The supportive AI framework: from recommending to supporting

Abstract

This paper presents the Supportive AI Framework, a conceptual framework for the design of human-AI collaboration to augment human cognition. AI-based decision support systems that are recommendation-driven (i.e. the AI makes a recommendation, and the human must decide whether to accept or reject it) often overstrain humans. The reason for this is the problem known as the ‘ironies of automation’, which occurs when humans are expected to supervise a technology that exceeds human capabilities. In terms of recommendation-driven AI, this is an impossible task for humans, as they must decide on AI-generated recommendations that take into account far more data and factors than humans are able to consider. Against this background, the Supportive AI Framework aims to go beyond recommendation-driven AI towards AI that explicitly supports cognitive processes of human decision-making, human learning, human trusting, and human motivation. This as a complement to providing comprehensibility through explainable AI and interpretable models. The Supportive AI Framework is theory-based and includes theories from the areas of natural decision making, experiential learning, intrinsic motivation, socio-technical system design and complementary function allocation.