Home // CYBER 2021, The Sixth International Conference on Cyber-Technologies and Cyber-Systems // View article
Authors:
Anne Coull
Keywords: Artificial intelligence; machine learning, explainability, stakeholder; acceptance; transparent; model; post-hoc; human-centred explanation; measure.
Abstract:
Artificial Intelligence (AI) applies algorithms to make decisions or support human decision-making. AI has the ability to transform every industry sector. While AI cannot yet reason abstractly about real-world situations or interact socially, it is responsible for transforming the online consumer industry, facilitating biometric access to mobile phones, and for bringing science-fiction into reality with driverless cars. Explainability is a major barrier to acceptance and utilisation of AI. This is most apparent in more conservative industry sectors, such as banking and finance, health and security, where the penetration of AI is nominal. Engendering greater user acceptance of AI requires an understanding of its stakeholders, who they are, and what they need to understand. Analysis of the current machine learning models identifies three main groups in the context of explainability: Those models that are transparent and easy to understand from their logical processes; Those models that can be adjusted to take a more human-logical approach that explains itself; and those models that are so complex they need to be explained post-hoc by interpreting their behaviour. Human-centred performance measures for explainability will facilitate continuous improvement and corresponding increased acceptability of AI models.
Pages: 106 to 112
Copyright: Copyright (c) IARIA, 2021
Publication date: October 3, 2021
Published in: conference
ISSN: 2519-8599
ISBN: 978-1-61208-893-8
Location: Barcelona, Spain
Dates: from October 3, 2021 to October 7, 2021