[ recap ]
Last week we looked at making the familiar strange: Viktor Shklovsky, the Nacirema, and what living in Nairobi taught me.
|
[ thoughts ]
On Tuesday, Meta launched Muse Image, an AI image generator with a feature that let any stranger tag a public Instagram account and use its photos as material for AI-generated images, without account owners being told.
Within minutes, activists were publishing opt-out guides, and by Friday the feature was gone, with Meta only saying that "we've heard the feedback that this feature missed the mark, so it's no longer available."
Meta opted everyone in automatically, and many of the people opted in want nothing to do with AI at all.
Around 1900, an orphaned teenager named Abigail Roberson had her portrait taken at a local studio for her own use. Some time afterwards, visiting a neighbour, she found that same portrait on an advertisement for Franklin Mills flour, one of 25,000 posters hanging in stores and saloons with "Flour of the Family" printed above her head.
Roberson sued at seventeen, won twice in the lower courts, and in 1902 New York's highest court ruled against her, four votes to three, the Chief Judge writing that many women would consider such use a "compliment to their beauty."
The public was disgusted, and the following year the state legislature made it a misdemeanour to use anyone's name or picture for trade without first obtaining written consent.
The courts have kept decisions narrow. It covers advertising and trade, nothing else. A photographer who shot his neighbours through their windows and sold the prints in a gallery won in court in 2015, because art is neither advertising nor trade. Being seen is fine, but being used is not.
|
[ creativity ]
In 1972, John Berger opened Ways of Seeing, his BBC series and book, with a history of the image. "Images were first made to conjure up the appearances of something that was absent."
The first images were magic, making the dead and the distant appear. Then, gradually, "it became evident that an image could outlast what it represented," and the image turned into a record, proof of how something once looked, until "an image became a record of how X had seen Y."
Photography arrived as a perfect version of it, so trustworthy that courts, newspapers and family memory treated it as proof.
The record could lie immediately. In 1840, barely a year after photography became publicly accessible, a Frenchman named Hippolyte Bayard staged the first faked photograph in history, posing as his own drowned corpse to protest Daguerre taking the credit for inventing the medium, with a mock suicide note on the back: "the corpse which you see here is that of M. Bayard."
What kept fakes like his rare for the next century and a half was the cost. Faking a photograph took staging, skill, and access to the subject, so the default photograph stayed a record of something that happened. Berger himself never called the camera neutral, every photograph being the photographer selecting one sight "from an infinity of other possible sights," but even his scepticism assumed the sights were true.
|
[ making ]
Artists met the being-used problem before the rest of us, because their work was scraped first.
At the University of Chicago, Ben Zhao's lab released Glaze in 2023: run your image through it before posting and a human eye sees nothing changed, but a machine studying the picture can no longer learn your style from it.
The same lab followed with Nightshade, teaching a scraping AI models wrong lessons about what it is seeing.
Between them the two tools have been downloaded almost nine million times.
Last summer, researchers at Cambridge, Darmstadt and Texas, broke both. Their method, LightShed, detects a protected image with 99.98% accuracy, stripping protections out, leaving images clean for training.
They published it as a warning, rather than as a weapon against artists.
The researchers who track these tools say to use Glaze / Nightshade a deterrent, not a guarantee.
|
|