Measurable Trust: The Key to Unlocking User Confidence in Black-Box AI
Year: 2024
Type: conference-paper
Abstract: Given the pervasive integration of artificial intelligence (AI) into our daily lives, establishing public trust is paramount for maximizing AI's benefits and ensuring its responsible use. This research proposes an investigation into the feasibility of developing a globally accepted, context-specific "trustworthiness score" for AI systems. We recognize that trust is a dynamic construct influenced by individual experiences, situational factors, and inherent user characteristics. We hypothesize that by quantifying behavioral manifestations of trust, such as user acceptance and confidence level during interactions, as well as incorporating expert assessments and ethical considerations, we can indirectly measure AI trustworthiness. This approach aims to create a standardized framework that can guide responsible AI development, mitigate potential risks, and empower users to make informed decisions about trusting AI systems. The proposed research is particularly relevant in high-stakes sectors like healthcare and finance where AI decisions can significantly impact individuals and society, underscoring the need for transparency, accountability, and robust mechanisms to evaluate and build trust in AI technologies.
Authors: Puntis Palazzolo, Bernd Stahl, Helena Webb
Institution: University of Nottingham
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
Awards: G301723715, G6136994381
OpenAlex ID: W4402659094