Changing Roles

This article was originally published in The Interline’s AI Report 2024. To read other opinion pieces, exclusive editorials, and detailed profiles and interviews with key vendors, download the full AI Report 2024 completely free of charge and ungated.

---

At a whole-society level, thought leaders and analysts are still split on whether breakneck improvements in AI, and what seems like its inevitable widespread adoption, are more likely to result in greater productivity, efficiency, and support for existing talent, or in longstanding roles being eroded away by automation.

And in both cases, the same commentators are also undecided as to which disciplines will be the most affected – with the prevailing sentiment being that essentially anyone, from the junior creative to the seasoned CEO, could have their position either augmented or usurped. Where once machine learning-linked job transformations were reserved for clerical and statistical roles, today a massive spectrum of creative and commercial jobs are potentially in AI’s sights.

What does this mean for fashion?

In researching this piece, the experts I spoke to confirmed the expectation that few jobs will be unaffected by AI. In the very near future, manufacturing, marketing, sales forecasting, designing, and even modeling will all necessarily leverage AI as either a creative aid, a productivity boost, or a way to minimize costs and reduce time to market.

And in the slightly longer term, AI does have the potential to subsume some of those roles entirely.

In every case, there’s also unanimity amongst analysts that fashion industry professionals who become fluent in these technologies will have the advantage in the job market – either as a lever to safeguard their jobs, or as a way to demonstrate their commitment to embracing new technology. Because, as the CEO’s current favourite adage goes, “AI might not take your job, but someone proficient in using AI might”.

Routine tasks and admin-intensive workflows will be first on the chopping block.

“AI is most likely to replace roles [primarily made up of] repetitive tasks,” explained Professor Linwei Xin, from the University of Chicago’s Booth School of Business, when we spoke this spring. According to Xin, this is an extrapolation of the existing trend of companies using automation in areas where they don’t have enough humans, or enough interested humans, to fill highly routine roles.

Until now, that automation has taken a more common and recognizable form: factory robotics, just-in-time supply chains, logistics, warehousing, and distribution. These are what we might characterize as the jobs where technology has been slowly augmenting and replacing human labor for decades. In the United States and China, where relative labor costs have skyrocketed, automation and robotics deployed up and downstream have been key components in reducing companies’ overall costs.  And as Xin pointed out to me, this same driving force is likely to be behind the adoption of AI across the fashion value chain – making its probable impacts easier to predict.

Some of the fashion industry’s manufacturing is already automated, particularly cutting, explained Juan Hinestroza, Cornell University Professor of Fiber Science and Apparel Design, who I also spoke to in researching this piece – although the global distribution of that automation is still uneven.

Depending on where garment and textile manufacturing is taking place, it is still currently cheaper for companies to employ humans to make and sew in markets where labor is cheap. But this trend, too, has a clear through-line from the original driver for automation: cost cutting. Fashion has, for several decades, lifted and shifted its sourcing and manufacturing from one region to another as micro and macro-economic forces evolve.

In industry parlance, this is what’s referred to as “chasing the cheapest needle,” and it has remained common practice even in the face of coordinated campaigns to “reshore,” or return production to consumption markets like the US, EU, and UK.

Hinestroza reminds us that, in the 1970s and 80s, fashion jobs boomed and busted in Eastern Europe, then they moved to China – and now they’re being relocated to Africa and South America where there is a net benefit to brands that spans both proximity (shipping is quicker and cheaper from those locations to key markets in Europe and North America) and labor cost.

“[As a brand] you are always racing to minimize costs and continue your production,” Hinestroza says. “But what do you do with people who that’s the only thing they know how to do?”

And this, really, is a critical question. It may be more likely for on-demand, digital production technologies and deployable, composable ‘microfactories’ to replace offshore manufacturing, but the principles and the unanswered queries behind AI adoption are set to be the same.

When you have an established system, or a part of one, that operates and sustains itself based on the inputs of people who have been exclusively trained on just that system – what happens to them when the commercial justification for letting AI do their jobs becomes stronger than the social and moral imperative to keep them in place?

This is more than just a theoretical tension, too, since there are entire segments of the fashion value chain where AI already has the capability to replace a swathe of human talent – and we are already seeing a pushback from communities (and the general public) that suggests that a delicate balancing act is currently taking place.

Will AI replace real live models?

The process of making final physical production samples in one part of the world, shipping them to another, identifying the right human beings to wear them, then booking studio space or locations for a professional photographer to shoot them in – and then translating the resulting images into marketing and eCommerce materials – is a long, expensive and circuitous one. And this makes it a prime target for AI automation, since a generative model could potentially condense many of these steps into one, and many of these people and skillsets into a single piece of software.

But is it quite that simple? The commercial argument may be compelling, but the cultural counterargument is equally strong.

