A human-centred framework for the development and evaluation of online systems applying artificial intelligence

Artificial intelligence offers numerous benefits, but can also cause harm when not adopted through a human-centred approach and addressing concerns regarding artificial intelligence (AI). This approach is known as human-centred AI (HCAI) and aims to augment human capabilities to create safe, reliable, and trustworthy solutions. The primary goal of HCAI is to enable a significant degree of human control and automation, allowing humans to influence decisions and events. Although definitions exist for HCAI, a consensus is required on how HCAI solutions are designed and evaluated. An HCAI framework needs to be employed for the development of AI solutions to ensure an ethical solution, where the results are objective, credible, unbiased, and fair. The use of a framework will facilitate uniformity in strategies, principles, and goals in organisations. This study employed pragmatism to understand what factors constituted HCAI, the existing frameworks for HCAI, how organisations apply human-centred design, and how organisations can effectively implement a HCAI framework. The knowledge gained helped develop a framework that can be applied to achieve human-centred design for AI solutions through a DSR approach. The DSR cycles of awareness, suggestion, development, evaluation, and conclusion were followed, utilising systematic literature review, questionnaires, and semi-structured interviews as data collection methods. The data were analysed using thematic analysis and descriptive statistics, including mean, mode, median, and standard deviation. The resulting framework consists of two sections: development and evaluation. The comprehensive framework comprises 15 categories with humans as the focus and humancentredness as part of the organisational strategy. The organisational strategy includes accountability measures to strengthen the human-centredness of the organisation. Context and requirements gathering are required, incorporating HCAI consideration, before design and development are approached. The approach of design, data, model, and analysis is suggested with testing, prototyping, and communication as an iterative loop. Through the design, data, model, and analysis approach, the following categories of grouped components need to be considered: human-centred values and ethics, UX and human interaction, data and model governance, technical robustness and performance, and AI system capabilities and design. The operationalisation of the framework guides the design, development, and evaluation of AI solutions to ensure human-centred AI solutions that are reliable, safe, and trustworthy.

Grantor dc:publisher

University of Pretoria

Year dc:date.issued

2026

Advisor dc:contributor.advisor

• Smuts, Hanlie

• UCTD
• Sustainable Development Goals (SDGs)
• Artificial intelligence
• Human-centred
• Human-centred artificial intelligence
• Human-centred framework

Statement dc:rights

• © 2024 University of Pretoria. All rights reserved. The copyright in this work vests in the University of Pretoria. No part of this work may be reproduced or transmitted in any form or by any means, without the prior written permission of the University of Pretoria.

Language dc:language.iso

en

DOI

https://doi.org/10.25403/UPresearchdata.31347985; https://doi.org/10.25403/UPresearchdata.31348018

Dc Identifier Other

A2026

OAI identifier oai:identifier

oai:repository.up.ac.za:2263/108299