Why Creating a Man Cleaning a Toilet Image is Harder Than You Think (Thanks to AI)

Gender Equity in Generative AI

By Samantha Pillay

When I set out to create an image of a man cleaning a toilet, I didn’t expect it to be a challenge, having created countless images previously. After hours of experimenting with different prompts and adjustments, I was left with images that felt off. The man I wanted to portray didn’t seem to know what he was doing, his posture awkward, his expression bemused. It wasn’t just a bad image; it seemed like the AI simply couldn’t grasp what I was asking for.

So, I tried a little experiment. I flipped the prompt and asked for a woman cleaning a toilet. The results were astonishingly easier to generate. The AI-generated woman looked natural, focused, and competent, effortlessly fitting into the scene. This was a subtle but powerful example of the gender biases baked into AI models.

The Root of the Problem: AI Training Data

Generative AI models are trained on massive datasets collected from all over the internet, including images and descriptions. If the training data disproportionately shows women performing domestic tasks, the AI learns to associate these activities more closely with women. It’s not that the AI is deliberately biased—it’s simply reflecting the world as it “sees” it through its training data. This is how stereotypes get embedded and, ultimately, reinforced.

AI models can perpetuate stereotypes if their training data isn’t balanced and inclusive. This experience made me wonder: am I, as a content creator, therefore, aiding to combat these biases? Does the content I share or create have the potential to retrain AI, either directly or indirectly, and move the needle on gender representation?

Can Content Creators Make a Difference?

There are two ways content like mine can impact AI models. First, if my images or videos are used in future training datasets, they could directly influence the AI’s learning. In addition, the content I share online, can indirectly shape public perceptions and influence what others create, adding to the broader cultural narrative. By putting out content that challenges gender norms, I’m introducing new, diverse examples into the digital world. With more content creators actively working to break these stereotypes, we can start nudging AI’s understanding in a more equitable direction.

AI as an Enabler, Not the Enemy

Despite its flaws, AI has been a powerful tool in my creative projects. I recently created a short film titled An Incontinence Truth, using AI-generated images, voices, and music to raise awareness about urinary incontinence—a topic often overlooked. Without AI, this film, and dance-pop soundtrack, would not have been possible. AI provided the freedom to portray the emotional impact of incontinence, ethically, without using any real patients’ words or images. It allowed me to create content quickly and cost-effectively, highlighting the potential for AI to be a force for good when used thoughtfully.

Why We Need to Be Engaged

AI use will only increase, and we need to influence the gender biases embedded in these models. It’s crucial that we generate content that presents a broader, more inclusive view of the world. With active participation, we can help reshape the digital landscape, one image at a time.