Abstract
Despite artificial intelligence (AI) being one of the fastest-growing fields due to its ability to enhance competitive advantage, there are concerns about the inherent fairness of AI algorithms. Bias has been noted in AI tools used in criminal justice, recruitment, health, and social service administration. As the use of AI increases there has been a call for organisations to consider the ethical and societal implications of biased AI. The traditional emphasis on fairness in AI algorithms tends to focus on developing fair standards, even though the field of AI is known for its rapid advancements and evolution. This means that the AI fairness standards and policies, which are usually rigid and prescriptive, could also rapidly become obsolete.
This study sought to understand how a fair process model can help reduce bias in AI algorithms. Specifically, the study adapted Jürgen Habermas’ critical theory of communicative action, and the lifeworld as an ideal mechanism to develop a process model for AI algorithmic fairness. The model engaged logical-semantic, procedural, and performative rules that can be applied to avoid power imbalances and domination by any entity or individual during the development process of AI. Jürgen Habermas idealised a better world where social changes occur within society in an evolutionary process free of domination and allowing for public discourse.
The study thus used a critical realism philosophy to investigate how AI management, developers, and society can contribute to the development of fair AI processes. The key findings revealed that due to the limited gender and social diversity in AI teams, there was little accountability to ensure fair AI, and limited avenues for society to participate in reviewing and critiquing AI models for potential bias. The study identified two predominant mindsets with regards to society’s participation in ensuring ethical and unbiased AI models: a dependent mindset, where individuals believe external entities are responsible for ensuring fairness in AI, and an independent mindset, where individuals believe in taking personal initiative on AI fairness.
Practical guidelines, including accountability measures such as data reviews, approvals, ethical clearance, fairness tests, and explainability and transparency are integral to the
v
logical-semantic, procedural and performative rules of a fair process model for developing fair AI. There is a need for a platform that will enable various AI stakeholders, including members of society who may be impacted by these systems, to engage, review and debate the development of fair AI. Responsible organisations must take deliberate actions to ensure that their AI developers adhere to fair AI processes. Additionally, while society needs to take an active role in promoting unbiased AI, the government needs to establish policies and mechanisms to ensure the interests of all social groups are considered.
The process model proposed in this study contributes to the advancement of fair AI development. The study provides practical guidelines for AI managers and developers to work with all relevant stakeholders, including society. The proposed model may impact the AI development process to allow for bias to be detected and mitigated before deployment. Ultimately the study seeks to promote the overall development of fair AI models.
Keywords:
Artificial Intelligence (AI), Algorithm, AI Ethics, Fairness, Jürgen Habermas, Theory of Communicative Action, Critical realism