I was doing a little research into LLMs for the Egregores in Ersatz, and the first thing you might stumble across is an essay by everyone's favourite post-internet artist turned AI shill, Trevor Paglen, and Kate Crawford. They claim to use AI to expose AI, explaining how it works and revealing its supposedly shady origins...
Except it doesn't do that. It's fallacious nonsense and Paglen is a hack! I knew he and his ilk couldn't be trusted.
For their exhibition "Excavating AI", they displayed a facial-expression dataset called JAFFE, presenting it as some kind of flawed precursor to modern machine learning and LLMs. But that's simply untrue, JAFFE was never intended for that. It was a facial-expression dataset created for academic research into emotion perception and human responses to facial expressions. It was used in what researchers call a "toy world", a constrained environment for research, not something intended for deployment into the real world.
What's crazy is the way Paglen and Crawford retroactively redefine what JAFFE was. They treat it as a case study in poor taxonomy, criticising its limited set of labels and alleged biases as examples of the limitations and biases of today's LLMs. But that makes no sense because JAFFE was never claiming to produce taxonomy of emotions for machine learning in the first place!
They then go on to suggest that this sort of research laid the foundations for machine learning as a whole, asking us to imagine what these sinister researchers might have done if they'd had access to the internet... But again, that framing is totally fabricated! JAFFE consisted of images created with the consent of participants, who were paid by the university for this specific research project. It was not a dataset collected to build an LLM.
But guess what the biggest irony of all is? In Crawford and Paglen's conclusion that machine learning requires greater oversight, they themselves take the JAFFE participants' images, enlarge them to massive scale, and plaster them across gallery walls as part of a sensational spectacle for their art show... with no consent from the participants! No permission!
So we're meant to trust these people to guide us towards the proper, non-slop use of AI? To act as its ethical stewards? Meanwhile they're showing no regard for consent themselves, appropriating other people's images for their own project.
And Trevor Paglen has a new book out now. For all I know, those images could be in there too. And his whole art practice is built around finding "interesting" images and constructing grand narratives about surveillance, tech, corps, acting as he's lifting the secret veil of the world's hidden power maps. Funny, then, that "Excavating AI" was privately funded by Prada.
Most of this comes from reading Michael Lyons' responses to and criticisms of Crawford and Paglen's interpretation of JAFFE, which I highly recommend you read yourself.
https://arxiv.org/abs/2009.01215
https://medium.com/@michael.lyons_85617/indiscretions-of-a-contemporary-artist-88c9528a3ec1
https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://philarchive.org/archive/LYOEARv1&ved=2ahUKEwjSqe6Pj-GUAxVYWkEAHcAwHm4QFnoECFsQAQ&usg=AOvVaw2c94VK1SYl9mpLkrDRvxmR
I find them convincing and compelling and I struggle to see how you could argue in favour of Paglen stealing and using the JAFFE images without consent, which is probably why Paglen has made no academic or solid defense. Unless you count this.
But what is all this, why am I sharing this? Well... Not gonna lie, it's drama, innit. I was trying to learn about machine learning and instead came across this mess on a artist I already thought was sus. I had to share...
But what I really want to know is... what the hell is the actual deal with Paglen and his circle?
It seems to me that Trevor Paglen wants to position himself as a leading figure in AI criticism. By presenting himself as a critical researcher exposing the "hidden truths and history" of machine learning, he gains this academic legitimacy and cultural authority. I mean, it's working, a lot of people take him seriously. Plenty of people I follow seem to worship the guy. I feel like I'm getting gaslite here!
But to me, it feels like he and his milieu are using the prestige of fine art and critical theory to basically "theory-wash" AI. They're constructing a narrative where AI is acceptable, viable, even inevitable, provided it's developed in the "right" way by the "right" people.
There's the so-called slop AI, and then there's good AI, the ethical AI that people like Paglen know how to use, while dismissing your criticisms as uninformed or superficial. Yet they don't even believe in consent themselves! And when you actually dig into some of their supposedly critical writing, it's weak bullshit, just a set of fake grand stories and histories that they're trying to hyperstition into reality. At the end of the day, it's just about normalising Ai, they're looking out for their class interests, and that's clearly in line with these capitalists and corporate powers.