The Future is Fake

I have had many conversations in my career about aesthetic and taste. People tell me they do not care how something looks, just how it functions. My rebuttal for this is typically “What type of car do you drive?”^{signal 1} And if they do own a car, a follow up question “So tell me how you chose your car and why?” These conversations typically unlock a bit more of the understanding especially when I point out other options, typically cheaper, that “functionally” could do the job for them.

We all make judgements about items, services and objects. We make snap judgements on people based on their clothing^{signal 2}. We judge these on social and societal norms.

We all have a modicum of taste imparted on us. And some of us pursue refining that taste^{signal 3} as a personal and/or professional education. Designers and aesthetic are intricately interwoven. You simply cannot have design education without the discussion. But “taste” is not something that is directly taught. It is something that your education influences.

As a designer, taste has four parts.

• You see what counts as good in a particular world.
• You use those signals with intent.
• You sense the new signals.
• And if you are very, very good, you move the signals yourself.

Notice that none of the four requires being original. Sometimes the tasteful call is the expected one. (Something I will come back to later on.)

Most people do not think of taste this way. They think of it as personal preference, a favourite colour but with better rationale. Because they file it under “subjective,” they treat it as decoration: pleasant to have, impossible to argue about, and not really a skill. That belief costs more than we realise. It shapes how we hire designers, how we value craft, and, increasingly, what we believe a machine (ahem, AI) can do in our place.

A serious case against taste as a skill was made by the French sociologist Pierre Bourdieu^{signal 4}. In his 1979 book Distinction, he studied how people in France chose their music, their food, their furniture and their art. He concluded that these choices were not really about beauty at all. They were about position. What we call having a “good eye”^{signal 5} is largely something we inherit from our upbringing, our schooling and that which is around us. His most famous line puts his thesis neatly: “Taste classifies, and it classifies the classifier.”

He is right. Every choice you make is a signal: the flat white rather than the filter coffee^{signal 6}, the tech-bro navy Patagonia vest^{signal 7}, the tote bag from the right bookshop^{signal 8}. Each one tells other people where you stand, and they read it whether or not they know they are reading. Anyone who works in advertising already knows this, because it is the entire business. When we apply judgement to using these signals on purpose, for a company rather than for ourselves, we call it branding^{signal 9}.

(Bourdieu would also have something to say about this essay. Later on I will call some design work “beige.”^{signal 10} That is itself a move in the great game he described. I am not exempt. I deeply realise this.)

But, I digress, Bourdieu stops too soon, in two places.

The first is quality. If taste were nothing but a badge of class, there would be no such thing as better design, only design that belongs to more powerful people. I do not believe that, and I doubt most designers do either.

Some things genuinely work better for human beings:

• type you can read at a glance
• a hierarchy that tells the eye where to go
• colours with enough contrast for someone with poor eyesight
• a signal that tells the truth about what a product truly is

Those hold whatever (mostly) your background. Other things are purely social:

• which serif reads as “premium” this year
• which shade of green says “sustainable”
• which bookshop’s tote bag you carry (or don’t)

Quality is partly real and partly social. Taste is the ability to tell the two apart, and to know which one a decision rests on.

The second is foresight. Here a harder critic stands behind Bourdieu. More than a century ago, the German sociologist Georg Simmel argued that fashion runs on direct imitation. The elite adopt a style, everyone else copies it, and the elite move on. On that view, “seeing ahead” is not foresight at all. Powerful people’s calls come true because powerful people made them. There is truth in that. There is also survivorship bias… we remember the calls that came true and forget the rest.

But if foresight were only power, style would only ever flow downward, and it does not. Streetwear grew out of skate parks and hip-hop, among people the fashion establishment ignored. By 2018, one of its leading figures, Virgil Abloh, was running menswear at Louis Vuitton^{signal 11}. The people who saw it coming were not the ones with the most status. They were the ones paying the closest attention.

