This article is part of the Future of AI, a series exploring how artificial intelligence will shape the fashion and beauty industries.
Earlier this year, tech leaders began praising the importance of taste. In February, Y Combinator co-founder Paul Graham predicted that “In the AI age, taste will become even more important. When anyone can make anything, the big differentiator is what you choose to make.” That same month, OpenAI president Greg Brockman called taste “a new core skill.”
For fashion and many other creative fields, the idea that taste hasn’t always been essential is absurd. However, the concept of taste has been amplified and reshaped in the era of generative AI, with this much-debated form of aesthetic judgment quickly becoming a buzzword. “Every company right now wants to talk about taste. Every thought leader in tech wants to write a Substack about taste,” says Andy McCune, founder of the visual inspiration platform Cosmos.
This trend serves to show that AI executives are not disconnected from the very human quality of good taste. But if taste and personal style are inherently human—cultivated through experiences with books, films, and people—can AI ever truly understand a user’s personal style or develop its own sense of taste? This is a crucial question for fashion, where such instincts influence everything from clothing design to outfit recommendations, many of which are already powered by AI.
Some in tech are confident it’s possible. “I hate to break this to everyone, but you probably don’t have better taste than the AI,” one product head posted on X. An AI CEO similarly quipped, “There’s a good chance AI will have better ideas than us within a few years.”
Those outside the tech bubble are more skeptical. “Taste and personal style is something you develop with time and with real-life experience,” says trend forecaster Mandy Lee. “Having no touchpoints to the real world is the antithesis of building personal taste. So whatever they’re talking about is not the same thing as taste and style.”
Shoppers also remain unconvinced. Currently, only 3% of shoppers surveyed by Vogue Business use AI chatbots for fashion inspiration, compared to 57% who turn to magazines (print and digital), followed by street style (47%), fashion blogs or Pinterest (36%), and influencers (35%).
Personal style has long been a way to signal aspirations and craft an individual identity within society, says Richard Thompson Ford, a Stanford Law School professor and author of Dress Codes: How the Laws of Fashion Made History. People do this by borrowing and combining references from various aspects of life—from different communities to historical periods—using familiar images from art, film, celebrities, or influencers as inspiration. Good taste isn’t about copying, Thompson Ford explains, but rather “quoting small parts of a familiar ensemble and putting them together with other things in order to express something that, at least to them, is unique and individual.”
As people increasingly turn to AI for discovery, it could reshape how they develop their personal style. Fashion-tech startups are optimistic about AI’s ability to streamline and simplify the process. The AI shopping platform Daydream aims to do just that. Its users aren’t necessarily high-fashion devotees, says co-founder Lisa Yamner. “The people who are finding us are quite needs-based; it’s more fashion enthusiast than ‘show me Loewe’s recent runway’ kind of stuff.”
With nearly half of survey respondents citing the overwhelming number of choices as their biggest challenge when it comes to shopping, AI’s role in curating and simplifying could become even more significant.As 45% of users focus on styling themselves by putting together outfits from their existing wardrobe and finding styles within their budget, it’s no surprise that over a third (36%) would consider using an AI tool to discover next season’s trends. But can AI truly understand taste and style well enough to match human curatorial abilities?
Industry veterans are skeptical. Lee, who has a decade of experience in trend forecasting and analysis, believes her work would suffer if she relied on AI, as it lacks insight into the cultural events and influences that shape fashion preferences. “With current technology, AI can’t fully grasp how events, socioeconomic factors, finances, and world politics impact trends and fashion,” she explains. “When you look beyond aesthetics, these are what truly drive trends. It’s not just about fashion—it’s everything surrounding it.”
So, can AI ever truly understand these real-world dynamics?
Input Limitations
AI’s biggest challenge in delivering tasteful outputs lies in its inputs. AI depends on datasets, and generic AI engines scrape vast amounts of data from across the internet, which can be noisy and unfiltered. Even AI tools built specifically for fashion face difficulties because fashion-related datasets often fall short, according to Yilu Zhou, an associate professor at Fordham Business School who has worked at the intersection of fashion and AI since 2013.
Zhou’s early research revealed that fashion taxonomy is poorly standardized. “Every designer speaks a different language. They might have two very similar designs but describe them in completely different terms—sometimes intentionally,” she says. The branding language embedded in product descriptions—like Haider Ackermann’s “glass effect” descriptor for a clear plastic blazer—poses a significant obstacle for AI in accurately predicting and interpreting fashion trends. Zhou emphasizes that standardizing data is the first step toward creating useful AI. “Otherwise, AI will be based on biased data, leading to biased and nonsensical outputs.”
