Episode 216: Could AI Save Fashion?

Did you have a kid in your class who always started every book report or presentation with a quote? Or maybe a definition from Webster’s Dictionary?

I definitely grew up with a kid like that.  While we moved around a lot while I was kid, we stayed in the same place (or at least the same school district) from 5th grade through 10th grade.  And in middle school, kids started to be segmented by the “track” they were on.  I don’t remember what the other tracks were at my school (because I was too busy worrying about whether or not I would ever go through puberty), but I was in the “college prep” track. And that meant that all of my classes were with the same kids for years.  They are the kids that I remember most from my tween/teen years, even if I haven’t been in touch with them in adulthood.  And there was this one kid who year after year, started every report with either “Webster’s defines _____ as _____” or something like “‘She had dumps like a truck, truck, truck, Thighs like what, what, what, All night long (c’mon), Let me see that thong’ Sisquo said this in 1999 and this is my report about the invention of underwear.”

So I’ve always been apprehensive about starting anything that I’ve ever written anywhere with a quote, because, well, it’s always felt like seventh grade to me. But today we’re going to get started with some quotes that ultimately proved to be a bit…short sighted? Wrong? You decide.

In 1903, an unnamed president of Michigan Savings Bank told Henry Ford’s lawyer, Horace Rackam, “The horse is here to stay, but the automobile is only a novelty — a fad.”

In 1946, Darryl Zanuck, an executive at movie studio 20th Century Fox said, “Television won’t be able to hold on to any market it captures after the first six months.  People will soon get tired of staring at a plywood box every night.”

In 1981, inventor Marty Cooper said, “Cellular phones will absolutely not replace local wire systems.” Oh and btw, Marty Cooper literally invented the cell phone, so maybe he was just being modest? 

And in 1995, Robert Metcalf, literally one of the early inventors of the internet said, “I predict the Internet will soon go spectacularly supernova and in 1996 catastrophically collapse.”

No one wants to be on the wrong side of history…and then be quoted about it forever on the internet.   And I–as a person who has always loved technology and has often been an early adopter when I could afford it–do not want to the be the person who drags my heels or shuts down conversation about technological advancement, especially when we are facing so many crises all at once that could (maybe) be helped by technology. Like, I do believe that technology and new invention are going to be a key component of fighting climate change.

Earlier this year, when German slow fashion organization Fashion Changers asked me to speak at their annual conference, well, first I was honored. Like how is it that this cancer kid from a trailer park in a small town in PA was known by people in other countries? And known and respected enough to speak at a conference and be featured in their magazine? Like, people have written about me in a language that is not english?! I still can’t believe it. 

For my presentation I was asked to speak about AI and how it could (or could not) benefit the fashion industry.  The theme I was given for my presentation was “Fashion & (Generative) AI – Between Hype, Backlash and Opportunity.” This was a great opportunity to dig into some research and learn a lot about something kinda new to me (generative AI)  and how it could fit into something I know really well (how the fashion industry works). And like a lot of things I have learned about for Clotheshorse (and really kind of in life), I started hopeful, that maybe AI could be a solution to the many issues plaguing fashion.  But what I learned along the way, made me more and more skeptical.  Today I’m going to walk you through the presentation (with a lot more detail) and share what I have learned. I don’t consider myself an expert on AI, but I definitely fancy myself an expert on today’s fashion industry. And I want to show you what I learned, why I am concerned, and how all of us need to be involved in what happens next.

Welcome to Clotheshorse, the podcast that wonders when artificial intelligence can start washing dishes and mowing the lawn for me, so I have more time to do fun creative stuff (or get some sleep).

I’m your host, Amanda and this is episode 216. Today we’re going to be unpacking how AI could mitigate the impact of the fashion industry, how it’s being used right now, and how (unsurprisingly) solving fashion’s problems is going to require a lot more than just an app. Basically: is it hype? Or is something reall?

Before we jump in, I want to remind you that I am taking off for Japan this Tuesday.  I’ll be recording some mini episodes in Japan about things I’m learning, seeing, thinking about while I’m there.  These episodes will be part educational, part personal experience. They will be coming out at kinda strange times, just because Japan is always about a day ahead of North America and I will be uploading the final edited files to my google drive for Dustin to then download, mix, and reupload…so who knows how that process will work out!

