Legible AI by Design : Design Research to Frame, Design, Empirically Test and Evaluate AI Iconography

Pilling, Franziska and Akmal, Haider Ali and Gradinar, Adrian and Lindley, Joseph and Coulton, Paul (2020) Legible AI by Design : Design Research to Frame, Design, Empirically Test and Evaluate AI Iconography. In: Common Good Framing design through pluralism and social values : Design as Common Good. Swiss Design Network, CHE, pp. 2442-2459.

Text (Legible AI by Design: Design Research to Frame, Design, Empirically Test and Evaluate AI Iconography)
Legible_AI_by_Design.pdf - Accepted Version
Download (1MB)

Abstract

Artificial Intelligence (AI) is becoming increasingly ubiquitous. Implemented into a wide range of everyday applications from social media, shopping, media recommendations and is increasingly making decisions about whether we are eligible for a loan, health insurance and potentially if we are worth interviewing for a job. This proliferation of AI brings many design challenges regarding bias, transparency, fairness, accountability and trust etc. It has been proposed that these challenges can be addressed by considering user agency, negotiability and legibility as defined by Human Data Interaction (HCD). These concepts are independent and interdependent, and it can be argued, by providing solutions towards legibility, we can also address other considerations such as fairness and accountability. In this design research, we address the challenge of legibility and illustrate how design-led research can deliver practical solutions towards legible AI and provide a platform for discourse towards improving user understanding of AI.