Proceed with Intent: AI for the DEI Landscape
The application of artificial intelligence (AI) was abundant at this year’s International Workshop. While AI continues to prove to be an extraordinary and accessible tool, one session aimed to explore the utility of the technology in the mission of improving diversity, equity, and inclusion in the pursuit of a better world via systems approaches. AI has revealed opportunities to reduce the time and resources required to analyze large datasets, uncover patterns, and tailor solutions that make information more accessible to people, including those with disabilities and otherwise marginalized communities.
The first act of the newly named INCOSE Assistant Director of Diversity, Equity, and Inclusion (DEI) was to co-lead a reflective workshop with EWLSE founder Dr. Alice Squires. “Empowering Women Interactive Workshop on Persistence in Leadership” encouraged attendees from around the world to self-identify at various levels of leadership proficiency and openly participate in exercises to actively reflect and share their personal and professional experiences. DEI-EWLSE is analyzing participants’ written reflections on enabling and inhibiting factors affecting the decision to persist in systems leadership. More information will be available soon!
Yet, the more immediate observation was the protrusion of bias in the imaging output generated by AI when asked to depict diversity, equity, and inclusion of strong women leaders in engineering systems. Commonly used AI packages like DALL-E, Playground, Leonardo.AI, and 123RF were challenged to generate representative photos and art. Despite the usage of carefully crafted combinations of keywords to require cliched characteristics of various groups in each generation, the task of capturing the group’s sense of DEI representation using AI tools proved quite difficult.
Participants expressed several forms of discontentment with the images uncovered:
“Why doesn’t she look anything like me?”
“Why is she dressed in such a provocative way [ex. heels, slits, visible cleavage]?”
“Is there a reason why a powerful female leader is an anatomically ‘perfect,’ white, young lady with long, flowy hair?
“Did we input the wrong prompt?”
“How many tries did it take to get to that?”
Hmmm.
Can AI be biased in its understanding of diversity or women in leadership roles?
Short answer: Yes.
Human understanding of the scope of diversity, facets of inequity, and the needs for inclusivity are still evolving. As a society, we must be intentional in acknowledging where gaps exist to craft a path forward.
Recommended by LinkedIn
How do we enhance AI’s ability to depict DEI?
Recommendation. We must ensure a diversity of perspectives, experiences, and thoughts among AI systems development, deployment, and sustainment teams.
How do we mitigate inherent bias that perpetuates and amplifies existing inequities?
Recommendation. More careful data selection and analysis are needed to mitigate bias in AI algorithms, data sets, and training.
How do we apply human oversight to improve the output of AI?
Recommendation. We must build and leverage our own understanding of DEI to intervene effectively and foster the development of informed prompts that impose certain features and dynamics in its rendering.
Using very lengthy and specific prompts, we were able to generate images with higher levels of gender, age, and race/ethnicity variance, but optimal representations of DEI in systems leadership should capture so much more.
Our charge to the INCOSE community is to address these challenges and embrace best practices to ensure that AI becomes a force for good not only in technical areas but also in promoting a more diverse, equitable, and inclusive future overall.
To read this article and more, check out the INCOSE Q1 Members Newsletter.