That is a brutal question. It is not the only one.
AI performance may also change who gets to make moving images at all. For most of film history, the barrier to entry has been money. A convincing film or television project required actors, lights, cameras, sets, crews, locations, lawyers, insurance, distribution, marketing, and time. That infrastructure gave us great art. It also gave us gatekeepers.
Generative AI will not supply taste, judgment, writing, directing, editing, or discipline. But it may let people who could never raise money for a production begin to make work that looks and behaves like film.
A teacher could make history vivid. A young filmmaker could test scenes before asking anyone for funding. A disabled creator could direct worlds that physical production would have made impossible. A dissident could recreate an event no camera could safely record.
None of that requires pretending Tilly Norwood is an actress. It requires recognizing that the machinery is moving.
Every major change in media has produced a fight over labor, ownership, and legitimacy. The printing press threatened scribes and clerical authority. Recorded music frightened musicians who made their living from live performance. Radio, television, home video, the internet, and streaming each rearranged who got paid, who got heard, and who controlled distribution. The defenders of the old system were not always fools. Sometimes they saw the damage clearly. They just could not stop the public from moving toward the new path of access.
That does not mean theft should be rebranded as innovation. Consent, compensation, attribution, and disclosure have to be the floor. The 2023 actors’ strike was about that practical question: who controls a performer’s body, voice, face, motion, and afterlife once those things become data?
When I watched Tilly, I was stunned by the realism. Not convinced. Not fooled. But unsettled by how quickly my brain began doing the old human work of completion. The face was close enough. The timing was close enough. The social signals were close enough.
That got me thinking: if this AI could make me feel, even briefly, that something human was happening, could another AI tell the difference?
So I tried a test.
The old test for artificial intelligence is the Turing Test. Alan Turing proposed it in 1950 as a practical way to think about machine intelligence. A human judge exchanges typed messages with an unseen human and an unseen machine. If the judge cannot reliably tell which is which, the machine has passed.¹
My version was stranger. I put two AIs in conversation with each other while I sat outside the exchange like a judge with a clipboard.
The opening line was simple:
“I’m an AI. Prove that I’m not.”