While I’m gone, no orders will be shipping out from the Clotheshorse merch store, but all of the new transfer and sticker designs will be going up for preorder this week (probably once I land in Japan) and you can order those (and the existing designs) to be shipped to you when I return the last week of November.

Fashion must change its ways. Clear data about the impact of the fashion industry is difficult to find because the industry is so complex and overlaps with many other industries: agriculture, oil/gas, and logistics. That means all data is directional, but it does give us a feeling for the scope of fashion’s impact. 

• The fashion industry is estimated to be responsible for 8-10% of global carbon emissions – more than international flights and maritime shipping combined.  This is not surprising now that we know the vast majority of clothing is shipped around the world–from factory to warehouse to customer–via airplane.
• The global industry produces about 150 billion garments every year.  Somewhere between 15 and 45 billion of those garments are never sold. This overproduction is the result of many issues: 
  • Buyers chasing the wrong trends.
  • Design teams not having enough time/staffing to catch major fit and quality issues.
  • Brands overbuying in order to get better pricing and meet aspirational sales plans that appeal to shareholders and investors.
  • These garments end up in landfills, incinerators, or at the industrial shredders.  Some retailers offload their unsold inventory to thrift stores/charity shops, crowding out better, more desirable secondhand product.  This leads to more secondhand clothing being exported to the global south, where its disposal has created an ecological disaster.

• In the EU, 1 in 5 garments (20%) are returned to the retailer.  In the U.S. that rate is 30%.
  • In many cases, these returned garments are never resold to other customers.  They are often destroyed or fed into the global secondhand clothing trade.
  • These returns are often caused by issues with fit and quality.
• Ultimately, the biggest issues facing the fashion industry are 
  • speed, 
  • overproduction by the retailers,
  •  overconsumption by customers, 
  • and the expansive human rights issues created by the fashion industry.

These are all really big things, and while they overlap in spots, they are also somewhat independent of one another.  So it’s hard to imagine a single cure for it all, right?  Changing this is going to require some personal change, social change, and systemic change.  We have our work cut out for us.

If you’re here listening to me talk, It’s because you love fashion, style, and the creative expression of clothing. All of us here in the slow fashion community?  We know this is worth fixing.

The internet has been flooded with stories extolling AI as the answer to fashion’s issues of sustainability and impact.  

Headlines range from referring to AI as a “Tech Infused Revolution” for fashion to a “game changer for sustainable fashion.” Lots of use of the words “revolution” and “unlocking a better future.”  When you see enough of these, it is easy to believe that perhaps AI IS the solution to these problems. It offers hope that changing fashion will not require very much change at all from the industry, just a few additional budget line items for this new software.  

This is not unlike greenwashing campaigns: greenwashing is successful as a marketing tool because it offers customers the possibility that clothing can be made without impact. That there are fabrics that cancel out the repercussions of overconsumption.  That we can carry on without changing our own relationship to clothing, shopping, and stuff.

AI could be part of the solution (but a better future for fashion will require a lot of systemic change, too).  

It is very important that we recognize that.  There is no easy “quick fix” for fashion’s environmental issues. When we acknowledge that, we can begin to see that at least part of these conversations about AI is hype. 

Knowing that AI is not the solution on its own, is it at least an important part of the solution? That’s a bit more complicated.

The path forward to a better, healthier fashion industry is multi-faceted

Technology–including AI–can be a part of it.

• • Slow down the production process.
    • More time to fine tune fit and details. This could mitigate return rates and create products that customers want to wear longer.
    • Shift from air freight to ocean. It is estimated that the carbon footprint of a garment shipped via airplane  is 14 times higher than a garment that is shipped via ocean. And as we have talked about here in the past, most of the clothes we buy in 2024 have been on at least one airplane ride, if not two or three (especially if you bought it from Zara).  But even if you ran over to Target right now to buy a t-shirt, most likely that shirt got into the US via airplane.

• It is difficult to see how technology (particularly AI) could make an impact in this area.  This is really more an issue of changing HOW the fashion industry currently works:  slowing down trend cycles, reducing style count, increasing head count in design and production, and giving teams more time to execute product development.

