SKIP

As we approach the two-year milestone of ChatGPT’s launch and the subsequent “Cambrian explosion” of functions and instruments, it becomes increasingly clear that two things stand out starkly: The potential for this technology to revolutionarily reshape our lives is straightforward, as are the perils of pervasive bias that pervade these models.

Within just two years, AI has rapidly evolved from providing routine assistance with tasks such as recommending online purchases, to making autonomous decisions on life-altering actions like determining insurance coverage, approving housing applications, assessing credit scores, and evaluating welfare claims. While some might initially find biases in these fashion styles amusing or even endearing, it is crucial to recognize that such prejudices become intolerable when they wield significant influence over the industries that shape our daily lives and economic well-being. 

How can we effectively address AI bias and develop more responsible fashion trends when the data used to inform them is fundamentally flawed? Can sustainable fashion truly prevail when those driving the trends are oblivious to the subtle yet far-reaching consequences of their designs, encompassing unconsidered drawbacks in all their multifaceted forms?

Can we increase diversity by recruiting more women, individuals from underrepresented minority groups, older professionals, and individuals with a broader range of AI expertise?

Early training and publicity

As I’ve navigated the world of STEM over more than three decades, I’ve consistently found myself in the minority, yet I’ve never considered it a reason to engage in radical or divisive discourse about AI’s potential range. Despite the remarkable advancements in housing innovation, a similar level of progress cannot be claimed for our workforce’s scope, particularly in data and analytics. 

Although the reported girls account for less than a third (29%) of all STEM employees, they comprise nearly half (49%) of total employment in non-STEM fields, a stark contrast that warrants further examination? In response to the U.S. According to the Division of Labor Statistics, a mere 8.7% of black professionals are employed in math and computer science roles, a stark reminder of the persistent underrepresentation of this demographic in these fields. Despite stagnant progress over two decades, women’s representation in leadership remains alarmingly low, with an unimpressive 12% occupying executive-level positions once entry-level roles are factored out.

To increase STEM engagement among girls and minorities, it’s essential we implement comprehensive strategies from an early age, starting in elementary school classrooms. I recall the heartwarming video released by toy manufacturer Mattel showcasing elementary school students, likely first- or second-graders, receiving desks filled with an assortment of toys for their enjoyment. While the majority of women opted for traditional “feminine” toys, such as dolls or figurines depicting ballerinas, they showed little interest in alternative playthings, including toy racing cars, which were typically associated with their male peers. The women were subsequently proven a video of their historic achievement: becoming the first lady to win the Argentinian Touring Car Grand Prix, thereby revolutionizing the ladies’ perspective on what was possible. 

This lesson underscores the profound impact of illustrations on shaping notions, serving as a poignant reminder to prioritize intentionally crafted messaging for young women in STEM fields. We must ensure equal access to exploratory opportunities and promotional platforms by integrating them into a shared curriculum and partnering with non-profit organizations like Information Science for All or the Mark Cuban Foundation. Let’s amplify and celebrate the trailblazing efforts of women like CEO AMD Lisa Su, OpenAI CTO Mira Murati, and AI researcher Joy Buolamwini, who founded The Algorithmic Justice League – paving the way for young girls to see themselves in STEM fields, where they can take the wheel. 

As technology advances, information and artificial intelligence will become integral components in nearly every profession, extending from athletes to astronauts, fashion designers to filmmakers, and countless others in between. We must address the injustices that hinder minority access to STEM education, and empower women by showcasing how STEM training can be a gateway to a fulfilling career. 

To effectively counteract bias, one must initially acknowledge its presence.

Biases can infect AI in two distinct ways: through the large datasets used to train algorithms, and through the personal biases or judgments of the individuals who design and implement them. To effectively eliminate bias, it’s crucial to initially recognize and accept its presence, assuming that all information is inherently prejudiced and that individuals’ subconscious biases are actively at play? 

Explore a range of cutting-edge image generators, including the most popular ones such as MidJourney, DALL-E, and Stable Diffusion. Reporters were asked to illustrate a “lovely lady,” but the results revealed a startling dearth of representation in body types, cultural backgrounds, and skin tones. In stark contrast, the notion of female magnificence, as dictated by these societal metrics, overwhelmingly favored the young and European, resulting in a stereotypical ideal of slender, pale beauty.

Only approximately 2% of the images exhibited signs of aging, while a mere 9% displayed darkened skin tones. Despite this finding, nonetheless, “bias originates”, The Submit’s evaluation revealed that widespread picture instruments struggle to depict realistic images of women outside the Western world? Additionally, research has shown that linguistic differences can lead to “covert bias” in determining a person’s intentions or recommending capital punishment.

What if refinement of bias is a mere illusion? In the late 1980s, I started my career as an enterprise systems specialist in Zurich, Switzerland. As a married woman, I was initially precluded from opening my own checking account, despite being the primary breadwinner in my household. In certain regions, a mannequin proficient in vast archives of women’s historical credit data may struggle to find any substantial records at all? What nuances do recruitment teams and hiring managers account for when evaluating candidates whose work history is punctuated by extended periods of maternal leave or childcare responsibilities, often spanning months or even years? While artificial intelligence-driven information might provide a solution to address this issue, it is contingent upon model developers having the foresight to consider these concerns.

It’s essential that diverse examples of women not only occupy seats on the AI committee, but also have an active voice in shaping, training, and governing these models. As technological advancements unfold without deliberate direction, they risk being shaped by the values and biases of a privileged few, leaving the greater global community vulnerable to the consequences of unchecked innovation.  

Extra range: A no brainer

As humanity’s insatiable pursuit of financial gain continues to fuel an accelerating pace, the entrenched biases embedded within our digital archives and personal histories pose a formidable hurdle to achieving complete eradication from AI development. However, this cannot suggest inaction or ignorance is acceptable. As STEM fields continue to expand their scope, it’s essential that AI models benefit from this growth, ultimately yielding more accurate and inclusive representations that positively impact us all.