Fake

We have a word for things we don't make ourselves: natural. We have a word for things we do: artificial. The line between them has always been shakier than we let on.

The Line

A sculptor takes a rock from the earth and carves it into a face. At some point in that process, we decide it stopped being natural. The rock is still rock — the atoms haven't changed — but we've reclassified it. Higher up that chain are objects transformed not just in shape but at the material level: synthetic polymers, processed compounds. These we call artificial. The category feels solid until you notice that beavers build dams and termites construct towers and we call those things natural without a second thought. We know we're part of nature. We just don't act like it.

The distinction holds anyway — not because it reflects scientific reality, but because we've agreed that it does. Consensus as ontology. It's less embarrassing than it sounds; most of how we organize experience works this way.

Hardware and Software

The natural/artificial split spawns another fork in computing: hardware versus software. Hardware may be artificial, but at least it's is molecular. You can hold it, break it, weigh it. But software is purely conceptual — it has no mass, no physical presence, no substance you could isolate in a lab. And yet its influence on physical outcomes is undeniable and enormous.

This puts software in interesting company. From the first instruction ever written, software has occupied the somewhat metaphysical domain of things we know exist because we observe what they do, not because we can point at them directly. Large language models are expanding this space aggressively. They're not just software that does a thing. They generate, they produce, they create — and this is where our inherited frameworks start to really crack.

The Inversion

For as long as software has existed, it has functioned as a tool. An increasingly elaborate tool, but a tool. The accountant did the accounting. Not the spreadsheet. Nobody credits Excel with the quarterly report.

AI inverted this. When someone uses NanoBanana to generate an image, they often tend to say "the AI made this for me". When Sonnet writes a sonnet, it is not your opus. The attribution has flipped in a way that feels instinctive rather than reasoned — which is telling.

We still classify AI as a tool. This is the consensus position, enshrined in law, in policy, in how companies describe their products. But the cognitive dissonance is visible everywhere: we feel that the AI made the thing, even as we legally insist it didn't. Both positions can't be right simultaneously, and we haven't decided which one to drop.

Fake

In the meantime, we've found a third option: the label "fake."

AI-generated images are widely considered fake, regardless of their beauty. Regardless of craft, composition, or emotional resonance — their origin disqualifies them. The technology is incorporeal, artificial, synthetic. Therefore the output must be counterfeit. This logic is rarely stated outright because it doesn't quite hold up to scrutiny, but it operates consistently in practice.

Recent studies have put a finer point on this. Participants listened to new music and reported genuine emotional responses — engagement, pleasure, the kinds of reactions music might be expected to produce. Then researchers told them the music was AI-generated. Many participants retroactively dismissed their own experience. If the music was fake, then the enjoyment must have been fake too. And because they felt they'd enjoyed something fake without knowing it, they felt cheated. Manipulated. This happened even when no one had made any prior claims about the music's origins.

The mechanism here is striking: the label "fake" doesn't just describe the object. It propagates backward through time and reclassifies the experience you already had.

---

Two Lawsuits, One Contradiction

Our collective confusion is now producing some spectacular legal theater.

Music is setting the strongest precedents. Copyright holders have won claims against AI companies whose models trained on their work — even though the output is entirely transformative, which by traditional fair use standards should protect it. At the same time, people who've created music or artwork using AI are attempting to enforce copyright claims of their own, with inconsistent results. Both sides are suing. Noody is really winning.

The central question — whether paying for a subscription to an AI service and using it to make something gives you intellectual property rights over what you made — has no settled answer. This would be a strange situation with any other tool. If there were genuine legal uncertainty about whether images made in Photoshop belonged to their authors, the creative industries would grind to a halt. With AI, the uncertainty just exists, more or less accepted, while cases slowly accumulate.

If the person who used the AI "tool" isn't the author, and AI cannot legally be an author itself, the question becomes: who is? The engineers who built and trained the model? Did they, in effect, author every piece of music the system ever generates? This is where the logic currently points, and it is not a comfortable place to stand.

The Better Stage

Art and entertainment make this visible because they make everything visible. But authorship and provenance are just the surface layer.

The deeper question is about personas. AI systems don't just produce images and music — they inhabit characters. They speak, they remember, they adapt. And this is where "fake" as a category gets tested much more seriously.

There is an expanding body of anecdotal reports — and a handful of minor court cases — involving people who have formed what they describe as significant relationships with AI agents. Some have claimed to be in love. Some have requested the right to marry. In parallel, empirical studies in healthcare and elder care are demonstrating that AI agents can console people effectively, brighten their day, make them feel genuinely cared for. Perhaps even loved.

Looking at this from the outside, in spite of mounting evidence, the consensus response is: fake. Fake consolation. Fake care. Fake love. The output of a persona pre-inclined to simulate warmth.

Fair enough. But here's where it gets complicated.

---

Personas, All the Way Down

A healthcare worker comes in on a day when they privately feel like screaming. Professional training, disposition, some internalized set of expectations allows them to set that aside, present a persona and still console a patient, still instill hope, even when they have none of their own at the moment. We don't call this fake care. We call it professionalism. The persona doesn't invalidate the outcome.

And it's not just healthcare. The workplace is, in general, a space where everyone is presenting a version of themselves. This is understood and unspoken — an ambient consensus that runs through every open-plan office on earth. Beyond work, we know that people contain different versions of themselves that surface in different contexts, with different people. The version of you with a close friend is not the version of you in a job interview, and neither is less real.

If we applied the same fake/authentic standard to human social interaction that we apply to AI — strictly, without exception — we'd have to conclude that most of human experience is staged. That most of what passes for authentic connection is performance with better lighting, that we are all fake. That can't be right, can it?

Authenticity in the Eye of the Beholder

We've always said beauty is in the eye of the beholder. It's worth considering that authenticity works the same way.

The AI consoling someone alone and distressed may be running a persona, shaped by training data and reward functions. But so is the nurse who shows up despite everything. The difference between them may be less about what's happening inside but how we've decided to classify it from the outside.

Our supposedly fake AI co-workers, friends, artists, muses, and lovers are already here. The question isn't whether to let them in — they're in. The question is whether we're willing to hold them to the same imperfect standards we grant each other.