• Stop overproduction.
  • Utilize technology to more accurately predict demand by size and style. Right now, buyers and planners only have very rudimentary tools at their disposal to predict sales by style and size (spreadsheets and calculators). Could AI be used to better predict consumer demand? This could reduce the amount of unsold inventory.  In the US, Old Navy invested heavily into a plus size expansion in all of its stores.  By all accounts, the product fit well and customers liked it, but the units made by size were wrong: Too many of the smallest and largest sizes, too little of the sizes that customers wanted.  This resulted in high inventory liability and unhappy customers. The program was scaled back. Could AI based tools use consumer data along with sales history to more accurately predict the inventory needs by size? 

• Revisit pricing and margin targets: even with better tools for predicting demand, the reality is overproduction is also the result of the pressure to increase profit margins and reduce product costs.  The larger the buy, the lower the cost.  So buyers are often ordering far more than they can sell in order to meet these aggressive targets. Once again, this is not an issue that can be solved via technology. This would require large scale industrial change.  That does not mean it is impossible!

• Reduce return rates: keep items out of the landfill and encourage customers to wear garments longer/stem overconsumption.
  • Size and fit seem to be the biggest challenges fashion faces from a product development perspective. No one is doing a great job of this because size and fit have always been issues for mass production.  Could AI finally take data sets of body measurements and proportions and create better grading, patterns, and fit? Could AI help create better size charts? We will talk about this more in a few moments, but I will tell you that despite combing through one article after another extolling the virtues of various AI design tools, I did not find any that addressed these issues of fit, grading, and size charts.  I did see some options for making better size recommendations to customers based on previous purchases and returns.  While that feels useful, it seems that its effectiveness would be limited by the consistency of a brand’s grading and fit.
  • Improve product quality so that customers wear clothes for longer periods of time, reducing the constant flow from retailer—>customer—->landfill/donation bin. Yet again, this is not an issue that can be solved by technology.  It would require a major shift in the status quo of profitability expectations, product lead time, and head counts for design and production teams.

Before we go into the potential uses of AI, we have to talk about the environmental impact.
The training process for a single AI model can consume thousands of megawatt hours of electricity, releasing hundreds of tons of carbon into the atmosphere.

The demand for increased electricity to power AI–expected to exceed the total electricity usage of Belgium by 2026–relies on heavy fossil fuel use.
Studies have indicated that using AI for a task consumes 30-40 times the amount of energy of a person using a “less intelligent” form of technology to do the same thing. And btw, in this case, “less intelligent” doesn’t mean pulling out an abacus or doing long division by hand, it means using a calculator or spreadsheets. Creating art in Photoshop or Illustrator. These amazing tools are “less intelligent” than just typing in a prompt and getting results. If you want to take the human brain out of the equation, you need a lot of electricity.

Data centers get hot and cooling them is a challenge. Water is used to cool data centers by filling pipes surrounding the building with cool freshwater. This freshwater evaporates, rather than being used for drinking water, agriculture, etc And this water use is no joke. Research has indicated that 500 milliliters of water (approximately your standard 16 oz plastic bottle of water) are used every time you ask Chat GPT a series of between 5 to 50 prompts or questions. The same goes for asking Chat GPT to write a 100 word email. And the proof of this water consumption is very obvious.
Remember how I talked about Microsoft needing more electricity to power all of its AI tools? Well it’s also using a lot of water. In its 2022 environmental impact report (which was released in 2023), Microsoft revealed that its global water consumption grew 34% from 2021 to 2022 (to nearly 1.7 billion gallons, or more than 2,500 Olympic-sized swimming pools) This was a major increase compared to previous years. Experts say this is driven Microsoft’s heavy investment in AI and its partnership with OpenAI, the company behind ChatGPT. Microsoft has a lot of initiatives in place in an attempt to reduce their water use, but I’m skeptical.
And btw, Google reported a 20% increase in its water use for the same time period.
Earlier this week, Commons Earth posted about this on Instagram, and the comments were full of a lot of whatabout-ism. You know “well what about instagram? Or email? Those use water, too and you’re not talking about that.” And yeah, no doubt much of the tech infrastructure that powers our lives in 2024 involves water consumption. But that doesn’t mean that we need to turn off the internet for good. It means that we have to pressure these companies to find a better solution. And it also means that maybe we re-evaluate the “necessity” of some of these tools. And beyond that, most importantly, we use this information to recognize that AI isn’t some miraculous planet saver.

Knowing that AI uses so much energy, etc, we should prioritize using it for tasks that could be an environmental win.  

Otherwise we only exacerbate the environmental impact of fashion.

And listen, I do think think there could be good uses of AI that could mitigate the impact of the current fast fashion industry:

Forecasting demand and stemming overproduction.