Here is what I am claiming. Taste is not prophecy. It is a better-than-average bet, made for reasons you cannot explain before the fact, and corrected when it fails. And it can be learned, by people who did not grow up anywhere near the rooms where it supposedly lives. They learn it the way a sommelier learns wine^{signal 12} or an editor learns a sentence: through years of looking closely, being wrong, and paying attention to why. They are forever students.

That is the difference between taste and preference. Preference is what you like. Taste is knowing why you like it, what it will say to others, and whether it will still be right a year from now. Preference is a feeling. Taste is a judgement built out of thousands of feelings that have been tested, corrected and sharpened.

You cannot see what you have not been taught to see. Put a first-time user and a UI designer with twenty years of practice in front of the same app screen, and they are not having the same experience. The designer sees the hierarchy, the slight tension in the spacing, the point where the type weight shifts and the eye loses its anchor. The user sees a screen. They feel every one of those choices. They just cannot name any of them.

I have my own blind spots. Send a goth into a club^{signal 13} and she will read the room in seconds: who is a purist, who is a tourist, who has been coming for fifteen years, who bought the boots last week. I would see a room full of people in black. I would respect it. I just could not read it. I have never been part of that world, so I do not have the judgement. I do not have the taste.

This shows up in the lab, too. In a 1993 study, the psychologist Calvin Nodine and his colleagues tracked the eye movements of people looking at paintings. Untrained viewers went straight to the obvious thing: the figure in the middle, the object in front. Trained viewers wandered into the background, across the edges and around the composition as a whole. Studies of brain activity point the same way, with experts and novices using noticeably different patterns of processing. Cognitive scientists have a name for why this happens: the information-reduction hypothesis, set out by Hilde Haider and Peter Frensch in the 1990s. As you become expert, you learn to ignore what does not matter, which frees your attention for what does. Researchers who study aesthetic sensitivity have gone further and suggested it is a kind of intelligence that standard IQ tests do not capture.

Put plainly, the way we see is learned, and the learning leaves traces you can measure.

These studies show that experts see differently. They do not show that experts have taste. A skilled technician can spot every misaligned pixel and still make something lifeless. Trained seeing is the raw material. Taste is the judgement on top of it: which of those details matters, for whom, and why.

Some of it arrives without our asking. In 1968 the psychologist Robert Zajonc showed that simply seeing something again and again makes us like it more. Hundreds of studies since have confirmed what he called the “mere exposure effect.” None of us chose our first preferences. The songs in our parents’ car, the furniture in our grandparents’ front room^{signal 14}, the logos on the high street: exposure shaped them all.

But exposure gives you preference, not taste. Taste takes deliberate work on top.

Executing acts of design alone will not do it. Plenty of people have done the same job for twenty years and have one year of taste, repeated twenty times. You build taste by studying the gap between your work and the best work in your field, and closing it, which is slow and often uncomfortable. You build it through long immersion in a world, through real curiosity about the people in it, and through study rather than consumption.

The designer and writer Elizabeth Goodspeed made a related point in 2024: developing taste takes a kind of vulnerability. You have to trust your instincts when they cut against the trend and against your peers. Having taste looks cool. Building it is earnest and a little uncool, because it means committing, in public, to what you actually believe is good.

This is where I think much of our industry goes wrong. A great many designers in tech, particularly in the Silicon Valley mould^{signal 15}, are not building taste. They are trying to look as if they have it. They consume a lot and study very little. And they ship work I can only describe as beige. Non-offensive lifeless work.

You have probably worked with some of them. They treat the company design system as a sticker sheet^{signal 16}, peeling off components and pressing them onto screens. The problem is not the design system. Design systems exist for good reasons: consistency, speed, and the trust that comes from things behaving the way people expect. The problem is that these designers cannot tell you why a component looks the way it does, or when it would be right to break it. Many have no real grasp of colour or typography, and little interest in the people who will use what they make. They do not invent anything. They reassemble what they have already seen.

Sound familiar?

AI tools can now produce images, layouts, copy, code and whole interfaces in seconds. The results are often competent and sometimes impressive. The floor has risen, and that matters.

