Ethical Implications of AI Bias for Engineers

The past decade has been marked by startling advances in the capabilities of artificial intelligence systems. However, data biases, an inevitable feature of human-coded algorithms, also emerge as a daunting challenge that must be confronted by the current society. They fuel political polarization, perpetuate the cycle of discrimination, and are often developed to prioritize the profits of the company over the welfare of its customers.

For example, Harvard Business Review has shared how a simple google image search for “professional haircuts'’ and “unprofessional haircuts'’ can reveal the problem. The former brings results that are almost exclusively white males, while the latter turns up faces with diverse racial, ethnic, and gender backgrounds. A child who performed this search could mistakenly believe that he/she can never receive a professional haircut solely due to his/her skin color. The result appears especially alarming when considering that stylish and business-appropriate haircuts are supposedly objective topics.

Since Google presents results based on articles written about such subjects, the outcome of this simple experiment clearly emanates from human editorial decisions to highlight the narratives of white men while neglecting those of others. This predicament, of course, aligns with the reality that the overwhelming majority of media outlets are owned by white males, who choose to propagate specific news stories and present them in such a way that won’t challenge the worldview of their predominantly white peers.

Although the downsides of biased data sets and algorithms have been clearly characterized, comprehending their underlying causes becomes more challenging. Haugen, a Facebook whistleblower who has previously worked as a product manager, criticizes the interest-prioritizing algorithms used on platforms managed by Facebook. She declares that “no one at Facebook is malevolent, but the incentives are misaligned.” Only through alluring the customers into consuming more content can Facebook maximize its revenues from advertisement, and unfortunately, the most efficient way of accomplishing this goal is bombarding their clients with provocative content.

Mitigating systemic AI bias requires the algorithm to be transparent, accountable, and explainable to the general public. Despite appearing unrealistic and overwhelming, there are a few practical steps that all tech companies could undertake. For instance, companies can assemble their machine learning and artificial intelligence teams from a more diverse group of applicants. By gathering individuals with distinct experiences and from different age, ethnic, and gender groups, the team can clearly construct a more representative model.

In addition, all companies should advertise platforms through which the users can give feedback. By having the end-users directly interact with the engineers behind the screen, the company can be more responsive to the needs of its clients.

Issue cannot be overcome by the computer scientists alone. Endeavor to create a mutually accountable and supportive community of experts from various disciplines is also demanded. For example, clinicians must be present while addressing AI bias in healthcare systems, while experts in the criminal justice system can also contribute to improving the performance of AI in assessing the offender’s risks or needs and predicting his/her rate of recidivism.

Disentangling the myths behind systematic AI prejudice would be challenging but also equally rewarding. For example, banks’ biased algorithms to predict credit scores have historically disadvantaged low-income and ethnical minorities and locked them out from the growing financial opportunities. Recently, however, many credit agencies have begun to challenge this stereotype. One such company with a vision to promote economic equality, the credit bureau Experian, has successfully developed the “Boost program,” which provides financial opportunities to those who have been known as the “thin-file” clients. Understandably, with meager economic resources and limited loaning history to augment their credit scores, those marginalized communities would have been automatically placed in the category of the least priority by algorithms. With this opportunity to finally prove their reliability in the banking system, however, the results were staggering. Since the launching of this program in 2019, more than sixty-one percent of the four million participants with “poor” credit have earned their upgrade to a “fair” rating, giving them far more access to buying a car, renting an apartment, or acquiring a business loan. The lenders who took the initiative to trust these thin-file customers also received the benefit of extending far more credit to other profitable businesses.

As stated in the National Society of Professional Engineers Code of Ethics, engineers shall all be guided by the “highest standards of honesty and integrity,” “acknowledge errors,” and “not distort or alter the facts.” These abstract words can have shifting meanings over time, but I believe that at its core, the ethical code asks for self-interrogation––are we really engineering towards a better human future? I suppose that now is the paramount time for the software engineers and computer scientists to reconsider this question. Without the capability to fully explain how their codes benefit both the wealthy and the poor, the software engineers must first acknowledge the potential flaws in their algorithms before striving to make their codes more accessible to the general public. Our collective empathy, consciousness, and morality as human beings must not be waived in the struggle for economic growth and productive efficiency.

References

[1] Greg Satell and Yassmin Abdel-Magied, “AI Fairness Isn’t Just an Ethical Issue”, Harvard Business Review, October 20, 2020. Accessed electronically on Oct 31,2021, https://hbr.org/2020/10/ai-fairness-isnt-just-an-ethical-issue

[2] National Society of Professional Engineers Code of Ethics https://www.nspe.org/resources/ethics/code-ethics

[3] Rachel Metz, “Facebook’s success is built on algorithms. Can they also fix it?” CNN Business, Oct 10, 2021. https://www.cnn.com/2021/10/10/tech/facebook-whistleblower-algorithms-fix/index.html. Accessed electronically on Oct 31,2021. Interview: https://www.youtube.com/watch?v=_Lx5VmAdZSI