Teaching AI taste: Training algorithms to understand aesthetics, not just clicks

New Delhi: We're used to AI being intelligent; of course, it is in the name. It can drive cars, build websites, write emails, fly rockets, create shows, and, if correct data sets are provided, can even predict the future. 

AI is learning to recognise faces, write poems, play chess, and whatnot. 

But it's a different story regarding something as nuanced as taste, what looks good, what feels right, and what resonates.

Most AI models are trained on the basic model of user interactions. They learn patterns based on what gets clicked, shared, or liked, which makes their learning simple, measurable, and scalable. But taste? That's a tough nut to crack. A post might go viral, but is it meaningful? A design might grab attention, but does it leave an impression?

This is where the gap lies—between what's popular and what's beautiful, thoughtful, or emotionally rich. Based on the data, AI may be able to recognize what's beautiful, but can it understand why something is beautiful and why it is different for different people? 

Clicks don't tell the whole story

Strictly in accordance with marketing's vision, we've been teaching AI to optimize for performance: more clicks, more time spent, and more conversions. It works well in areas like online ads or product recommendations, but aesthetic judgment doesn't fit that formula.

People don't always click on things they love. Sometimes, it's curiosity. Sometimes it's a habit. And often, the best work whether it's a piece of writing, a photo, or a design - isn't made to go viral. It is made to resonate emotionally with people. 

Yet, we keep training AI on surface-level signals. The result? AI has no clue about aesthetics and taste.

Taste is complicated, and that's the point

Here's the thing about taste: it's messy, subjective, and deeply human. It's shaped by our culture, symbolism, where we grew up, and what we've experienced. It changes with time and mood context. And it rarely agrees with the algorithm. It isn't quantifiable by clicks alone. It emerges from a complex interplay of the traits mentioned above and much more. 

Two people can look at the same design and have entirely different reactions. One might find it calming, and another might find it cold. One person's minimalistic style might be another's boring.

So when we say we want AI to "understand taste," we're really asking it to do something deeply human—to make sense of subtlety, emotion, and culture. That is not something you can get from just tracking clicks.

Teaching AI the 'Why,' not just the 'What'

If we want AI to grasp aesthetics, we must change how we train it. Instead of flooding it with spreadsheet data, we should expose it to work that's been curated by humans - the kind of content selected for quality, nothing else. 

Think about award-winning ads, editorials, museum collections, and playlists made by music editors. These are shaped by people who know what they're doing. 

We must train AI on multimodal datasets, pairing beautiful visuals with curatorial text or critical reviews. A good example is NIMA (Neural image assessment), which is trained on large datasets of images rated by humans. 

From creating to curating

Generative AI is improving at making things - images, videos, writing music. The real challenge is helping AI figure out what's worth keeping. It's like moving from being a painter to being a curator, from typing a sentence to editing a story. AI not only needs to know how to create a hundred versions but also how to say, "This one works. This one feels right."

Culture and emotion aren't one-size-fits-all

Another layer that AI struggles with is cultural sensitivity. What looks elegant in Paris might feel over-the-top in Delhi. A soft color palette might feel luxurious in one market and bland in another. Taste is local, emotional, and often personal.

That emotional layer is something current algorithms still find difficult to grasp.

Building AI that understands us — Not just data

To move forward, we need AI built with humans in mind—not just performance metrics. That means more collaboration between data scientists and creatives, more involvement from human curators, designers, and storytellers, and letting go of perfection in favour of intuition.

We're not trying to make machines that replace human taste. We're trying to make tools that understand it - and support it. That's a very different goal.

The best technology isn't the one that gets the most clicks; it's the one that understands people- their preferences, differences, and stories. It is the one that can help them create something that lasts.