Ethics of creating A.I. images in spotlight

Last weekend, millions of Twitter users saw Pope Francis appear on their feeds, modelling what seemed to be a custom Papal puffer coat.

In reality, though, the head of the Catholic Church never wore that designer, Balenciaga-like jacket: The image was nothing more than a hyper-realistic A.I. generation.

Behind this work of forgery was Pablo Xavier, a construction worker in Chicago who asked media not to use his last name for fear of backlash, reports Chris Stokel-Walker for Buzzfeed News.

“I just thought it was funny to see the Pope in a funny jacket,” Xavier tells the publication.

(The story and Xavier’s interview were published before Pope Francis was hospitalized with a respiratory infection and will likely spend a few days there, though he does not have Covid-19. Since this announcement, his condition has been improving.)

Xavier’s four pictures of a “dripped-out” pope were created using the popular A.I. tool Midjourney, which generates images based on text prompts.

However, like lots of computer-produced artwork, the photos of the pope are not perfect.

His smeared hand, misshapen glasses, and blurry cross necklace are all telltale signs of A.I. interference.

Nevertheless, the depictions tricked people, including model Chrissy Teigen.

“I thought the pope’s puffer jacket was real and didn’t give it a second thought. no way am I surviving the future of technology,” tweeted Teigen on Friday.

While the pope wearing a perfectly tailored, arctic white puffer could command a laugh out of even the most stoic internet users, the portrayal of public figures in realistic, A.I.-generated art has real-world implications.

A.I. pope in puffer jacket

Though he created the images for fun, Xavier tells Buzzfeed he had instant regrets once they went viral. He saw posts criticizing the Catholic Church for unnecessary spending, citing his A.I. creations as evidence.

Beyond the images of the pope-looking fly, Midjourney-generated content has tricked the internet on other occasions.

The same program is responsible for recently circulated phony photographs of Donald Trump getting arrested and Queen Elizabeth doing her own laundry.

While many were quick to debunk the Trump arrest photos due to their political implications, some never questioned the image of the pope and just kept scrolling.

“If Trump has been publicly arrested, I’m asking myself, why am I seeing this image, but Twitter’s trending topics, tweets and the national newspapers and networks are not reflecting that?”

Mike Caulfield, a researcher at the University of Washington’s Center for an Informed Public, tells the Atlantic’s Charlie Warzel.

“But for the pope, your only available heuristic is would the pope wear a cool coat? Since almost all of us don’t have any expertise there, we fall back on the style heuristic, and the answer we come up with is: maybe.”

Creating fake images is not the only problem that users are having with A.I. programs.

For one, the models also have coded biases.

When prompted with words like “CEO” or “director”, image generator DALL-E 2 churns out pictures of white men 97 percent of the time. Continue reading

Additional reading

News category: Analysis and Comment.

Tags: , , , , ,