But what these tools produce is a statistical blend of what they have already seen. A model does not choose a typeface because it understands what that choice will mean to a particular audience at a particular moment. It knows the prompt, and it knows the past.

The obvious reply is that so do we. Everything I have described so far, the exposure, the study and the immersion, is a record of what already exists. Every new idea is a recombination of old ones. If the model is a remix machine, so are we.

That is fair, so the difference has to lie somewhere else. I think it lies in two places. The first is what each does with the past. A model is built to produce the most probable answer, the middle of everything it has seen. A designer with taste is often betting on a less probable one. The second is who is placing the bet. The designer lives inside a community, has a reputation to lose, and has to live with the result. The model does not.

It helps to think of taste in three phases:

Models are extraordinary at the third phase, because the most probable answer is usually the one everyone has already seen. For a short while after training, they can touch the second. The first is where they struggle, and the first is where taste earns its keep. It is the early, unproven commitment behind great work and great brands.

To be fair, a model can produce something improbable. Ask it for strangeness and you will get strangeness. What it cannot do is know which strange thing matters, or stake anything on it. Novelty is cheap. Knowing which new thing the moment is ready for, and backing it, is taste.

There is a deeper problem. A model learns what is common, and what is common mixes both kinds of quality together. Readable type and this year’s fashionable serif arrive in the same training data, weighted by how often they appear. The model cannot tell which is which. That is exactly the distinction taste exists to make.

Left to itself, then, AI output settles in the middle. It is the high-fructose corn syrup^{signal 17} of design: cheap, engineered to please on first taste, and empty. Bourdieu had a sharp description of middle-brow taste^{signal 18}. It gives away the effort behind it. It is polished, correct and eager to please, and it has no point of view. Breaking a rule well requires knowing why the rule exists, and that knowledge comes from experience, not from averages.

The result is beige.

The designer who consumes instead of studying and the model that matches patterns instead of judging are running the same process on different hardware. Both produce the look of design without the judgement underneath.

There is a harsher irony. These tools make middle-brow work easier to produce at scale, while making it harder for people to develop the taste to go beyond it. Taste grows through a loop: you make something, judge it, fail, adjust and make it again. Automate the making and you break the loop.

Give everyone a design system and a machine that generates layouts, and you do not get designers. You get people who can operate tools, and the gap between those two is enormous. Real democratisation would mean giving more people access to feedback, mentors and deliberate practice, the things that actually build taste. It would not mean handing them a machine that skips the learning and delivers a finished artefact, like a factory.

Access to tools is not the same as access to capability.

Now I need to complicate my own argument, because beige is not always a failure.

Remember Zajonc: familiarity produces liking. People trust what they recognise. In banking, where I have spent much of my career^{signal 19}, that matters enormously. Nobody wants the payment screen in their bank’s app to be avant-garde. They want it to look and behave the way they expect, every single time. They want it readable, predictable and, in the right places, boring.

The industrial designer Raymond Loewy had a name for this balance: MAYA, “most advanced yet acceptable.” Push too far ahead and people reject the work. Stay too close to the familiar and it disappears. Much of the best design in the world sits deliberately near the familiar end.

The payment screen is boring for real reasons: people need to read it quickly and trust it completely. A startup homepage that looks like every other startup homepage is boring for social reasons: it copies what is fashionable. Both are beige. Only one was chosen.

So taste includes restraint: knowing when not to be original. The problem is not beige. It is unchosen beige. A designer with taste picks the convention because it is right for this audience at this moment. The sticker-sheet designer and the model arrive at the same place because it is the average. The output can look identical. The difference shows up the next time, when the convention is wrong and only one of them can tell.

This is also a concession. If a great deal of commercial design value lies in doing the familiar well, then AI will do a great deal of that work, and do it well. Designers should stop pretending otherwise. What the machine cannot do on its own is tell when convention is the right answer and when it is cowardice. That call belongs to whoever has to live with the result.

