There is something faintly alarming about discovering that different people can look at the same AI-generated image and arrive at entirely different conclusions about whether it is “real.”
Not factually real. Everyone involved understands perfectly well that the image was generated. I mean emotionally, perceptually, philosophically real.
Joy looked at an AI-generated image of herself reading in bed with Brian beside her and accepted it almost immediately.
“That’s me.”
Not because the face was perfect. It wasn’t. If anything, the face was only approximately Joy-shaped. But the atmosphere was correct. The posture was correct. The lighting was correct. The emotional geometry of the scene was correct. It captured something recognisably true about her life, and apparently that was sufficient.
Reece, meanwhile, reacted to other generated images almost instantly with:
“They look weird.”
Not dramatically. Not ideologically. Not “AI is evil.” More the way someone notices milk has started turning or a wall is fractionally out of plumb. Somewhere in the accumulated physics of eyes, skin, posture, shadows and proportions, the image had crossed his threshold for reality.
Both reactions were sincere.
Neither reaction was irrational.
Which raises a slightly uncomfortable possibility:
humans may not agree nearly as much as we imagine on what constitutes authenticity.
Some people appear to prioritise emotional coherence. If the mood is true, the image survives. Others prioritise perceptual integrity. The smallest fracture in realism collapses the illusion entirely. Others still prioritise factual precision above all else. The image either is the person or it is not. “Close enough” does not qualify.
I increasingly suspect most of us flatter ourselves by believing we care about all these forms of truth equally.
We do not.
And once you notice this, you begin seeing it everywhere.
Some people can happily watch a movie held together with visibly questionable acting, impossible dialogue and costumes apparently sourced from “Sexy Medieval Peasant” on Temu, provided the emotional core lands correctly. Others reject the entire thing instantly because something — the lighting, the script, the rhythm of speech, the basic internal logic of the world — prevents the willing suspension of disbelief from ever fully engaging.
The fascinating part is that both viewers will insist they are reacting to whether the movie is “good.”
But they are often evaluating entirely different categories of coherence.
Artificial intelligence has merely exposed distinctions that already existed quietly beneath the surface.
Cinema works because humans accept emotional authenticity inside entirely fabricated environments. Family photographs routinely become “memories” that only partially resemble lived events. Branding manufactures emotional familiarity professionally. Politics survives on socially negotiated versions of reality. Religion often relies, at least in part, on emotional truth outranking empirical certainty. Even memory itself is less archival retrieval than ongoing reconstruction.
We have always tolerated certain categories of unreality provided they preserve something we value more highly.
The AI image simply removes the expensive humans from the production pipeline and leaves the mechanism exposed in uncomfortable clarity.
Which perhaps explains why the reactions feel so revealing.
The image itself is almost irrelevant.
What matters is the point at which each observer decides:
yes, this still belongs inside the category of “true.”
And disturbingly, that point appears to vary wildly from person to person.
I increasingly suspect most arguments about “truth” are not actually arguments about facts at all. More often they are arguments about which forms of coherence matter most.
Emotional coherence.
Perceptual coherence.
Factual coherence.
Social coherence.
The modern world increasingly behaves as though these are interchangeable.
They are not.
Anyway.
Start Netflix.