More than 15 New Yorker cartoonists have recognised their own signatures on images generated by ChatGPT: beyond stylistic imitation, this raises the question of attribution.
An investigation by Nieman Lab shows that ChatGPT's image generator does not only imitate the style of the New Yorker's cartoons: it sometimes puts the signatures of real cartoonists. More than fifteen authors, current or past, have been identified, including Brendan Loper, Harry Bliss, Emily Flake, Joe Dator, Pat Byrnes, Peter Vey and Jason Adam Katzenstein, but also historical figures such as George Booth or Saul Steinberg.
The case of Brendan Loper, who signs "BLOPER," made the problem visible: a generated cartoon featuring Dolly Parton and Tim Curry circulated in August with his signature, to the point that readers wrote to him to find out if he was the author. For him, a signature is a certificate of authenticity. Emily Flake compares the experience to being given a quote that you never said.
From style to identity
The debate goes beyond the usual question of stylistic imitation. A drawing "in the manner of" can pass for a tribute; a drawing signed with a real name presents itself as an authentic work, especially when detached from its context on social networks. The behavior suggests that the model associated certain visual codes with the names that accompanied them in its training data. After requests from journalists, OpenAI added warnings when a cartoon in the style of the magazine was requested, without the signatures disappearing.
Read the full article on the original source
Go to the source →