Experts also note that the data AI relies on can be misleading, especially in fashion. “Something might go viral on social media, generating lots of shares, but that doesn’t necessarily mean people will buy it,” says Francesca Muston, chief forecasting officer at WGSN. Examples include 2023’s visible panty trend on the runway or the predicted bra trend for 2025—both were expected to generate more clicks than actual sales. Additionally, AI often fails to account for seasonality unless specifically programmed to do so. “If a data analyst doesn’t understand fashion, they might misinterpret a trend as gone when it’s actually set to return next year,” Zhou explains.
This is where human judgment becomes essential—to interpret the stories AI often misses. “Humans can better contextualize disparate pieces of information, connect the dots, and recognize opportunities for future trends,” Muston says. Zhou agrees, adding that human expertise is needed to identify when AI is off-track or “hallucinating” to fill gaps in incomplete datasets.
Purpose-built AI models, like Daydream’s, aim to offer a better foundation by aligning with users’ personal styles. Daydream’s Yamner recalls users’ reactions when introducing the platform: “They’ll say, ‘I tried the same search on ChatGPT and got horrible results.’” AI trained specifically on fashion data should, in theory, provide more relevant and accurate insights.Daydream uses its own brand mapping system to understand how brands connect in terms of style, aesthetic, and positioning. Combined with individual user signals, this allows the platform to recommend brands that feel both relevant and unexpected—without any advertising or paid placement influencing the results.
Programming Taste
Even if AI is trained with the right data, some doubt it can match human taste. “With the right training data, AI might approximate what individuals do in certain situations, but I think it will always lag behind,” says one expert. “As a person, your influences come from the street, chance encounters, and a wide range of sources—some digitized and available to AI, and some not. I doubt all the influences that shape taste, especially for someone with a strong aesthetic sense, are immediately accessible to AI.”
AI also tends to focus on broad trends that dominate social media or shopping headlines. But more interesting are the styles that emerge locally, which are harder for AI to detect, notes Zhou.
Programming AI to match a user’s specific style can also become too narrow and prescriptive, limiting discovery beyond one’s usual preferences. “If you don’t have a thousand dollars to spend on a bag, we won’t show you the thousand-dollar bag,” says Yamner. Yet many fashion enthusiasts have been inspired by iconic designs like Nicolas Ghesquière’s early-2000s Balenciaga City bag or Phoebe Philo’s Celine, even if they couldn’t afford them at the time. That inspiration still matters. While Daydream filters by price for purchase intent, it can use Philo’s aesthetic as a signal to surface pieces with a similar vibe at more accessible prices—a form of democratization important to the platform.
Similarly, focusing only on fashion brands and trends is limiting, since personal style is often shaped by other cultural areas. For example, WGSN’s trend predictions improved once they started tracking industries like food and sports. “People don’t just wear clothes. They also eat food, live in a house, use cosmetics, and engage with other areas like consumer tech or sports,” says Muston. Focusing solely on the product means missing much of what drives a trend.
Some founders believe AI can be developed to identify “good taste” as defined by human input. McCune of Cosmos thinks AI can learn taste with the right programming. His goal for Cosmos is to be an “anti-slop platform.” “AI can support creatives in areas like search and recommendation,” he says. Cosmos’s machine learning team built an “aesthetic prediction model” that determines what users see. It was trained on images saved by the first 10,000 beta users—including designers, creative directors, and architects—along with “really bad” datasets used as negative samples. Now, every uploaded image is scored against the aesthetic standard set by these samples.
“We set a bottom threshold, and anything below it is deprioritized in search and recommendations,” McCune explains. He emphasizes that it’s not about imposing a single taste but about elevating quality.Visual culture is becoming more uniform, but Cosmos doesn’t just highlight top-tier content. “We’re using it more as a baseline filter to remove the junk and low-quality material,” he explains. Machine learning has played a crucial role in Cosmos’s curation, offering a small-scale response to Zhou’s criticism of fashion’s ‘bad data’—though Cosmos isn’t limited to fashion.
Receipt-sharing app Selleb is also optimistic about combining AI technology with human oversight. Co-founders Chloe and Claire Lee see AI as a foundational tool that will ultimately be reviewed by people. Users share receipts not only for fashion, as the founders initially expected, but also for cafes, transportation, flights, groceries, and more. “Our broader vision is to map all these different products online and track everyone’s taste across various factors that get closer to capturing that elusive aspect of taste—something I still believe is hard to define,” says Claire.
Selleb emphasizes the importance of cross-category data to better understand a person’s taste, preferences, and style. New users connect their email and submit thousands of receipts. “Those receipts—when they were made, what categories they fall into, how much I spent—reveal a lot about me as a shopper and my unique identity,” Chloe notes. Users follow what the sisters call their “taste doppelgangers”: people with similar preferences across categories, based on the “taste graph” the app is building. By analyzing users’ receipts comprehensively, the backend AI can identify patterns that aren’t visible from publicly available online data, leading to personalized fashion and other recommendations.
Looking Back
AI predicts and identifies trends based on past data, meaning it cannot look beyond these inputs, no matter how advanced the technology becomes. “AI isn’t great with novelty—and trends often depend on novelty,” says WGSN’s Muston.
