In my last blog post, “Behind the Scenes: How We Get Downsview SDA’s Divine Services Online,” I shared a glimpse into our process of bringing our church services to YouTube. I talked about the steps involved, from gathering information to crafting descriptions with the help of AI tools like Gemini.
However, what didn’t make it into that post was a rather unexpected and, frankly, frustrating challenge: finding a suitable featured image that authentically represented our diverse communications team at Downsview Seventh-day Adventist Church in Toronto.
You see, our reality is that our church community, and indeed our dedicated team working on these online initiatives, is richly diverse, with a significant representation of Black individuals. It’s something we value and want our online presence to reflect.
So, when it came to choosing a featured image for a blog post highlighting our tech-savvy efforts, I naturally wanted an image that included Black people engaging with technology in a collaborative setting. This shouldn’t have been a difficult task, right?
Well, it turned out to be surprisingly so.
Initially, I tried using general prompts for AI image generation, hoping the algorithms would naturally produce a diverse group. However, the results consistently featured predominantly white individuals. This wasn’t reflective of our team or our Toronto community, and it certainly didn’t align with my intention of subtly showcasing Black engagement in the tech space – something that is sadly still underrepresented in broader media.
I then attempted to be more specific, using terms like “diverse team working with video equipment” or “church media team collaborating on laptops.” Yet, even with these broader diversity cues, the generated images often lacked Black representation or featured individuals in ways that felt tokenistic rather than authentic.
This iterative process became quite disheartening. Each time, I hoped to see a reflection of our team’s reality, but the AI seemed stuck in a loop of producing images that didn’t align with our church’s demographic. It highlighted a frustrating reality about the biases that can inadvertently be embedded in AI training data – often underrepresenting certain ethnic groups in various professional or collaborative settings.
It made me acutely aware of how crucial it is to be mindful of representation, not just in the content we create, but also in the visual elements we use. The difficulty in generating an image that simply reflected our team’s actual composition underscored the ongoing need for greater diversity and more inclusive datasets in AI development.
Ultimately, I had to make a choice. While I strive for authentic representation, repeatedly prompting for specific ethnicities can also be problematic for AI models, potentially leading to stereotyping. In the end, I opted for an image that focused on the equipment and the setting, hoping the blog post’s content would still convey the dedication and diverse perspectives behind our online ministry.
This experience has been a valuable learning curve. It has reinforced the importance of:
Being intentional about representation: Actively seeking visuals that reflect our community.
Understanding the limitations of AI: Recognizing that AI models are not yet perfect and can reflect societal biases.
Continuing the conversation: Raising awareness about the need for better representation in media and technology.
While the journey to find that “perfect” image was more challenging than anticipated, my commitment to showcasing the vibrant diversity of Downsview SDA, both in our congregation and our dedicated teams, remains unwavering. We will continue to strive for authentic representation in all that we do.