Dutch AI studio Lalaland has become a prominent name in fashion thanks to its pipeline, which can generate lifelike fashion photoshoots featuring artificial models by training an algorithm on actual photos and on a brand’s specific products, as their CEO and Founder Michael Musandu explained to me. At a first glance, their work raises clear questions about whether fashion brands, which spend thousands to millions of dollars a year to hire models alone – without even factoring in the costs of all the other photoshoot elements I mentioned earlier – could cut back on the money they spend on photoshoots, and the models, photographers and other professional skills and materials needed for them, and reinvest that income elsewhere.

Instead, Musandu believes their AI models are helping to create equity for brands of all sizes and different budgets, allowing them to scale their photoshoots without those overheads becoming constraints. “By saving 70% of the costs associated with traditional shoots, we’ve seen brands reinvest these savings into more marketing campaign videos and images featuring real models,” he notes.

“Lalaland’s AI-model supplement does not replace real models,” Musandu pointed out to me. “We believe human models will continue to play a vital role in the fashion industry, establishing genuine connections with consumers. Our technology aims to support this.” Musandu argues that the industry will not only need to continue employing models and photographers and lighting engineers and booking studio space, but it will actually need more of that talent to come from more diverse and underrepresented groups, and that AI-enabled automation can be a lever to help that happen.

[The complicated status of AI and diversity and inclusivity is analysed in more depth in another feature contained in this report – Editor]

And this raises an important consideration for any organization looking at AI as a way to augment or automate any creative, technical, or commercial role: does that job create a pinch point by simply existing, or could the time and effort the person filling that role expends create greater value elsewhere? Or, in other words, is the opportunity a matter of throughput and speed, or of releasing untapped potential?

The answer to this question is likely to fall somewhere in the middle, but the template being set by AI-generated product photography does suggest that fashion is going to see an evolution of traditional roles rather than a wholesale replacement of them. Although the net result may still be fewer jobs in some areas, including those that have traditionally been considered sacred in fashion.

The industry will trend toward hiring fewer, more AI-fluent designers.

Despite the public perception, technology-enabled or technology-adjacent automation is rarely the proximate cause of a shift in the jobs market. Instead, analysts and thought leaders suggest, the real replacement for an entrenched role is talent with a different, more contemporary skillset.

“It’s not that the worker gets replaced by a robot or a machine in most cases, especially for desk jobs, it’s that some better or more educated worker can do that job because they can be twice as productive or three times as productive,” Code.org Founder and CEO Hadi Partovi said during this year’s World Economic Forum. “The imperative is to teach how these tools work to every citizen, and especially to our young people.”

And Hinestroza reports that he’s been using AI for almost two years himself, as well as teaching generative and non-generative tools in his classes because his students “need to be prepared [for the] reality of the industry”. Designers in particular need to be fluent, Hinestroza told me, because the largest improvements in the underlying technology have been observed in generating visuals.

And Hinestroza points out, that at the beginning of last fall semester, the output of generative image models went from lackluster quality to experiencing an “incredible leap of images and technology”. When he attended the biannual Canton Fair, one of the world’s largest trade shows, he noticed nearly every company designing was using text-to-image generators like Midjourney and Dall-E as a way to cope with the demands of creativity in a fast-paced industry. “If you look at a butterfly, and want a collection based on butterflies in Mexico, [AI models] will create an entire collection for you,” he says.

Does this mean the end of non-AI fashion design as an occupation? Hinestroza reminded me that we simply don’t know yet. One potential outcome is that designers and AI become co-pilots, with generative models providing creative inspiration and assistance – and even potentially helping with technical design and patternmaking tasks. On the other hand, companies are signaling that they won’t need as many designers, with the hope being that the use of AI tools could help to open up other roles for those professionals – even if the prospects are potentially bleak for junior designers who may no longer get the opportunity to prove themselves before being elevated to mid-weight and senior roles.

In an use case that may feel dystopian or unremarkable, depending on your perspective, Hinestroza described a hypothetical example to me whereby a brand might fully automate a spring collection based on fruits in Asia, for instance, and then a designer could create a separate portfolio leveraging different tools – and another executive could choose which workflow and which set of tools did the better job. And while this may not be completely a case of pitting people against machines, it would, in the very least, be an instance of machine-assisted humans being set against machine-native creative pipelines.

Crucially, the timeline for this kind of experimentation is short, as Hinestroza told me:

“Most of the companies are using these tools. Some admit it. Some don’t. [I] don’t know why they don’t admit it. Maybe they want to pretend to be more human-centered, but at the end of the day, you have deadlines [to meet] and [AI] can accelerate the process.”

AI optimizes. Humans make decisions.

Earlier this year, Massachusetts Institute of Technology labor economist David Autor argued in Noema magazine that the fear of AI replacing jobs is misplaced and, instead, the adoption of AI could instead serve as a catalysts for what we might term ‘capability elevation’ – enabling a larger set of workers to perform higher-stakes decision-making.

