are often described as black boxes, "something goes in and a response comes out, and it's not clear why the model gave that particular response instead of another." (https://www.anthropic.com/research/mapping-mind-language-model).

Humans trust formation is complex, and there are several theories about how it happens. Three factors have consistently been linked to trustworthiness: ability, benevolence, and integrity (Borum, Randy, “The Science of Interpersonal Trust” (2010). Mental Health Law & Policy Faculty Publications. 574, Pg. 13).

Where humans have developed the ability to judge trustworthiness organically, we require a deliberate method for judging the trustworthiness of an AI Agent.

Furthermore, unlike people, agents lack reliable continuity. With people, if a person was trustworthy today, they will be the same person tomorrow and thus, we may start from a positive trust. But AI Agents are software. This software changes from time to time, without outward signs. These changes can fundamentally change its trustworthiness. How does one maintain trust with this thing, that has the same name, but different features.

Scope / Open Questions

1. What are the moderators of (e.g., features that change) an AI agent's trustworthiness?

Related Work

• https://digitalcommons.usf.edu/cgi/viewcontent.cgi?article=1573&context=mhlp_facpub
• Siuda, S., Schlösser, T., & Fetchenhauer, D. (2022). Do We Know Whom to Trust? A Review on Trustworthiness Detection Accuracy. International Review of Social Psychology, 35(1): 20, 1–16. DOI: https://doi.org/10.5334/irsp.623
• Keng Siau, Weiyu Wang, Trust is hard to come by: Building Trust in Artificial Intelligence, Machine Learning, and Robotics, Cutter Business Technology Journal, 31(2), 47-53.
• Afroogh, S., Akbari, A., Malone, E. et al. Trust in AI: progress, challenges, and future directions. Humanit Soc Sci Commun 11, 1568 (2024). https://doi.org/10.1057/s41599-024-04044-8

Related Themes

• https://github.com/FG-TIDA/themes/issues/32
• https://github.com/FG-TIDA/themes/issues/29

Discussion