Implementing artificial intelligence across task types: constraints of automation and affordances of augmentation
Information Technology & People
ISSN: 0959-3845
Article publication date: 29 May 2024
Issue publication date: 3 December 2024
Abstract
Purpose
This study aims to uncover the constraints of automation and the affordances of augmentation related to implementing artificial intelligence (AI)-powered systems across different task types: mechanical, thinking and feeling.
Design/methodology/approach
Qualitative study involving 45 interviews with various stakeholders in artistic gymnastics, for which AI-powered systems for the judging process are currently developed and tested. Stakeholders include judges, gymnasts, coaches and a technology vendor.
Findings
We identify perceived constraints of automation, such as too much mechanization, preciseness and inability of the system to evaluate artistry or to provide human interaction. Moreover, we find that the complexity and impreciseness of the rules prevent automation. In addition, we identify affordances of augmentation such as speedier, fault-less, more accurate and objective evaluation. Moreover, augmentation affords to provide an explanation, which in turn may decrease the number of decision disputes.
Research limitations/implications
While the unique context of our study is revealing, the generalizability of our specific findings still needs to be established. However, the approach of considering task types is readily applicable in other contexts.
Practical implications
Our research provides useful insights for organizations that consider implementing AI for evaluation in terms of possible constraints, risks and implications of automation for the organizational practices and human agents while suggesting augmented AI-human work as a more beneficial approach in the long term.
Originality/value
Our granular approach provides a novel point of view on AI implementation, as our findings challenge the notion of full automation of mechanical and partial automation of thinking tasks. Therefore, we put forward augmentation as the most viable AI implementation approach. In addition, we developed a rich understanding of the perception of various stakeholders with a similar institutional background, which responds to recent calls in socio-technical research.
Keywords
Acknowledgements
This paper forms part of a special section “Sharing Work with AI: Introduction to the special issue on the futures of work in the age of intelligent machines”, guest edited by Dr. Kevin Crowston, Dr. Ingrid Erickson and Dr. Jeffrey Nickerson.
The authors are thankful for the feedback received when presenting an earlier version of this paper in the “Digitalization in Sport and Personal Health” track at the 30th European Conference on Information Systems (ECIS) and for the financial support provided by HEC Liège Research – PRISME. The authors acknowledge the assistance of Celine Decoster, a former student at Ghent University, in engaging stakeholders in their study. Finally, the authors are grateful to all the informants for taking the time to be interviewed for the study. The first author would like to dedicate this paper to her beloved mother, who passed away during the review process. She was a continuous inspiration and would be very proud of this work.
Citation
Mazurova, E. and Standaert, W. (2024), "Implementing artificial intelligence across task types: constraints of automation and affordances of augmentation", Information Technology & People, Vol. 37 No. 7, pp. 2411-2440. https://doi.org/10.1108/ITP-11-2022-0915
Publisher
:Emerald Publishing Limited
Copyright © 2024, Emerald Publishing Limited