And the experts I interviewed were all clear that AI’s impact will not just be felt in narrow roles, but that AI should, at least in theory, enable mid-level to executive-level fashion industry professionals to make better–and fewer–decisions.

“Imagine selling 100,000 products. You have to manage each of them, every week [deciding] how many of this shirt [to order from] from this supplier. If you’re selling thousands and thousands of products, it would be difficult and tedious for human buyers,” Xin explained to me. Which is equivalent to asking how can one person make all of the decisions that need to be taken across the extended product lifecycle, and in the compressed timeframe needed to bring new styles to market when they still have a good chance of selling.

Xin described to me an example of how AI used past sales data to do a better job of forecasting so that “your human buyer can focus more on best sellers” and continue the job of communicating with suppliers.  A prime instance of something considered an intuitive or ‘higher order’ task that is, in reality, another matter of efficiency, time, and accuracy – underlining the extent to which technology can progressively shift the window of what is considered a task that’s suitable for automation.

Instead of having people fly around the world and identify trends by visiting runway shows or conducting field trips to fashion capitals, now algorithms work more effectively and efficiently. They process social media photos of what people are wearing in Singapore, London, or Rio and, in real-time, decision-maker are able to “ask the machine” to create a collection based on the next trend.

Large brands use mass customization instead of traditional forecasting to reduce the lead time to produce specific items, Xin explains. In this process, AI predicts trends and optimizes their scheduling from design to delivery. When customers choose preferences for the product they’re purchasing, the brand effectively delays customization until the last step in choices such as color and material. This concept is known as postponing.

Casting the net wider, Xin could not comment on whether or not the fashion sector would cut marketing jobs. Hinestroza, though, told me that he believes most marketing now can be 100 percent automated with AI tools. His students are already learning how to use AI to market clothes to different ages and different markets, and earlier comments from Musandu at Lalaland reinforce that the choice between tradition and automation is not always a binary one – suggesting that this next generation of talent will be the one to navigate the fine line between an evolving technology frontier and cultural attitudes.

AI can also optimize shipping routes and manufacturing, Xin says, even predicting disruptions rather than leaving brands needing to react to them. The push to make supply chains more robust has ramped up after COVID and the Russia-Ukraine war. For the 20 years prior, companies wanted to minimize operations, leaving them vulnerable to disruptions and limitations in supply chains, he explains. Even with AI taking on a powerful role in planning and monitoring supply chains, though, Xin argues that there would still need to be high-level executives managing the human connections and relationships that keep these supply chain functioning.

Xin did express concern about how automation is replacing some formerly human-powered jobs. However, he repeatedly noted that there will be many jobs AI is less likely to  replace, particularly those that involve human relationships and their psychology. Xin used chess moves as an illustration: AI chess engines can tell a player what the best move is, but only a human coach can tell a player why it’s the best move. And from a disclosure and trust-building perspective, that transparency is likely to matter a great deal.

Brands—and stockholders—benefit. Tech companies are next.

With such an unclear picture of how and where AI is going to supplement or supplant human labour, the question remains: who is financially benefitting from the proliferation of AI tools in fashion?

Hinestroza says tech companies aren’t making huge amounts of money from it yet–but they will soon enough. This is likely down to the high cost of training and running inference for large AI models, but there is already evidence to suggest that these costs are being driven down, and that the sliding scale of customer revenue to cost will soon tip over into profitability.

Brands, though, are already successfully minimizing their cost, Hinestroza believes. In a new workflow, AI generates designs quickly, the brands choose the best designs for human creatives to either iterate on or approve, and then the manufacturers create prototypes very fast. This also, in effect, minimizes production because it reduces the overall amount of prototypes developed and increases the adoption rate of design ideas and samples to finished products.

And Hinestroza also reminded me that, in his experience, most of the designers for big brands are already contractors. In the future, brands may only need to hire a reduced number of full-time people to take care of AI programs. And while Western brands may hesitate to adopt similar workflows in their own operations, Hinestroza told me that when he was in China, he saw substantial third-party design companies leveraging AI design tools to create their new product catalogues, and those designs then being sourced by brands and supplying clothes all over the world, knowingly or otherwise.

So while obvious AI use by domestic brands is the tip of the iceberg, the reality is that AI has likely already deeply infiltrated the fashion value chain and is creating a commercial edge for suppliers – one that brands themselves will want to replicate.

Quite how this develops culturally, is currently very difficult to say. The impact on fashion’s talent base is clearly imminent, but how that impact will be perceived by the shopping public is hard to predict given the backlash against the use of AI in creative fields.

“If you’re in the business of fashion, you have to sell clothing to people, not computers. It’s a complicated phenomenon,” Hinestroza concludes.