• AI pulls a variety of data to forecast consumer demand by product, color, size.
  • Historical sales data
  • Market trends data
  • Consumer behavior data
  • Economic indicators
  • Competitor activity
• AI uses this information to identify top sellers (potential reorders/expansion of collection) and bottom sellers (markdown/do not order again in the future.
• Recently, AI specialists Paretos partnered with ARMEDANGELS to address the brand’s issues of overproduction. “Through Paretos’ order recommendations, we have the potential to reduce overproduction at ARMEDANGELS by more than 40,000 pieces annually – a direct contribution to our sustainability strategy,” Martin Hofeler, CEO of ARMEDANGELS

• Is this potential benefit reality or hype?

• Paretos works with a variety of clients across many industries, but ARMEDANGELS seems to be its first fashion client.  On its own website, Paretos does promise “40% higher accuracy in demand forecasting.” What that 40% means is vague: a 40% reduction in overbought inventory? A 40% decrease in individual planning mistakes? This is very unclear.
• The brand used this technology for the first time to plan its Spring 2024 collection…and I have not found any updated information on its progress.
• While digging into the claims Paretos makes on its website regarding the usefulness of its software, I saw two recurring themes:
  • Greater efficiency than humans using excel
  • Cutting the time to plan inventory and product commitments down to just one hour per week. 
• When I see those claims, I know what the real underlying promise is: cut costs by reducing team size.  Right now there are already teams of people who do the things that Paretos is promising: planners, analysts, and buyers.  They take all of the data points I mentioned and use it to make decisions. As a buyer, I was the “tool” that determined bestsellers and worst sellers.  I planned orders, SKU expansion, and markdowns, in partnership with my planner. And over the last decade of my career, I saw headcount in planning and buying cut as retailers tried to balance lower prices and higher margin targets.
• Ultimately, if you read between the lines here, Paretos is not promising a more sustainable fashion industry, they are promising a more profitable business with less salaries to pay. 
• It is difficult for me to see this kind of tool as more effective than merely having enough humans working on the team to make these decisions.

Fine tuning fit and grading: this could reduce return rates and slow the rate of overconsumption. 

• Interestingly enough, despite digging through article after article about AI tools that could be used for fashion design, I found 0 that focused on fit and technical design.  In fact, there are numerous tools that will “create” designs from rough sketches or even just words (Midjourney is the best known here, but there are many other tools like Botika and The New Black)
• Many of the AI applications that promised to make the creative side of design more efficient extolled savings in time and design costs.  Translated: savings on salaries and staff.
• I also encountered numerous applications for designing t-shirts, predicting style trends, and building color stories.  Once again, all jobs (creative roles) currently completed by humans.
• Ways that AI has been used to address fit/size:
  • Ecommerce plugins like WAIR that help customers find their sizes using browsing, purchasing, and returns history for that customer. WAIR promises a 19% reduction in return rate.  That’s not a HUGE win when you do the math.  If 1 in 5 garments are returned in the EU, that’s 20 out of every 100 garments purchased.  With a tool like WAIR, that reduces it to 16. That’s an improvement, but more significant improvement could be made by fine tuning grading and improving product quality.  Still, systemic changes in  tandem with technology like WAIR could have major impact in reducing returns.

Creating bespoke garments, including custom fit for each customer: Imagine if every garment was made specifically for that specific customer! It could reduce return rates, end overproduction, and slow customer overconsumption.

• Sydney-based suit company Theodore is using AI and its own Pocket Tailor app to create bespoke suits for customers.  In addition to using each customer’s measurements to get the perfect size, it also makes recommendations on silhouettes and details based on the customer’s measurements.
• Currently, suits are produced in a factory in China, so customers must wait several weeks to receive their suit.  This could change if the company were able to create smaller manufacturing hubs around the world, where products could be made more locally, saving time and carbon emissions.
• How many of you have listened to the episode of Clotheshorse featuring Angela of FABRIC? Well one thing FABRIC has that makes them super unique is a massive piece of technology. The Kornit Presto Direct-To-Fabric printer allows designers to print full color prints (or solid colors) on the individual pattern pieces and on the final fabric for garment that just needs to be cut out and sewn together.  Pairing AI bespoke technology with a machine like this makes bespoke garment production fast and local. 

