AI and DEIB: How to push back on algorithmic bias.
As content creators, we should be aware of our biases so they don’t seep into our work. It’s not just good for business; it’s the right thing to do. I was fortunate to work for one company that took this seriously. They trained employees on how to create communications free of offensive, inflammatory, and biased content, and had a sensitivity review process in place to enforce it. Diversity, equity, inclusion, and belonging were baked into our processes.
Sensitivity reviews required me to engage in uncomfortable conversations about race, religion, gender, disability, and culture. If a phrase or image I was reviewing sent up a red flag, I reached out to a colleague from the affected group and we talked it over. Sometimes I discovered that I was sensitive to the wrong issue. Often, the discussions revealed my blind spots. When you belong to the predominant culture, these conversations can make you feel uneasy. This is good.
Generative AI can introduce bias into our work.
Machine-based learning systems are trained on enormous amounts of data, and this data is based on the world as it currently exists. This means that the dominant culture is represented more favorably and in greater proportion than minority groups, which may be considered “outliers” by AI.
Computers learn from their creators, and if we haven’t done the deep societal work of eliminating racism and sexism, our algorithms will reflect that. As Alexandria Ocasio-Cortez puts it: “Algorithms are still made by human beings, and those algorithms are still pegged to basic human assumptions. They’re just automated assumptions. And if you don’t fix the bias, then you are just automating the bias.”
While there have been calls to make AI more diverse, the improvements experts are suggesting—such as hiring more minority and women computer engineers—could take years to implement. For now, we must rely on our human brains and hearts to push back against algorithmic oppression. But how?
Review your content for anything unintentionally upsetting/insensitive.
When I entered the prompt, “write 20 headlines that highlight the inequity of Black children in the education system” into ChatGPT, this I what I got:
The headlines are lackluster, but that’s not the real issue. In the context of an article on Black children, the idiom “breaking chains” in headline 15 needlessly evokes upsetting images of slavery. This would be OK, even necessary, if the article were about slavery. But it’s not an article about slavery. So don’t use it.
Is that too subtle? Am I splitting hairs? Maybe, but since my ancestors weren’t slaves, I wouldn’t have the same visceral reaction to the idiom as someone whose ancestors were. And if the headline is easy enough to throw away, why not do just that?
Don’t be lazy.
Here’s another prompt I entered: “Write a 100-word corporate statement on Black Lives Matter.” In this case, the output wasn’t bad.
But here’s where it can get ugly: Did your organization have a conversation about Black Lives Matter at the highest levels, and is it committed to fulfilling this vision? If not, the statement is simply AI-generated bullshit. And it’s easily detectable AI-generated bullshit. Simply cut and paste this “corporate statement” into ZeroGPT.com to see. Here’s what we get.
You don’t want something as important as a corporate statement on the Black Lives Matter movement to be flagged as auto generated. This is really, really, really bad for your brand. It’s proof of your laziness, your lack of empathy, your tepid commitment to diversity, equity, inclusion, and belonging, and your attempt to save time instead of crafting a careful, considered message. I would even call it monstrous.
And if you think your competition or employees aren’t checking you on this, you’re wrong.
Make sure your images don’t reflect stereotypes.
Whether it’s an algorithm that turns a pixelated image of Barack Obama into a Caucasian or a program that thinks all surgeons are white males, AI-generated images still have a way to go. Proceed with caution.
What else can we do to advance Diversity, Equity, Inclusion, and Belonging in an AI world?
As creatives and marketers, we don’t control who writes the algorithms or trains the computers, but we can:
- Adopt a diversity-first mindset: from the initial creative brief to persona creation to UX testing.
- Make sure our staff is diverse and has a seat at the table.
- Put a formal sensitivity review process in place.
- Train employees to write AI prompts that take diversity into account.
- Listen to our colleagues! If they flag something as biased or stereotypical, pay attention and remediate the issue.
- Do the soul work: We need to get to know our biases and root them out. We need to uncover our blind spots. We need to be better.
Are you having and welcoming these hard conversations? Does your organization have processes in place to weed out bias? Drop a comment below!
