What We Talk About When We Talk About AI

How we address AI in conversation and news coverage is at least as important as the technology.

Recently, the news publication and events company Semafor launched a morning newsletter into a crowded field. The Middlebrow is generally selective about which morning roundups get read because they are repetitive and often contradictory. Bloomberg’s morning “five things,” remains the top choice but Semafor generally does a good job. But this headline threatens to throw them out of my rotation:

My guess is that you see the problem write away — “AI Piaf to Narrate Her Own Biopic,” implies that somebody has made a biographic movie about an AI trained to imitate the long-dead singer Edith Piaf. But unless Spike Jonze is heading the project, we know that’s now what’s going on. The writer doubles down on this assertion with the line, “The robo-Piaf will narrate her own story…”

I don’t think I am being pedantic by pointing out that the writer is giving agency to the AI that does not and cannot exist. Of course, you can write a biography of an AI programmed to imitate a real person, in the voice of that AI, which is something that science fiction writers have been doing forever. But the writer here is implying that the real Piaf, through the medium of AI, is telling her own story from the wrong side of the River Styx, and that is nonsense.

I don’t mean to argue that there’s no artistic merit in using computing technology to reconstruct lost voices from history, but we should not pretend that we are resurrecting them or letting the dead speak. It’s a bit like saying that a comedian with a talent for imitation actually is their subject. “Gee, boss, you said it would be nice to have George H.W. Bush around to talk about his experiences dealing with Middle East issues from the 70s to the 90s, but he’s dead. I did get you a meeting with Dana Carvey...”

Specifically to this project, another way to film the biography of Piaf would be to use an actor. But nobody would ever write a headline like, “Emily Blunt to Narrate Her Own Biopic as Edith Piaf” because that would be very confusing even though all that is happening here is that a role that would be typically given to a human has been given to a computer program.

Inevitably, because it’s already been the subject of so much science fiction, people will create AI’s meant to model great figures of history. They will claim we can ask questions of Albert Einstein or Abraham Lincoln and people will fall for this. It will hopefully be obvious to people that any AI, subtly or not, will simply be using the authority of bygone people to push the agenda of its programmers. If AI-braham Lincoln starts commenting on current affairs, its users are going to have to ask who built the thing and why.

News media has a real responsibility here to cover AI for what it is, and here are some principles I think should be at least considered by writers and editors dealing with stories about it:

1. AI has no agency. It makes no decisions and it does not hold beliefs, think or possess reflective self-conscience.

2. AI is not a stand-in for real people, living or dead. It can imitate them for artistic or educational purposes, but that is mimicry.

3. All AI reflects the agenda of its creators, either subtly or obviously. It cannot achieve “objectivity.”

It is very easy to imbue AI with soul and magic. This should be resisted, lest we make ourselves marks for the Mechanical Turk of the 21st century.