The Creativity Gap

Neil deGrasse Tyson has that apt argument against chalking up gaps in scientific knowledge to the power of God: “If to you God is where science has yet to tread, then God is an ever-receding pocket of scientific ignorance.” (He says something that that effect here around 2:08.)

I fear a similar argument around the conversation of creativity and machine learning. At one point, “creativity” was beyond the purview of machines. It’s what separated humans from practically everything else in the universe. Animals were governed by instinct and the spheres by laws of physics and randomness, but Man was made in the image of God and with that came the act of creation itself. Creativity is of God, and the scientific machine could never touch it.

Between the 1960s through the early 2000s, the painter Harold Cohen coded his program AARON, hooked it up to a plotter, and bid it paint. And yet this was not interpreted as a creative act on the role of the machine. It was the original coder that set the parameters that enabled the machine to create recognizable flowers and figures.

But what if you fed every human work from history into that machine and asked it to paint something, allowing it to randomize aspects within defined rules of compositions? We may refer to it as a “probability machine,” a visual auto-correct or predictive text, as we do now. But the provenance of the art piece is so completely blurred, we don’t know to whom to assign the creative act. Modern artists would say that’s because it was “stolen,” but that’s a topic for a different post. And some folks on the internet say it’s the prompt itself that is creative, but that feels . . . self-serving.

Creativity is a kind of magic of the human experience. We don’t know exactly how it works. Just read a Creative Act by Rick Rubin or Big Magic by Liz Gilbert, and you’ll see how they lean on mystical ideas of universe antennas and personal geniuses to communicate about art-making and creativity; accurately describing it is an act of creativity in itself.

So each time a large language model “writes” “better” music or “paints” images or adds motion, we’re stuck in the position of continually moving the goal-post, trying to define “true creativity” as something other than whatever the heck that was.

Is that the position we’re in now? True creativity is that receding extra bit of lived experience, embodied nature, ability to create connections that differentiates our works from machine learning? A spark of the divine? A human inheritance?

I’d imagine someone reading would disagree. They’d say no, it’s just a matter of a big enough dataset and electricity.