How Generative AI Could Perpetuate Fashion Bias

In March 2021, before the general enthusiasm around generative artificial intelligence exploded, a pair of researchers published an article on how biases can show up in AI-created images.

Using an AI tool, they created five male-looking faces and five female-looking faces. Then they fed them into another AI tool to complement with the bodies. For female faces, 52.5% of images returned by the AI ​​featured a bikini or low-cut top, they wrote. For male faces, 42.5% were completed with suits or other career-specific clothing.

Bias in the AI ​​or rather in the data on which these models are trained is a well-known problem. There’s even a mantra: trash in, trash out. The idea is that if you enter wrong data, the output will reflect those faults. Since the generative AI tools available have typically been trained on giant volumes of data mined from the internet, they are likely to reflect internet biases, which can include all conscious and unconscious biases in society. The researchers speculated that their production resulted from the sexualized portrayal of people, especially women, in internet images.

Fashion should be very careful. As it begins to use generative AI for everything from producing campaign images to powering online shopping assistants, it risks repeating discrimination based on race, age, body type and disability he’s spent the past few years loudly proclaiming he wants to move beyond.

For example, when I entered the prompt model in a black sweater into DreamStudio, a commercial interface for the AI ​​Stable Diffusion image generator, the results were thin, white models. This was the case for most, if not all, of the models every time I tried it. In the hive mind of the internet, this is always what a model looks like.

Ravieshwar Singh, a digital fashion designer who has tried to raise awareness of the issue, even organizing a small protest during the recent AI Fashion Week, said the current moment is particularly important in tackling these issues.

What we’re seeing now is building those standards in real time with AI, he said.

Except now, brands won’t be able to fall back on the justifications they’ve used in the past for not launching certain types of models or not representing different groups. Where they might have previously pretended they couldn’t find the right curvy pattern, they are now able to generate the look they want, Singh pointed out. While they might have claimed in the past that producing a range of samples to fit a range of bodies was prohibitively complex or cost prohibitive, there is no longer any cost nor of major additional complexity. (This raises the related question of whether brands should use AI instead of hiring human models, but the reality is ignoring technology won’t make it go away.)

So the question for me becomes: Why are we making these choices in the first place? Singh said.

There are factors beyond technology at play. Brands often try to present an ambitious image that replicates what society deems more broadly desirable. On the other hand, fashion is also more influential than most other industries in defining what desirable looks like.

For the industry to deviate from its paradigms and ultimately change them would require additional thought and effort. It will be up to individual brands and creatives to introduce more diversity, and there is no guarantee that will happen. Fashion has tended to resist even small changes in the past, and if that means more work, there might be those who won’t put in the effort, which means fashion will continue to reinforce the same patterns.

The tech industry is still grappling with its own bias issues in AI. There are many well-documented examples of AI treating white males by default, with consequences such as voice recognition does not work well for women Or image recognition error black men. Generative AI adds its own risks, like perpetuating negative stereotypes or erasing different groups simply by not including them. A problem with some image generators is that they can defaults to a white male for just about any promptpositive or negative.

Technical experts and researchers believe a possible way to solve the problem is reinforcement learning from human feedback, a technique that, true to its name, involves a human provide feedback on the AI ​​model to steer their learning in a desired direction without the human having to specify the desired outcome.

I’m optimistic that we’ll arrive at a world where these patterns can be a force in reducing bias in society, not reinforcing it, Sam Altman, CEO of OpenAI, the company behind ChatGPT and image generator DALL- E, told the rest of the worlda global technology news site, in a recent interview.

Singh thinks AI could also have a positive influence on fashion. If someone creates an AI campaign with a South Asian model, or includes someone with a body type that hasn’t been the fashion norm in the past, a casting director might see it and having the idea of ​​doing the same thing in a physical casting.

However, fashion companies using generative AI must first think beyond the history of default decisions and the technology they take for them.