In reality, people’s style and taste evolve with changing contexts and cultural shifts that AI cannot anticipate. “Trends are highly complex and move in many different ways,” Muston explains. “How often have people said, ‘I would never wear XYZ,’ and strongly opposed a trend due to past associations? Yet, when that trend reappears in a new context, it suddenly becomes appealing.” If past data suggests a look is likely to fail, AI will take that at face value. Humans, however, can question the context and recognize why a comeback might be possible.
Interest in certain brands or aesthetics is often sparked by random or statistically unlikely events that AI cannot account for, says Madé Lapuerta of @DataButMakeItFashion. She cites a surge in interest in Van Cleef & Arpels last November when Dodgers player Miguel Rojas—who wasn’t even supposed to bat—hit a game-changing home run, winning the World Series. “Because AI-driven predictive models rely entirely on past patterns, they can’t foresee the future or understand what will resonate.”
This is Lee’s main concern with using AI to predict trends or anticipate changes in taste. “The way AI ‘predicts’ trends isn’t really prediction—it’s just reflecting what’s happening now,” she says.
The Human Edge
This human advantage is crucial. AI’s reliance on historical…The data shows that AI can identify content based on the ‘what,’ but not the ‘why.’ As Muston puts it, “AI can codify taste, but only in a synthetic way.”
Thompson Ford agrees that it’s too imitative. “It’s one thing to say, ‘I want to look like Ralph Lauren’s collection from last year’—AI might manage that. But if I want to look like someone’s collection this year that hasn’t even been created yet, I doubt AI can do what a designer does, or what a stylish person does.”
Even techno-optimists like McCune are questioning this. “By nature, models have to be trained on something from the past,” he explains. “Humans can look forward and create new trends and aesthetics. Models will always reflect the past—I believe only humans can truly look into the future.”
Experts suggest the only way AI could imitate this is if it gained sentience—a hotly debated possibility—and even that isn’t certain. McCune adds, “I believe generative AI will be able to cultivate taste and style, but it will be the taste and style of the now or the past. It won’t look into the future and create new things that feel on-trend.”
Lee, who is less optimistic about AI, agrees that its inability to look forward is a major limitation. For her, this means AI—without sentience—will never cultivate real taste or style. “You have to go outside, see what people are wearing, hear what they’re talking about, watch movies, listen to music, pay attention to current events and politics,” Lee says. “These are what shape fashion and style. It’s not just the clothes—it’s everything about you as a person. If you’re relying on AI to tell you who you are, you’ll never have style.”
Even if AI one day gained sentience and broke free from human input, it would still lack one thing: a human body. Without a body to operate from and to dress, cultivating taste and style seems nearly futile. “The one thing AI doesn’t have is a body,” Thompson Ford notes. “It’s hard to imagine AI developing the intuitions that come from moving through the world in your own body and interacting with others—except, again, through imitation.”
Lee agrees. “I’m sure it will improve, but AI will never be human. So it’s impossible, I think, to truly make sense of world events and how they translate into fashion,” she says. “I’ve been doing this for ten years, and sometimes even I’m wrong or behind on certain things. There’s no way a robot will ever be better at it than me.”
Frequently Asked Questions
FAQs Can AI Truly Understand Taste
BeginnerLevel Questions
1 What do we mean by taste in this context
We mean the complex human experience of flavor which combines smell taste texture temperature and even personal memory and emotion Its more than just a chemical analysis
2 Can AI taste food like a human
No not in the human experiential way AI doesnt have consciousness or subjective feelings It cant enjoy a meal or have a personal preference Instead it analyzes data about taste
3 So what can AI do related to taste
AI can process massive amounts of datalike chemical compounds in food recipes consumer reviews and sensory panel resultsto predict flavor profiles create new recipe combinations optimize food products and recommend dishes you might like
4 How does AI learn about taste
Its trained on datasets For example it might be shown thousands of recipes labeled with flavor descriptors or data linking chemical structures to perceived flavors It finds patterns in this data to make predictions
5 Are there any realworld examples of this
Yes Companies use AI to develop new snack flavors craft beer recipes or create personalized nutrition plans Apps like plantbased meat companies use AI to analyze molecular structures to mimic the taste and texture of meat
Advanced Practical Questions
6 Whats the main limitation stopping AI from truly understanding taste
The hard problem of consciousness and qualia Taste is a subjective firstperson experience AI can correlate data but cannot experience the sensation of sweetness or the nostalgia a flavor evokes It lacks embodied subjective awareness
7 Can AI account for cultural and personal differences in taste
It can try but its a challenge By training on diverse culturally specific datasets AI can learn common preferences within groups However capturing the deep personal emotional and cultural context behind an individuals favorite food is extremely difficult
8 What are the benefits of using AI in food science and development