Algorithmic solutions, subjectivity and decision errors: a study of AI accountability
Digital Policy, Regulation and Governance
ISSN: 2398-5038
Article publication date: 26 November 2024
Abstract
Purpose
The purpose of this paper is to explore the challenges of implementing accountable artificial intelligence (AI) systems in India, focusing on the need for algorithms to justify their decisions, especially in subjective and complex scenarios. By analyzing various government projects, documented biases and conducting empirical case studies and experiments, the study highlights the limitations of AI in recognizing the nuances of India’s unique social landscape. It aims to underscore the importance of integrating political philosophy to ensure that AI systems are held accountable within India’s sociopolitical context, urging policymakers to develop frameworks for responsible AI decision-making.
Design/methodology/approach
The research adopts a mixed-methods approach to address the five research questions. It begins with an extensive literature review, focusing on AI’s transformative potential, algorithmic bias and accountability in the Indian context. Data is collected from 15 AI use cases in health care, education and public safety, 13 government automated decision tools and five bias cases, including facial recognition and caste-based discrimination. Additionally, ten case studies and three experiments on ChatGPT are analyzed. Content analysis is used to interpret and categorize the data, identifying patterns and themes. Specific case studies and experiments on autocompletion in search engines further support the findings.
Findings
The study revealed significant limitations in current AI systems when applied to India’s complex socio-cultural landscape. Analyzing 15 AI applications and 13 government projects, the research identified multiple instances of algorithmic bias. Experiments with Google’s autocomplete and ChatGPT showed that these systems often reinforce social stereotypes and struggle with nuanced, subjective situations. The findings emphasize the accountability gap in AI-driven decisions, highlighting the need for rigorous oversight, particularly in welfare projects where errors could lead to severe consequences. The study recommends developing regulatory frameworks, improving AI design and raising public awareness to address these challenges.
Originality/value
In the context of complex societies like India, a pressing concern arises: who should assume responsibility for the repercussions stemming from algorithmic failures to comprehend subjective complexities? To this end, there exist no serious scholarly works toward which present paper tries to shed new insights. It draws upon insights from the corpus of political philosophy literature, encompassing both classical and contemporary notions of responsibility, and seeks to establish connections between these concepts and the unique sociopolitical structure of India. The work is unique in the focus of the paper and is original in the direction projected.
Keywords
Citation
P.R., B. and O., G. (2024), "Algorithmic solutions, subjectivity and decision errors: a study of AI accountability", Digital Policy, Regulation and Governance, Vol. ahead-of-print No. ahead-of-print. https://doi.org/10.1108/DPRG-05-2024-0090
Publisher
:Emerald Publishing Limited
Copyright © 2024, Emerald Publishing Limited