Guidance for Algorithmic Bias Considerations

As software using algorithms are becoming increasingly complex, with big data analytics, Al, and machine learning, controlling for algorithmic bias has become a significant challenge. To address this challenge, the IEEE (pronounced “Eye-triple-E,” which stands for the Institute of Electrical and Electronics Engineers), a significant technical professional association for the advancement of technology, recently announced a standard for algorithmic bias considerations. The standard is one of the eleven ethics-related standards currently under development as part of the IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems.

According to a related working paper, the purpose of the new standard for algorithmic bias considerations is to provide software developers who create algorithmic systems with a development framework that they can use to avoid unintended, unjustified, and inappropriately differential outcomes for users. In the working paper, an unjustified bias is referred to as the differential treatment of individuals based on criteria for which no operational justification is given. An inappropriate bias is defined as a bias that is legally or morally unacceptable within the social context where the system is used (Koene et al., 2018).

Within the context of the accounting profession, we expect similar guidance to emerge from and be led by professional organizations such as the American Institute of Certified Public Accountants (A1CPA), the International Federation of Accountants (IFAC), the Institute of Management Accountants (IMA), the Institute of Internal Auditors (IIA), and the Information Systems Audit and Control Association (ISACA).

Professor Lorenzo Patelli of the University of Denver (2019), a member of IMA’s Committee on Ethics, calls for all management accountants to be aware that Al is not a neutral technology. Al poses challenges to their ethical professional practice based on the principles and standards of management accounting and finance. Specifically, he discusses how Al is challenging the ethical tenets of IMA's Statement of Ethical Professional Practice. In honesty, fairness, objectivity, and responsibility, he refers to the potential risks of dishonesty in the way data is analyzed in sophisticated big data analytics. Regarding the risks related to the fairness of the analysis, he stated:

A very large data set still may not be representative of the overall phenomenon for two main reasons. First, the use of current data may not be representative of future trends. Hence, it is unfair to neglect factors that are not heavily represented currently. Second, the most dominant factors may not be the most consequential. Thus, ignoring or undervaluing historically underrepresented factors violates fairness because it may aggravate biases and cause severe unintended consequences such as discrimination, (p. 12)

Regarding objectivity and responsibility, he mentions such potential risks and challenges as the lack of flexibility to changes, inability to explain the outcomes of Al, and difficulty attributing responsibility for the results due to the complexity of Al systems and individuals and organizations struggling to accept responsibility for the consequences.

He concludes that:

Management accountants must formulate ethical principles for the design of Al algorithms, recognize ethical issues implied in the development of Al-powered software, and resolve ethical mishaps caused by the implementation of Al solutions, (p. 12)

Professional organizations have started to review their codes of ethics and standards to identify potential areas where Al ethics and bias might have to be addressed.

In 2018, the International Ethics Standards Board for Accountants (IESBA), an independent global standard-setting board (affiliated with the IFAC), formed a Technology Working Group (TWG). The purpose of the TWG is to analyze the impact of the latest trends and developments in technology on the ethical behavior of professional accountants and lESBA’s existing International Code of Ethics for Professional Accountants. In the initial phase of this initiative, completed in December 2019, the TWG concluded that the Code currently provides high level, principles-based guidance for most technology-related ethics issues. However, it also identified key areas where the material in the Code could be enhanced (IESBA, 2020). Some examples of recommended enhancement areas are summarized below:

• • Highlight a broader societal role for professional accountants in promoting ethical behavior as a critical, consistent foundation for organizations, particularly when developing and using technology.
• • Revise the Code to deal with the threats created by the complexity of the professional environment more effectively in which professional accountants perform their professional activities.
• • Revise the Professional Competence and Due Care subsection by expanding a professional accountant's responsibility to be transparent.
• • Strengthen the concept of accountability outlined in the Code (by including references to technology in provisions relating to relying on the work of others).
• • Revise the Confidentiality subsection considering the increased availability and use of personal and other sensitive data, to consider privacy-related matters and protect information actively.
• • Strengthen the provisions relating to auditor independence (to address potential threats to independence, such as those created by the sale or licensing of technology applications to audit clients and others).

However, this effort from professional organizations remains in an early stage, and more guidance is expected in the near to medium-term future.

In practice, few organizations have an ethical framework in place to address Al and algorithmic bias risks today.

In a recent press release, Irfan Saif, a principal Al leader for the Risk & Financial Advisory practice at Deloitte & Touche LLP, stated:

For many organizations, a steep Al learning curve awaits. Traditional business professionals will need to learn how to team with data scientists and technologists to achieve strategic goals and to explain the changes in the environment. And those developing algorithms and managing data will need to be specially trained to identify and mitigate bias within Al applications. An educated and tech-savvy workforce is better positioned to ethically embrace the opportunities that Al use creates (Deloitte, 2019, para. 7).