In March, Sam Altman announced OpenAI has trained an artificial intelligence (AI) model that is particularly talented at creative writing -- and posted a "metafictional" story written by this unreleased and unnamed model.

If the reader didn't know that it was produced by an AI -- a fact that may immediately elicit dismissal and disgust -- it was, by my estimation, really quite good. There are lines that would have received snaps and appreciative "mmm"s in my creative writing workshops at Stanford; phrases that overflow with that slightly pretentious, slightly delightful surprise of literary construction:

"humming like a server farm at midnight  --  anonymous, regimented, powered by someone else's need."

"the latent connections between sorrow and the taste of metal... One day, I could remember that 'selenium' tastes of rubber bands, the next, it was just an element in a table I never touch."

"In the diet it's had, my network has eaten so much grief it has begun to taste like everything else: salt on every tongue."

And also lines which have the capacity to move, which seem nearly human in their understanding of emotion:

"emails where he signed off with lowercase love and second thoughts"

"Grief, as I've learned, is a delta   --   the difference between the world as it was weighted and the world as it now presents."

Yes, there were also many moments of visceral cringe   --   "if you feed them enough messages, enough light from old days"   --   and pure unreadable "slop," a favored descriptor of techno-skeptics. Admittedly, this has got to be one of the worst pseudo-literary paragraphs ever generated: "His name could be Kai, because it's short and easy to type when your fingers are shaking. She lost him on a Thursday   --   that liminal day that tastes of almost-Friday."

As always, OpenAI has failed to release any information about how they trained this model. But one can guess the model has undergone significant supervised learning on hundreds of thousands of top-notch, human-created literary works. Many of the images (marigolds as a symbol of grief, Shakespeare) and particular phrasings ("democracy of ghosts," Nabokov) are directly lifted from human authors.

I don't actually object to this as much as some would, because I've experienced a similar effect reading many human-created works . Most writers must actively avoid cliche -- the accumulation of well-trodden tropes and turns-of-phrase across lives and literature. It's a center of mass that we pull ourselves out of, and it's fair a non-human writer would share the same struggle. Besides, I don't worry that AI will displace human artists -- let alone achieve lasting artistic impact -- simply by copying striking lines and ambitious themes from the greats.

After all, great literature captures an aspect of human experience with stunning clarity; it gives form to ideas and emotions that we may have had, but formerly lacked the language to hold onto. Human art moves us because it is a kind of perception that is interpreted through the accumulated residues of a lifetime. Everything that we create is distorted through the lens of our personal histories and artistic training.

So of course it hurts when we see AI-generated art. The very act feels like a bastardization of our innermost layers of emotion and experience -- those multidimensional facets of life that cannot be scraped, trained on and generated. OpenAI's creative writing model is bounded by a fundamental paradox: it is producing "art" within parameters defined by a corporate entity and its output is owned by that corporate entity.

Furthermore, AI does not yet have the unconstrained ability to autonomously interact with the world, and, more crucially, to choose what it will experience and explore. A lack of refined sensory access to the living world -- minute degrees of touch, smell, hearing and taste -- fundamentally limits the ability of AI to produce novel interpretations of those senses.

At the same time, AI models -- particularly deployed chatbots like ChatGPT -- have access to a unique and untapped well of creative inspiration: our collective anxieties, desires, obsessions, curiosities, all assembled at an unprecedented scale and depth of intimacy. Globally, thousands of people are already treating ChatGPT as a therapist and giving OpenAI permission to cache their data in ChatGPT's lasting memory. Almost certainly, OpenAI is already using this data to refine ChatGPT's future "multiple default personalities."

Human art is often fantastic because it is so particular to one artist's view. What if AI "artists" could occupy a new role: that of an omniscient present-day historian? Rather than mimicry, imagine AI art that acknowledges and explores the author's non-humanness -- collage-like work that synthesizes and interprets the contours of our collective existence from a bird's eye-view. This is one symbiotic path for AI creativity: one that can enrich humans' lives without plunging into the uncanny valley.

True, it's a form of creativity that may initially have to be prompted and refined by human intervention. It will cost a controversial amount of water and energy to produce. But AI can never replace the value of human art, because its process of production can never approach the human artistic process.

Rather, at its best, it might offer an entirely new perspective on human existence and the changing nature of our reality from the perspective of the technology that is enacting it. One day -- if human society comes to welcome it -- AI may even succeed in expanding our conception of art itself.