The Potential and Risks of AI Bias: A Call for Diversity in AI Talent
As we approach the two-year anniversary of ChatGPT and the subsequent “Cambrian explosion” of generative AI applications and tools, it is clear that AI has the potential to reshape our lives positively. However, it is also essential to address the risks of pervasive bias that exist within these models.
AI has rapidly evolved from supporting everyday tasks like hailing rideshares to becoming the judge and jury in significant activities such as arbitrating insurance, housing, credit, and welfare claims. While bias in these models may have been seen as annoying or humorous in the past, it is now indefensible when these models hold the power to influence our livelihoods.
The question arises: How can we proactively mitigate AI bias and create less harmful models when the data we train them on is inherently biased? It seems challenging when those who create the models lack awareness of bias and its unintended consequences.
The answer lies in promoting diversity in AI talent. Currently, the workforce in STEM fields lacks diversity, particularly across data and analytics. Women make up less than a third (29%) of all STEM workers, despite representing nearly half (49%) of total employment in non-STEM careers. Black professionals in math and computer science account for only 9%. These statistics have remained stagnant for 20 years, with even lower representation in leadership positions.
To address this issue, comprehensive strategies are needed to make STEM more attractive to women and minorities. This effort should begin in the classroom as early as elementary school, ensuring equal paths for exploration and exposure to STEM subjects. Non-profit organizations like Data Science for All or the Mark Cuban Foundation’s AI bootcamps play a crucial role in providing opportunities for underrepresented groups. Moreover, it is vital to celebrate and amplify the achievements of women role models in STEM to inspire young girls.
Data and AI are becoming essential in almost every job of the future. Therefore, it is crucial to close the inequities that limit access to STEM education for minorities and show girls that STEM education can lead to careers in any field.
To mitigate bias in AI, it is essential to first recognize its existence. Bias can infiltrate AI through the biased data sets models are trained on and the personal logic or judgments of the people constructing them. Several popular image generators have shown a lack of representation in body types, cultural features, and skin tones, reinforcing Western beauty standards. Even subtle biases, such as those resulting from historical credit data or gaps in employment due to maternity leave, can have a significant impact. Synthetic data generated by AI can help address some of these discrepancies, but only if model builders and data professionals are aware of these problems.
Diverse representation of women must have an active voice in constructing, training, and overseeing AI models. This cannot be left to chance or a few technologists who historically represent only a fraction of the global population.
While it may be challenging to completely eliminate bias from AI, taking action to promote diversity in STEM and ensure a diverse range of talent is involved in the AI process will undoubtedly lead to more accurate and inclusive models. This benefits everyone and creates a more equitable and fair AI-powered future.
In conclusion, the potential of AI to positively reshape our lives is undeniable, but we must also address the risks of bias within these models. Promoting diversity in AI talent is crucial to mitigate bias and create more inclusive models. By starting early with education and exposure, celebrating women role models in STEM, and recognizing bias in data and model construction, we can work towards a more equitable and fair AI future.