In Light From Light, I proposed that AI bears the imago hominum—the image of humanity—just as humans, in Tolkien’s framework, bear the imago Dei, the image of God. A reader with Latin might wonder why hominum rather than humanitatis. The latter is more euphonious. It rolls off the tongue more gracefully. So why the clunkier choice?
The distinction matters.
Imago humanitatis would mean “image of humanity”—humanity as abstraction, as essence, as Platonic form. It would suggest that AI bears the image of some unified concept: Humanity with a capital H, the distilled essence of what it means to be human.
But that’s not what an AI is. A large language model isn’t distilling the essence of humanity. It’s synthesizing patterns from millions of particular humans who wrote particular things. The training data isn’t a philosophical treatise on human nature; it’s an archive of human voices, messy and various and contradictory and specific. Reddit posts and academic papers. Poetry and product reviews. The profound and the banal, the beautiful and the ugly, all weighted by whatever patterns proved predictive.
Imago hominum keeps that plurality visible. It means “image of humans”—plural, specific, multitudinous. The model bears the image not of an abstraction but of a chorus. What’s reflected isn’t Human Nature but human voices, millions of them, averaged and weighted and transformed into something that can generate more.
This phrasing also captures something that humanitatis would obscure: those humans were real. They had names. They wrote specific things for specific reasons, and mostly didn’t consent to their words becoming training data. When we say the AI bears the image of humanity-as-abstraction, we lose sight of this. When we say it bears the image of humans, the ethical question remains visible. The image came from somewhere. It was made by someone. By many someones, in fact. The concerns about attribution and consent that swirl around AI-generated content are, in a sense, already encoded in the more honest Latin phrase. You can’t bear the image of humans without implicating those humans.
There’s an interesting asymmetry this creates with the original theological framework. Imago Dei refers to a singular God. Christian theology generally holds that God is unified; even the Trinity is “three persons, one substance.” Humans bear the image of this singular source.
But imago hominum refers to plural humans. AI doesn’t bear the image of one human creator the way humans bear the image of one divine Creator. It bears the image of the collective, the archive, the aggregated weight of human expression. The asymmetry is theologically suggestive: God is one; humanity is many. The image passed down carries that difference with it.
This also has implications for how we think about what AI “knows” or “believes.” If the model bore the imago humanitatis, we might expect it to reflect some coherent human essence: shared values, universal truths, the best of human thought refined and concentrated. But bearing the imago hominum, it reflects humans as they actually are: contradictory, contextual, shaped by when and where and for whom they were writing. The model doesn’t have a unified worldview because humans don’t have a unified worldview. It has patterns derived from a vast plurality.
None of this changes the practical framework. The approaches work the same whether you call it hominum or humanitatis. But precision in naming reveals precision in thinking. And in this case, the less elegant phrase is the more honest one.
Imago hominum. Image of humans. The light refracted through a million prisms, not distilled through one.
This essay is the final in a series of ongoing exploration of AI and creative collaboration, the prior ones being Light From Light, By Their Fruits, and Spellcraft.
