The Importance of Inclusive Approaches to Artificial Intelligence

Every day during my senior year of high school, I walked through the main building’s doors and was met with a temperature scanner that would ideally prevent people with high body temperatures from spreading COVID-19. It seemed innocuous enough: it simply mapped your face, using target areas like the inner corner of your eye, and scanned your temperature using infrared rays. One thing I noticed, however, was that the scanners always seemed to falter when attempting to scan darker-skinned students such as myself. There were countless instances where I would watch students in front of me pass by with ease, only to be prevented myself. Eventually, the school nurse had to manually scan my temperature using the infrared guns; every time, she did it with pure kindness, but I knew it was a hassle for both of us. I began to think of reasons as to why the issue existed in the first place. I knew it wasn’t because I was wearing a mask–the scanner was primed to work exclusively while masks were on its target’s face. I also knew that it wasn’t because of the type of mask I wore, either: people wore countless different kinds of masks in a myriad of different colors every day. I decided to delve into the inner workings of those machines and found the likely source of my issue: biased facial recognition software.

Errors in facial recognition software are the result of their training models heavily reflecting the people that develop them: primarily white men. In one report regarding an algorithm that caused the accidental arrest of a Black man, it was found that Black and Asian faces were falsely identified by the algorithm a shocking 10 to 100 times more than white faces. This is in direct violation of tenets II and I.i of the Institute of Electrical and Electronics Engineers (IEEE) code of ethics, by which engineers are expected “to treat all persons fairly and with respect...and to avoid injuring others” as well as “to hold paramount the safety, health, and welfare of the public [...] and to disclose promptly factors that might endanger the public or the environment.”

Over the past year, the imbalance in tech has only been compounded further, as seen in the case of Google. In December of 2020, the seemingly omnipresent tech giant fired the head of its artificial intelligence ethics team, notable AI researcher Timnit Gebru, as a result of a dispute involving a paper in which she criticized its facial recognition software. Google certainly has quite a bit of history with bigotry. For example, searches in the engine have displayed horrifically oversexualized results when searching “black girls” due to algorithms loaded with bias, poor regulation of search query data collection has guided people down deep rabbit holes of bigotry and hatred, and its employees have become fearful of reporting experiences concerning sexual harassment and discrimination. Gebru’s firing (which Google likes to call ‘resignation’) only fanned the flames of the already burning workplace culture at the company. The company’s heavy-handed response to valid criticism of their practices regarding artificial intelligence further proved just how little we know about the dangers that it possesses. But why is that so?

As a result of human nature, the ethics of AI are very complicated: trying to impose any specific group’s set of morals onto an AI could have unintended consequences on others. But these consequences are exactly why large tech companies need researchers like Gebru and her teammates: people who are willing to traverse the treacherous terrain that is artificial intelligence are the ones who allow us to map that terrain and potentially come out with a better understanding of how to apply it to the major problems that our world is mired in. Although we may be fearful of what lies in the darkness that is the future of tech, we also must be careful not to extinguish the bright lights that are meant to guide our path; Google’s firing of Gebru and the effective dissolution of their AI ethics team most certainly was a detriment to the field as a whole and is definitive proof that we as engineers must carefully re-examine our approaches towards emerging technologies and those wielding them.

References

[1] Ghaffary, Shirin. “Dozens of Google Employees Say They Were Retaliated against for Reporting Harassment.” Vox. September 09, 2019. Accessed October 26, 2021. https://www.vox.com/recode/2019/9/9/20853647/google-employee-retaliation-harassment-me-too-exclusive.

[2] Hao, Karen. ““I Started Crying”: Inside Timnit Gebru’s Last Days at Google-and What Happens next." January 06, 2021. Accessed October 26, 2021. https://www.technologyreview.com/2020/12/16/1014634/google-ai-ethics-lead-timnit-gebru-tells-story/.

[3] “IEEE Code of Ethics.” IEEE. Accessed October 26, 2021. https://www.ieee.org/about/corporate/governance/p7-8.html.

[4] Noble, Safiya. “Google’s Algorithm: History of Racism Against Black Women.” March 26, 2018. Accessed October 30, 2021. https://time.com/5209144/google-search-engine-algorithm-bias-racism/.

[5] The Regulatory Review. “Facing Bias in Facial Recognition Technology.” The Regulatory Review. August 06, 2021. Accessed November 01, 2021. https://www.theregreview.org/2021/03/20/saturday-seminar-facing-bias-in-facial-recognition-technology/

[6] Young, R., & Hagan, A. (2021, September 30). Search engines like Google are powered by racist8, misogynist algorithms, says MacArthur Fellow. Search Engines Like Google Are Powered By Racist, Misogynist Algorithms, Says MacArthur Fellow | Here & Now. Retrieved February 26, 2022, from https://www.wbur.org/hereandnow/2021/09/30/safiya-noble-internet-research