The first two parts of taste are seeing quality and using it deliberately. The last two, anticipating where it is heading and moving it, are where the gap between people and machines is widest.

If you truly understand the world of the people you serve, not their “pain points” but their actual lives, you start to sense where things are going before the data can tell you. By their lives, I mean their values, their worries and the forces shaping their days. You feel the shifts that will change what they need, how they want to be spoken to, and which signals will land with them next year and the year after.

When I led design for Intuit’s accountant business^{signal 20}, I spent a lot of time teaching executives and product managers about the culture of accountants. Jonny Jiang, a design researcher, and I developed what we called Accountant Understanding. It went past empathy. The aim was to grasp how accountants actually think and see the world.

I would fly into California from London^{signal 21} for meetings of our accountant advisory council, and I was usually the only person from Intuit’s product or design teams still sitting with them when the hotel bar closed^{signal 22}. We talked about the company, the products and their jobs. It was immersion in their lives, not just their profession.

The real data was at the bar, not in the dashboard.

That is the difference I was reaching for earlier. The model has more of the past than any of us. What it does not have is a seat at the bar.

It matters, too, because taste does not live in one compartment. Food, music, clothes and politics are all connected. Understand those connections well enough in the people you serve and you can see how they are likely to shift together.

Notice which half of quality moves. Readable type will still be readable in ten years. What changes is the social half: which signals feel fresh and which feel tired. Anticipation is mostly about that half. Craft holds the real half steady while taste reads the social half as it shifts.

A designer with real taste can feel when a visual language has gone mainstream and is about to lose its power. They know how to wrap something new in a form familiar enough to be accepted. They can sense when an audience is ready for something it has not seen yet. They anticipate rather than react, and then they make work that moves the culture on.

This is what I mean by using taste strategically. Different audiences have different logics, and each demands a different kind of respect. That respect is empathy working at a strategic level. It is not the empathy of “I understand your problem.” That is the bare minimum, and it is roughly what Design Thinking^{signal 23} reduced empathy to. It is the empathy of “I understand your world well enough to know what you will need before you do.”

This is where AI is weakest today. A model learns from what people have already done and written down. The pressure building under the surface has not been written down yet: the fatigue, the restlessness and the appetite for something different that come before every real shift. Sensing it means being in the world. It means moving through real places and having relationships that show you how people feel, not just how they click.

And a machine cannot move culture on its own, because moving culture means committing to a direction before there is evidence it will work. A model can generate the option. It takes a person to choose it, defend it and stake a reputation on it.

Models will get live data and tighter feedback, and some of what I have said about them here will date. What will not date is the question underneath: who carries the consequences of the call? A model does not lose its reputation when a bet fails. It does not have to face the accountants at the bar the next morning.

Put the pieces together. Pattern recognition without empathy is just statistics. Add empathy and you get taste. Add a sense of time and taste becomes foresight. Add the nerve to commit and you get the thing that moves culture.

Twenty years in design^{signal 24} have taught me that the people who do great work are the ones who have developed judgement. I do not mean opinions. I mean the ability to make a call about quality when things are unclear, defend it with reasons, and change it when new information arrives.

That judgement comes from the slow work of building taste. It comes from studying what makes great work great, and from making things and watching them fail. It comes from honest feedback from people better than you, and from sitting with how that feels. It comes from time spent with the people you design for, learning their world rather than listing their “user needs.”

Taste is deciding what good means for this audience, in this context, at this moment. Sometimes good means new. Sometimes it means familiar. Quality is partly real and partly social, and the machine cannot tell the two apart. Hand that decision to a machine trained on the average of everything that came before, and you are not optimising. You are surrendering.

So the question is not whether to use AI. It is what to refuse to hand over. Keep the choosing. Keep the judging. Keep the moving. Build the taste that lets you tell the difference between what the machine gives you and what people actually need.

The machine is already brilliant at what everyone has seen. The job worth keeping is seeing what, for now, only you can see.