Our regular analysis selects one or more news stories from fashion technology, and presents The Interline‘s take on why they matter to our global brand and retail audience – as well as what they might mean for the longer-term future of fashion. As always, this analysis is also delivered to Interline Insiders by email — and signing up continues to be the best way to get a fresh look at the fashion technology news, completely free, in your inbox.
AI-generated models: inevitable job cuts clash with inauthentic inclusion
Artificial intelligence (AI) retains its grip on the headlines this week in both predictable and surprising ways, proving just how pervasive it is and is likely going to be for the foreseeable future. Two different companies, Deep Agency and Lalaland.ai for Levi Strauss & Co, are currently in the spotlight for their AI-generated models. And although they are each taking slightly different approaches to the technology – and both seem to still be in the early-ish stages of commercialisation – growing awareness of just how quickly AI models are being rolled out in place of people already has many debating the potential harm.
From a pure business efficiency point of view, AI-generated models are a clear win. Unlike photoshoots incorporating real models, generative ones negate the need to have real production samples in-hand, to book studio space, to hire photographers, and to hunt for models with the right look across the full spectrum of gender, ethnicity, body type, etc.
In their response to some of the initial backlash, Levi’s did indicate that their experimentation with AI-generated models was intended to supplement the continued use of human models, and that it is their intention to continue to represent a range of body types, ages, and skin tones. But not every brand will be so magnanimous when the scale of the cost-cutting potential becomes clearer, and while the first brand or retailer to exclusively use generated models will no doubt be using them to deliver an ersatz version of “representation,” the uncomfortable fact remains that AI-generated diversity is not real diversity.
It is, of course, not the AI industry’s fault that the fashion and modelling industries have been exclusionary for far too long, but the timing of the deployment of AI models, and the disruption they represent, could scarcely be worse. Just as fashion is making headway towards inclusivity and representation, the modelling sector is set to experience a historic upheaval.
To understand why this will make generated models such a lightning rod for criticism (beyond the obvious legal quagmire of using real people in the training datasets behind them) think of it this way: AI is all but guaranteed to eliminate a lot of modelling jobs because the bottom line argument is just too compelling for any commercial enterprise – especially a publicly-traded one – to ignore. But access to those jobs has not been equally distributed prior to the looming extinction event.
Losing your job as a model – to an AI model in both sense of the word – will be devastating. Losing your modelling career to AI when people who look like you have only recently started to earn proper representation in the first place, is on an entirely different level – like having a door held shut, then opened to you briefly before it’s slammed in your face again.
Currently, brands are – rightly – held responsible for under-representation or exclusion of certain groups of people by the lens of public scrutiny. It’s a straightforward task, as someone from a marginalised group, to look at a brand’s marketing material and recognise that you’re not reflected in it, and to want better. But in the very near future it may become difficult – then impossible – to know if the representation you’re being shown is real. We might see incredibly diverse, inclusive brand campaigns, catalogues, and e-commerce frontends, but have no idea whether or not they represent authentic or artificial inclusion.
And as we’ll see from the rest of this week’s AI-focused stories, this combination of huge upheaval and rapid timeline characterises a lot of AI initiatives – and is at the root of a wide spectrum of stakeholders’ concerns around them.
Generative fashion and virality: a double-edged sword?
A common counterargument to AI taking over the modelling industry would be this: most people can distinguish between real images and generated ones, and therefore generated models aren’t ready to replace real ones. But this week’s most viral AI story suggests that that’s already not the case. And as The Interline has seen it put recently: remember that the generated images of today are the low bar for the generated images of the future.
The viral story in question? It was a fashion meets faith moment as Pope Francis apparently stepped out in a stylish puffer jacket that had many people, celebrities included, fooled. The picture first appeared on Reddit and was later revealed as having been created with generative AI model, Midjourney.
The key takeaway here isn’t necessarily that AI images are fully believable and real – although they’re certainly headed in that direction – but more that in the context within which most people consume media today, they are already believable enough. Recent Microsoft research concluded that the human attention span has dropped to 8 seconds, dipping down from 12 seconds in 2000, and in this terse timeframe, AI-generated images may well go undetected even if they have flaws that are visible on closer inspection.
This level of at-a-glance verisimilitude was enough to prompt a statement from the Pope himself, who put out a surprisingly even-handed response, saying that: “I am convinced that the development of artificial intelligence and machine learning has the potential to contribute in a positive way to the future of humanity […] I am certain that this potential will be realised only if there is a constant and consistent commitment on the part of those developing these technologies to act ethically and responsibly.”
This opinion is also shared by some high-profile people who believe generative AI has been unchained too quickly. This week, the non-profit Future of Life Institute issued an open letter that calls on “all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4.” Amongst the thousands of signatories are Tristan Harris, Steve Wozniak, and Elon Musk. And their perspective is perhaps not surprising given that a new study by Goldman Sachs, published this week, speculated that AI could replace the equivalent of 300 million jobs. According to the research note entitled ‘The Potentially Large Effects of Artificial Intelligence on Economic Growth’, AI could “replace a quarter of work tasks in the US and Europe, but may also mean new jobs and a productivity boom [and] could eventually increase the total annual value of goods and services produced globally by 7%.”
