It’s been just over a year since Kering introduced Madeline, a ChatGPT-powered shopping assistant that consumers can use to search for products and get product recommendations on KNXT. KNXT is an e-commerce site secretly used by luxury conglomerates as a testing ground for digital innovation. .
At the time, KNXT promoted it on Twitter (since renamed to X) as the end to the endless scroll for the perfect luxury item. The reality was less dramatic. In early testing, Madeline’s answers proved to be limited and robotic, suggesting products that weren’t necessarily the best fit for the occasion and sounding like marketing copy. A notice on the site states it is currently under maintenance and does not list a reopening date. Kering did not respond to a request for comment.
Madeline appears to have suffered the same fate as other generative AI-powered experiments that hit the market after the release of ChatGPT in late 2022, amid growing excitement about the capabilities enabled by large-scale language models. Companies immediately began testing technology to conjure up design concepts. , create images for marketing campaigns, write product descriptions, and chat with customers. McKinsey estimates in 2023 that generative AI could increase operating profits in the fashion and luxury sector by up to $275 billion over the next three to five years.
There remains a lot of optimism about the potential of generative AI, with billions of dollars of investment flowing into startups trying to make it a reality. But rapid adoption seems less certain. The U.S. Census Bureau said in a March report that, “While further analysis is needed, the rapid development of generative AI suggests that an explosive increase in AI use in enterprises between September 2023 and February 2024 is unlikely.” It’s not connected yet.”
Chatbots aren’t the only ones facing challenges.Levi said fashion business The company said in a statement that it has no plans to expand on a pilot program announced last March that uses AI-generated models to increase model diversity on e-commerce sites. Although this was one of the earliest and at the time most ambitious uses of the technology, it was criticized by critics who pointed out that a minority of humans already had a hard time getting modeling jobs. faced severe flames.
“We do not believe this pilot is a means to advance diversity and is a substitute for actual actions taken to achieve diversity, equity, and inclusion goals. “We did not think so, and it should not have been portrayed that way,” the company said. That statement.
This decision may have been motivated not only by technical flaws, but also by the criticism the company received. However, the limitations of generative AI are becoming increasingly apparent. More than a year has passed since generative models first gained public attention, but even the most advanced models still fabricate facts, make basic mathematical errors, and are subject to physical or historical inaccuracies. Generating accurate images.
While these can be useful for certain jobs, they often involve, as tech supervisor Molly White recently wrote, “delegating some tasks to inexperienced and sometimes sloppy interns. in the same way that is sometimes useful.
It is too early to declare that generative AI is a failure. But questions are growing about whether it can live up to its extraordinary expectations.
“Although AI will ultimately be transformative, GenAI has many technical problems, especially with respect to reliability, and is unlikely to live up to the current hype,” said a prominent AI skeptic who recently wrote Gary Marcus, who wrote about the possibility of an AI bubble bursting, said: The next 12 months are stated in the email. “It may take years or even decades for most of these promises to materialize.”
hype cycle
Emerging technologies often follow a similar trajectory. It’s so similar that Gartner, a technology research and consulting firm, codified it in his 1995 and named it the “Hype Cycle.”
A new innovation is emerging and is generating a lot of excitement and attention. Based on some high-profile successes, expectations build and reach a high point. But if early experiments don’t work, a period of disillusionment ensues. If all goes well, there will be an eventual recovery as the next generation of technology emerges and the benefits become clearer and adoption increases, but there is no guarantee that will happen. So-called one-way mirrors in stores will never catch on, and the dream of the Metaverse may never come true.
“We are placing GenAI in the retail hype cycle, and we are right at the peak of it,” said Sandeep Unni, Senior Director Analyst, Retail Practice, Gartner.
Some retailers hope LLM will revolutionize online shopping, allowing them to understand not only their questions and intentions, but also their surroundings, such as a wedding being formal but a picnic not being formal. We hope to enable truly conversational chatbots that understand You can then answer questions and recommend products.
It will be a while before most retailers want one of these chatbots. The chatbot’s knowledge is derived from general data collected from the internet. Amazon, which has invested billions of dollars in generative AI, received a moderate rating for its shopping bot Rufus, which began public testing in February. washington post They deemed it “of little use” and said they did not trust its recommendations. Amazon said it will continue to improve the bots it is developing.
