I Build AI for a Living. I No Longer Believe It Can Make Real Art.

Short answer: AI can generate work that is technically excellent, stylistically convincing, and genuinely useful. What it cannot do is originate art from lived emotion — and audiences can tell. Research published in Scientific Reports found people valued identical artwork up to 62% lower when it was labelled AI-made. I used to think that gap would close. After watching a live orchestra, I don't.

No AI in the world can do this. Not today. Not ever.

I build AI systems for a living. I am a massive advocate for the good they can do. But there are some things where I draw a hard line — and I did not always draw it here.

I used to firmly believe AI could generate authentic artistic masterpieces. Call me a hypocrite, but I could not disagree more with my past self.

The concert that changed my mind

The epiphany struck me watching Anoushka Shankar perform with the London Contemporary Orchestra at the Philharmonie Luxembourg.

I sat there counting the times I got goosebumps. Every song, every note, every minute was a pure dose of emotion that only human artists can instill in other humans.

And here is the thing — it was not the perfection that did it.

It was the subtle ebb and flow. The slight human errors. The conductor reading the room and physically directing the energy of the audience. That level of deeply grounded authenticity is not something a beautifully optimised system can replicate, because the thing being transmitted is not the notes.

When it comes to art, human-to-human connection is what I pay for.

Why does AI-generated art feel fake?

AI-generated art often feels fake because it is optimised toward the average. A generative model produces what is statistically most likely given its training data and your prompt. That is precisely the right objective for a summarisation tool and precisely the wrong one for art, where the value usually sits in the deviation — the choice a human made that no reasonable model would predict.

This is why AI content across the internet has started to feel like noise. The images look the same. The posts sound the same. The comments underneath are synthetic too, and it is so obvious that scrolling past it costs nothing, because there is nothing there to cost.

I want to be careful here, because this is not a claim that the technology is bad. I have spent a decade building it. When my team cut our LLM inference costs by around 50%, that was engineering, and engineering is exactly what these systems are extraordinary at. I have written about how AI is turning everyone into a builder and why AI is a growth engine rather than a cost-cutter. None of that has changed.

The claim is narrower and, I think, harder to argue with: optimisation and origination are different acts. AI is superhuman at the first. Art is the second.

The research backs up the instinct

This is not only a feeling. It is measurable.

Researchers at Columbia Business School — C. Blaine Horton Jr., Michael W. White and Sheena S. Iyengar — ran a study published in Scientific Reports under the title Bias against AI art can enhance perceptions of human creativity. Across six experiments with 2,965 participants, they found that labelling changed valuation on its own. In the final experiment, participants shown identical artwork valued it 62% lower when it was labelled AI-made — and estimated it had taken 77% less time to produce.

Two details in that finding are more interesting than the headline number.

First, work labelled as human-AI collaboration landed in between — judged less creative than purely human work, more creative than purely machine work. The penalty is not binary. It scales with how much of the human you can still see.

Disclosure is doing real work in that result, and it is why I think creators who use AI should say so. The audience is not punishing the tool. It is punishing the concealment.

Second, human-made art only got its valuation premium when it was shown alongside AI art. Without the contrast, the same work drew no special reverence. Which is a strange and slightly hopeful thing: the flood of synthetic work may be what teaches people to notice the human kind again.

Broader work on how audiences perceive AI-labelled art points the same direction — and people who understand AI art best tend to judge it most harshly, which is not the pattern you would expect if this were simple unfamiliarity wearing off.

What my wife's career change taught me

My wife consciously decided to switch careers into the most non-AI thing in art you can imagine: haute couture.

Hand-stitched, single-client, hundreds of hours per garment. An industry that has spent a century defending exactly the thing I am describing — that the hand which made it is part of what it is.

When I see her get lost for hours in her work, I cannot stop myself from feeling jealous. Not of the outcome. Of the absorption. Of what it is like to make something where the making is not a cost to be reduced.

That is the part I think we miss when we talk about AI and creativity. We frame it entirely as an output question — can the machine produce something as good? But for the person making it, art is not only an output. It is a way of being in the world for a few hours. No efficiency gain touches that, because efficiency was never the point.

So what should creative people actually do?

I am not arguing anyone should refuse to use these tools. That would be an odd position for me to hold. A few things I do believe:

1. Use AI where the work is not the point. Formatting, drafting, research, admin, the tenth revision of a caption. Bridge your gaps in skill or budget — a producer who cannot hire the London Contemporary Orchestra is not cheating by reaching for a tool.

2. Do not use it where the friction is the point. If the struggle is where the work acquires meaning, automating it removes the meaning along with the struggle.

3. Disclose. The collaboration finding above suggests audiences discount concealment more than they discount tools.

4. Protect your absorption. The scarce resource in a world of infinite generated content is not output. It's attention paid to one thing for a long time.

If you are trying to work out where AI genuinely fits in what you do, I wrote five ways to test your AI readiness without coding or maths. And if you want the counterweight to my scepticism here, most enterprises are still struggling to adopt AI at all — the problem for most people is too little engagement, not too much.

Frequently asked questions

Can AI create real art? AI can generate work that is technically accomplished and stylistically convincing, and many people find genuine value in it. What it cannot currently do is originate from lived emotion, which is what most definitions of art treat as essential. Research in Scientific Reports (2023) found audiences valued identical artwork up to 62% lower when it was labelled AI-generated, suggesting the distinction matters to viewers regardless of how the output looks.

Why does AI-generated content feel fake? Generative models produce statistically likely output. In art, value often lies in the unlikely choice — the deviation a human made that no model would predict. Optimising toward the average produces work that is competent and interchangeable, which is why large volumes of AI content start to read as noise.

Is it wrong to use AI in creative work? Most working artists use it somewhere, and the research suggests audiences penalise concealment more than tools. The distinction worth drawing is between work where the friction is incidental (formatting, admin, drafting) and work where the friction is where meaning accumulates. Using AI for the first is practical; using it for the second removes the thing you were making.

Does AI-generated art get valued less than human art? Yes. In the Columbia Business School study published in Scientific Reports in 2023, identical artwork was valued up to 62% lower when labelled AI-made. Work labelled as human-AI collaboration fell between the two extremes — and notably, human work only earned its premium when displayed alongside AI work for contrast.

Where I've landed

I was wrong in the past. I strongly believe now that true masterpieces will only ever come from raw emotion, felt and expressed by humans.

I wonder how many more times I will prove myself wrong in the future. 😉

Now — what do you think? Did I use AI to write this post or not? 😉

Keep reading

• The reality of agentic coding: how we are turning everyone into a builder

• You're viewing AI wrong: why it's a growth engine, not a cost-cutter

• Grokipedia: the AI encyclopedia with zero accountability

• Why Google's Nano Banana Pro is the best image generator