Leaflet for Artificial Intelligence Systems (LAIS): a novel artefact for promoting transparency in AI

Abstract

The adoption of AI (Artificial Intelligence) systems by the general population led to many juridical and ethical problems. Researchers have proposed a set of widely accepted ethical principles to guide the development of AI systems. However, it is not clear how to translate the principles into practice. In this dissertation we aim to address the transparency principle and investigate how the lack of this principle may impact on the operation of AI systems, leading to problems such as the unclearness of who is responsible when a negative output is generated. Moreover, we aim to explore the viewpoints of developers as well as users of AI systems, as they often do not understand nor know the system's limitations and implications. We propose a Leaflet for Artificial Intelligence Systems (LAIS), an artefact that aims to support transparency in AI systems. This artefact was crafted inspired by the pharmaceutical leaflets, and it was developed due to the recognition that the current software licenses are lengthy and not intuitive. We adopted the Design Science Research Methodology to guide the development of the artefact. To design the leaflet, we analyzed AI regulations initiatives, ethical AI literature, and pharmaceutical leaflets. To validate our artefact, we conducted a survey with experts from different backgrounds, including developers, users and researchers in the field of AI. The LAIS achieved a 94% approval rating, with participants affirming its relevance as a document that fosters transparency. Therefore, we consider that the LAIS represents an original and innovative contribution which operationalizes transparency principle into a concrete artefact, and it was crafted inspired by a long tradition of the pharmaceutical industry in communicating with end-users.