“Our biggest learning is that people want reassurance from experts,” says Co-founder and CEO of Goodsort, a startup focused on AI shopping assistants that recently pivoted its approach. Managing Director Jake Stark said.
The company, formerly known as ShopWithAI, initially had an AI chatbot that recommended clothes based on the styles of various celebrities. That product wasn’t scalable, Stark said. The company still offers men’s fashion options, but it’s also branching out into watches, where AI uses writing from a panel of experts to shape its recommendations.
One of the most promising uses for generative AI in fashion is design, where factual accuracy is less of an issue. This technology allows the designer to quickly generate new ideas and also maintain his own style by training his AI based on past works. Designer Norma Kamali is fully committed to leveraging her AI, and she’s working hard to create a system that will help carry on her legacy even after she leaves the label. Startups are racing to build their own fashion-specific tools from image-generating AI models.
But whether AI-driven design will become widespread remains an open question. While some designers may reject it, rightly or wrongly, because it displaces or devalues human creativity, it’s a sentiment consumers can share. A brand called Selkie had already faced backlash from customers for using AI to create images. There are also unresolved issues regarding intellectual property issues. Revolve was an early adopter, releasing a small collection designed by AI, but the company declined to say whether it would continue to use the technology.
Hilary Taymor, founder and creative director of brand Collina Strada, quickly integrated the image generation tool Midjourney into her design process and continues to use it, but she is unimpressed with the advances made in the latest version of the tool. What drives AI’s creative capabilities is often the same unintended consequences that cause problems when generating text. As developers work to reduce these hallucinations, they could also take away from AI’s creativity. Perhaps AI is better at reproducing stereotypes of dress, but that’s not necessarily what design labels want.
“I don’t feel as creatively inspired as I used to, so I keep using the older version,” Taymor said.
The future of generative AI
These problems may ultimately prove surmountable. AI systems allow users to adjust something called “temperature” (basically the amount of randomness in the output), potentially defining how creative they want the AI to be. For obstacles faced by chatbots, such as product knowledge or hallucinations, which can turn the AI’s output into something false or absurd, retailers can fine-tune the model with specialized training and search tools that the bot can use. We use methods such as a technique known as augmented generation to address them. To pull answers from another knowledge database.
“I see the future direction of this technology as how we create a customer experience with this, reduce the downsides like hallucinations, and leverage the unique benefits.” said Tian Su, vice president of personalization and recommendations at Zalando. About the application of AI.
Last year, Zalando introduced its own AI shopping assistant for region selection. Su acknowledged that the technology is not perfect, but said it still adds value for users. The 500,000 customers who talked to bots don’t always know the search keywords they use to find what they want, but through their interactions with the AI, they can narrow down their results or discover new products. can. Sue said there is no other technology that allows these conversations with all customers at the same time.
Established capabilities could also benefit from generative AI, she added. Zalando has developed a search format that updates results in real time as users type in the search bar. This is like a chatbot that eliminates the need for chatting altogether.
There are also simple tasks for which generative AI can already be expected to work, such as writing basic product descriptions. Adobe recently introduced tools to Photoshop that allow users to fill spaces with generated images and create backgrounds that can be used in marketing assets. Taymor said he regularly uses ChatGPT to write work emails.
And technology continues to evolve. Gartner predicts that generative AI will reach a “productivity plateau” within about five years, where there will be viable products with mainstream adoption. The challenges of resolving illusions, intellectual property, security issues and regulations could all derail it, Gartner’s Uni warned. Retailers have an additional hurdle in finding talent to help them test, measure, and scale generative AI projects that deliver real benefits. But unlike fanciful concepts like the metaverse, Unni says generative AI is actually a broader extension of his AI, and the value proposition there is much more established. I did.
However, there are real limitations to overcome first, and there is no guarantee that they will be resolved. But a product doesn’t have to be innovative to be useful. Generative AI may or may not change the way online shopping and brand designs and images are created. Subtle uses may be found behind the scenes to evolve what already exists, as algorithms already do.
“I think I’m probably more excited than I actually am,” Sue said, emphasizing that even her “maybe” is a question mark. “But there’s something real about it.”
