One of the best parts of my job is being an outward-facing technologist, representing an amazing team of coworkers at CDW. Part of that role is writing this blog post, but it also includes engaging in live events.
Last week, I was in Portland for an executive dinner with colleagues, partners, and customers to discuss security in the context of AI.
This week, I hosted a live “Ask Us Anything” on AI with Caitlin Berger, one of our certified AI badasses. If you missed it, here’s a link to the on-demand recording.
Interestingly, one topic emerged in both events: bias in GenAI models. This was a hot-button issue a few years ago, but it seemed to fade into the background.
I’m glad to see it resurfacing, because it’s more important than ever. Let’s talk about why that is, and what responsible technologists can do about it.
The “Nature and Nurture” of AI Models
During our webinar, Caitlin shared an experience from a real-world project that involved using an LLM to process résumés. The model’s output would directly influence hiring decisions. Hiring is fraught with potential bias, and any organization that hasn’t buried its head in the sand has likely worked hard to minimize it.
But are LLMs capable of avoiding bias themselves?
In response to that question, I shared an analogy.
Like humans, large language models are shaped by both nature and nurture.
As a parent, I accept that my kids are growing up in a world I don’t fully control—a world that includes joy and kindness, but also injustice and racism. My job as a parent is to help them make sense of that complexity. That’s the nurture that shapes who they become.
Similarly, LLMs are shaped by the massive datasets on which they’re trained. That data reflects our world—flaws and all. This is the nature of a foundation model.
As technologists, we must acknowledge that bias may be inherent in these systems, even with the best intentions of their creators.
We must be responsible stewards of this transformative technology, fully aware of its limitations and unwavering in our commitment to using it ethically.
