In a recent interview, Ben Affleck said that “AI can’t create anything meaningful.” In broad strokes, his argument is that AI-written work is mediocre, that AI is not capable of building something meaningful from the ground up, and that it will ultimately settle into the role of a tool, like VFX. But if you dig a little deeper, there is also a more thought-provoking claim underneath it: that anything produced by AI will always carry less value than value produced and assigned by human beings.
The first argument has already been debated plenty. The second, though, feels like it touches something more fundamental—almost a basic principle of economics. Since I find that second line of thought more interesting, let’s reframe the question. Instead of asking, “Can AI create something meaningful?”, let’s ask: Can AI keep generating things that people want to read meaning into?
Kasane Teto, the diva born from a lie
Kasane Teto is famous as a Vocaloid-like character created as an April Fools’ joke. Someone on Japan’s 2ch posted a fake announcement claiming that a new Vocaloid had been released, along with a bundle of random character settings: 31 years old (self-proclaimed 15), a chimera, likes French bread, and so on.
But then something strange happened. People began to treat that lie as if it were real. There was fan art, songs, derivative works, and eventually Teto’s voice was implemented in UTAU, an actual singing voice synthesis engine. Later she was even officially released for a VSynth engine, and now she is arguably popular on a level comparable to Hatsune Miku. It is not much of a stretch to say that Teto became a real cultural icon while carrying the identity of a “diva born from a lie.”
Teto’s value did not come from whether she was “real from the beginning.”
That value emerged the moment people began treating her as real. So what about lies thrown out by AI? In other words, what about hallucinations accidentally produced by AI—a fake person, a fake interview, a fake debut article?
Let’s compare.
Whether the original source was human or AI matters to some degree, but it is not decisive. Once the seed begins to sprout, it is already a collective creation, and it starts to follow the grammar of existing cultural media.
“Human-assigned value is always greater,” really?
If we unpack that claim, it does seem to follow the logic of traditional cultural media: humans assign value. But does that value necessarily have to flow back to a human author? In Kasane Teto’s case, engines like UTAU and VSynth certainly helped bring the character to life, but conversely those tools functioned more as tools for expressing the character than as symbols of the character themselves. In that sense, Kasane Teto, though randomly generated, did not “replace” humans, but became a new kind of star competing on a different axis from them.
So now let’s look at Affleck’s “replacement” frame. The reason his argument sounds persuasive is that it assumes replacement in the strict sense of human artist = AI taking that exact place. To argue against that frame, one would have to prove that AI can perform every role a human can.
But what if randomly generated outputs follow the same cultural-media grammar as human creations, just on a different plane? There is no need to fall back on stale analogies like photography versus painting, or CGI versus physical sets. Kasane Teto did not replace human singers, yet she still became a competitor on another axis and grew into an industrial actor in her own right.
Now we arrive at the final link in the chain. Can AI-generated random outputs really have no meaning?
“AI converges toward the average, so it can never surpass the value of human-created meaning or art.”
Even if AI converges toward the average, selection does not.
Most cultural hits are not created through “one stroke of genius,” but through a process like this:
- Massive sampling (many attempts)
- Selection (fandom favoritism, meme reinforcement, algorithms, community consensus)
- Accumulation (remixes, parodies, derivative works)
Even if the generated material is bland, if you scale up sampling and apply weighting, something can still emerge from the tail. Much like evolution, out of countless mutations, one survivor becomes something meaningful.
That is exactly how Teto worked. She began as a fake, but people made a playful act of selection, and as that selection accumulated, an identity took shape.
What about AI hallucinations? If AI spits out a thousand fake idols, that is fundamentally the same process as random generation. If one of them becomes a meme, and the meme becomes a fandom, and the fandom becomes an IP…
AI can generate thousands of “fake settings” every second. Quite naturally, value comes not from who created it, but from who chooses to treat it as real. And people already know how to treat lies as real. They did in 2008, and they will in 2025.
If AI can endlessly generate things that people want to assign meaning to, then doesn’t that change the very definition of “meaningful creation” itself?