Five Obstacles to AI Trustworthiness Alignment and How to Deal with Them: Introducing the DELIB-TRUST Framework

Abstract

Aligning the perceived trustworthiness and trusting behavior of users with the capabilities and characteristics of AI systems is considered central to the adoption of AI systems in society. However, achieving such a trustworthiness alignment faces major practical obstacles, at least with more complex notions of trust that go beyond mere reliance on an AI system. The present article identifies five fundamental obstacles that arise in the trustworthiness assessment and alignment process and that threaten to make any trustworthiness alignment unfeasible in practice. To address these obstacles, we introduce a novel DELIB-TRUST framework which goes beyond the conventional causal-behavioral framework of trust calibration and incorporates elements of deliberation and reason-giving. We evaluated our framework by conducting an expert study with 12 HCI/AI practitioners and provide actionable steps for its implementation. Through its evaluation, we show how the framework can help close important gaps that exist due to unacknowledged value-laden decisions and interpretative steps in efforts to achieve trustworthiness alignment.