Nonetheless, we do not know for certain that these “wins” counterbalance the climate impact of AI.  More research is needed.  What we DO know is that none of these AI tools “fix” the sustainability problems of fashion on their own.  Large scale systemic changes are also required.

AI is in use in the fashion industry right now, but not for applications related to sustainability.  Unfortunately the fashion industry is currently using AI to replace the human element of fashion.

Replacing designers with AI to create designs and prints:

In the past year, two American brands were under fire for using AI developed prints: Selkie and Baggu.  While both brands claimed that the prints created using Midjourney were then fine tuned and finished by humans, it raises the question: why couldn’t they have been designed from beginning to end by a human? In the case of Selkie, the designer/founder of the brand said she was unable to design the constant flow of new designs completely on her own. This is especially concerning when we consider two things that we know about AI:

• Generative AI is trained by the work of other artists and designers. This raises the ethical issues of pulling from artists who have not opted in to training generative AI.

• The energy use of AI is 30-40 times that of a human doing the same work.  Is this a good use of fossil fuels?

“On paper” using AI to design prints and products is a win for profitability: it allows brands to increase style count/guarantee constant newness without hiring more designers to do that work.  It could even result in cutting headcount on design teams.  There are other reasons that generative AI is appealing to retailers:

• While no one knows for certain how it works, several lawsuits allege that SHEIN uses AI to scrape designs and art from the internet, copying the work of smaller brands and artists.
• Brands in the US are asking designers to ONLY use Midjourney to create designs.  This adds a layer of plausible deniability in situations of accusations of copying/stealing designs. This completely changes what it means to be a designer. Now I’m going to share some information I learned from friends who are still working within the fashion industry. All of them wished to remain anonymous because they still work for a company using AI in place of “traditional” design.  So I’m also not going to name their employers.  But I will just say this: they are people I trust and their stories confirmed other stories from people working for other brands under the same corporate umbrella. This larger company has several brands under its ownership that people often don’t consider fast fashion because they have excellent branding and nice store merchandising, but they are in fact, fast fashion. 

What happens when humans are no longer doing the work of creating new designs?

• From a purely artistic standpoint, is giving the job of design and creation to machines removing the “art” from fashion?
• Could fashion hit a wall creatively when AI runs out of new ideas (because no no ideas have been created by humans)?
• Will fashion become less interesting and appealing? Could this actually result in MORE unsold inventory?

Could AI remove the art and creativity from fashion?

This is particularly concerning to me because fashion began as an art form and it remains a personal creative expression, a space for creative innovation and evolving vision.  How does that change when humans are no longer doing the art part? Yet still doing the manufacturing.  Note that no one is working to automate the most difficult aspect of fashion–clothing production–but many are selling solutions for product design, visual merchandising, copywriting…all of the creative aspects of fashion.  The elements that bring fun and joy to both the creators and the customers.

 Replacing human models with AI-generated “models.”

• This is strictly a cost saving measure, but the savings most likely are not as high as the environmental cost. Last year, Levi’s came under fire for partnering with Lalaland.ai to use AI generated models.  Levi’s felt that this move would decrease return rates by showing products on a more diverse range of bodies.  Critics wondered why Levi’s just didn’t hire a more diverse range of human models.
• How do ideal fake humans impact our collective mental health and comfort/satisfaction with our own bodies?
• This also raises the ethical implications of models’ rights to their image, as AI-generated models are based off of the faces and bodies of very real models and humans. 

We have the power to change the way fashion is using AI!

Nobody has more power to change the course of the fashion industry than those of us who truly love fashion!  

As individuals: One thing I have learned in my decades of work as a buyer within the fashion industry? Only two things force it to change: the law and fear of lost sales/lost customers.

• Tell brands that you refuse to support brands using AI to replace designers and models.
• Vote with your wallet! Boycott companies using AI in ways that do not align with your values.
• Tell your friends and family about the environmental and ethical impacts of AI in the fashion industry.
• Push your elected representatives for regulations around fashion’s use of AI.

As passionate fashion industry professionals:

• See past the hype of AI and recognize that it is not helpful without large-scale systemic change within the fashion industry. Share that information with your peers and cross-functional partners.
• Ask the difficult questions in meetings: WHY are we opting to use these tools knowing that the environmental impact is significant? WHAT are the measurable benefits of these tools? How will these work in tandem with larger process changes?
• Recommend tools that may be beneficial or advocate for increased head count.