Separately from AI’s impact on jobs and culture, though, is its direct impact on fashion, which is a cornerstone of both this story and an important brand announcement, also from this week.
While a lot of the attention has, understandably, been focused on how the generated image of the Pope managed to trick so many people, it’s also important to consider the other response it elicited: that the man looked chic. This raises the question of the role that generative AI is potentially going to play in trend creation (which might be an odd term, given that large models are typically accused of plagiarising existing work, rather than creating anything novel) when viral moments can be created around a piece of clothing that doesn’t really exist. In that scenario, should brands run with the moment and make that garment?
There are two elements to consider here: does the generated item bear too many style hallmarks of another brand (i.e. could it open your brand up to copyright infringement claims, depending on the prompt) and is it something that could potentially be producible? If a generated style passes both those tests it should – at least in theory – be subject to the same process as any other viral trend. The end result being that someone – and possibly several someones – will want to bring their own versions of it to market.
In fact, this whole cycle was wrapped into a neat package this week by G-Star Raw, which used Midjourney to generate a denim collection before going on to physically make one of the 12 top results, putting it on display in their retail location in Antwerp.
G-Star Raw’s designers seem keen on the idea of using AI as a tool to aid in their creativity, saying the following: “From a designer perspective, we can compare the impact of AI design to the effect Adobe Illustrator had on hand-sketching: it improved and accelerated the process, but always as a tool.” This is a common refrain among AI proponents, too, but it hinges on one key thing: the designer being conscious of the fact that they’re using AI as inspiration and assistance.
We should also point out that G Star’s experiment is different from “the Pope’s” AI-generated fashion in that this was all an in-house project, with no consumer influence until the real garment had already been made. Still, it’s a short hop to consider a scenario where consumers generate their own denim collection, have it go viral, and then either the original brand (or another opportunistic brand) decides to actually make one or more pieces from it.
Either way, it seems that fashion is destined to have AI be much more closely involved in trend creation and design assistance than anyone could have predicted would happen this quickly. And as generated images start to float to the surface of social media and other channels more often, we’re likely to eventually (and potentially very soon) see brands unknowingly using AI as inspiration.
Fast and frightening: AI-assisted product discovery, and outpacing regulations
Speaking of rapid advancement, we’ve also seen a massive leap in the link between AI and product discovery, which is something we wrote about a month ago as being an important question to address at an indeterminate point in the future before shopping became dominated by AI.
But that future point came around phenomenally quickly, and the past week saw shopping become directly integrated with AI as part of Open AI’s first wave of ChatGPT plugins. You can find out more about how plugins work, and why they represent such an important step in AI’s incredibly swift development by reading Open AI’s own article. But the key thing to consider here is that ChatGPT was only released at the end of November 2022, and by the end of March 2023 we have a direct plugin for product discovery, search, and pay-later shopping from Klarna. So when we consider the timeline from the average person saying “I can distinguish generated pictures from authentic ones” to “only trained experts and specifically developed programs can tell them apart” we should bear in mind that we’re probably talking months, not years, for the next milestone advancement.
As you have probably gathered, developments being made in AI are now moving far too quickly for rules and regulations to keep pace – with the consequence that some regulators seem to be opting to turn the other way.
As a case in point, regulators in key markets like the UK apparently consider the AI arms race to be so important for commercial and economic prosperity, that they’ve flat-out said no dedicated regulators will be appointment, and that instead, AI will be governed under existing frameworks. While this will no doubt play a part in realising the Goldman Sachs vision for heightened regional productivity and output, and is charitably intended to avoid curtailing innovation, the question remains whether innovation or harm will be AI’s most pronounced impact on the world.
Fashion, of all industries, perhaps has the most to gain from playing in that new creative space. But it also has a lot to lose if AI proves itself capable of replacing chunks of its direct and indirect workforce. So whether ours is an industry that ends up with more in the way of benefit or harm, only time will tell.
But we probably won’t have to wait too long to find out.
The best from The Interline:
This week, we published an exclusive feature from Bryce Quillin, PhD and Jessica Quillin, PhD, as well as a feature from the Founder of NOFORM.
In their first exclusive for The Interline, Bryce Quillin, PhD and Jessica Quillin, PhD, Co-Founders & Principals of It’s A Working Title, LLC shared their experiences on how this year’s SXSW became a nexus for a debate about where digital fashion stands, and where it should be headed.
Rahul Verma, Digital Fashion Design Consultant, and Founder of NOFORM, asks if digital working automatically equates to sustainable working? And if digital-only fashion is creating a false perception of what comprehensive DPC should really mean?
