The more I’ve gotten into especially Midjourney v5, but also AI-generated images in general, the less useful and relevant I think the term “deepfakes” is any more. To me, that term is forever tied to a sort of early alpha version of what we now know as generative AI that mostly has to do with things like face-swapping in digital images and video. I just don’t find it terribly descriptive anymore for the current wave of uses that goes well beyond that.
Likewise, I never liked the even more vanilla term adopted by some quarters of industry, “synthetic media.” I get that it goes to a wider range of provenance and use cases than deepfakes, and cuts out the judgemental part of that older terminology. But I just never vibed with it.
“Generative AI” I like in general to describe the field. I plugged the 40 item request idea from Simon Willison and landed instead on the term “generative media” to describe inclusively, text, audio, video, images created by generative AI models. “Synthetic” still feels somewhat negative/judgemental to me, but I feel the word generative has positive implications: something is being generated, created, etc.
Likewise, I asked ChatGPT for 40 names for a subcategory of generative media which attempts to depict real persons, places, events, but which have no correspondence to actual things that took place in “real life.”
I’m still wrapping my head around the theory of it according to the semioticians, but it seems like simulacrum kinda points toward this distinction:
French semiotician and social theoristJean Baudrillard argues in Simulacra and Simulation that a simulacrum is not a copy of the real, but becomes truth in its own right: the hyperreal. According to Baudrillard, what the simulacrum copies either had no original or no longer has an original, since a simulacrum signifies something it is not, and therefore leaves the original unable to be located. Where Plato saw two types of representation—faithful and intentionally distorted (simulacrum)—Baudrillard sees four: (1) basic reflection of reality; (2) perversion of reality; (3) pretence of reality (where there is no model); and (4) simulacrum, which “bears no relation to any reality whatsoever”.[7] In Baudrillard’s concept, like Nietzsche’s, simulacra are perceived as negative, but another modern philosopher who addressed the topic, Gilles Deleuze, takes a different view, seeing simulacra as the avenue by which an accepted ideal or “privileged position” could be “challenged and overturned”.[8]
All that is a little too heady and complicated for everyday use, however.
Some of the better terms ChatGPT came up with here were:
- Virtual likenesses
- Artificial renderings
- Digital fabrications
- Virtualized representations
- Computational depictions
- Synthetic reenactments
- Generated replicas
- Constructed portrayals
- Simulated scenarios
- Invented visuals
- Algorithmic simulations
There’s probably some combination of the above which would be appropriate, depending on the exact use case. Maybe “Simulated portrayal”? Will sit with these more, but am thinking in terms of what might make sense as a label for something like Trump & Hillary kissing.
Perhaps:
Generative media:: Simulated portrayal
Tim B.
seeing it on the page:
`Generative media:: Simulated portrayal`
The first part seems to tell me:
– How it was made (using generative AI models)
The second part tells me:
– Its relation to reality (or my claim about its relation, anyway, which may be easily falsifiable)