Deep Hysteria (2023)

Deep Hysteria is a still image series that repurposes algorithmic bias in the service of unraveling a deep human bias.

Side by side deep-learning-generated images of male and female versions of a young adult. Male caption: Calm-Looking Male, 20–28. Female caption: Confused-Looking Female, 23–31.
Side by side deep-learning-generated images of male and female versions of a young adult. Male caption: Calm-Looking Male, 18–26. Female caption: Fearful-Looking Female, 19–27.
Side by side deep-learning-generated images of male and female versions of a young adult. Male caption: Calm-Looking Male, 14–22. Female caption: Suprised-Looking Female, 12–20.
Side by side deep-learning-generated images of male and female versions of a teen or young adult. Male caption: Calm-Looking Male, 12–20. Female caption: Sad-Looking Female, 14–22.
Side by side deep-learning-generated images of male and non-binary or female versions of a young adult. Male caption: Calm-Looking Male, 23–31. Non-binary/Female caption: Surprised-Looking Female, 14–22.
Side by side deep-learning-generated images of male and female versions of a young adult. Male caption: Calm-Looking Male, 19–27. Female caption: Confused-Looking Female, 24–34.
Side by side deep-learning-generated images of male and female versions of a young adult. Male caption: Calm-Looking Male, 21–29. Female caption: Sad-Looking Female, 22–30.
Side by side deep-learning-generated images of male, non-binary, and female versions of a middle-aged adult. Male caption: Calm-Looking Male, 41–49. Non-binary caption: Fearful-Looking Female, 45–51. Female caption: Surprised-Looking Female, 47–53

The people in these artworks are “AI”-generated twins that vary in gender presentation — titled as another “AI,” trained on human perceptions, identifies the emotion on their faces.

The faces are generated using deep learning algorithms trained on still frames of thousands of YouTubers speaking to the camera. Generated individuals are then algorithmically regendered and the variations fed to commercial deep learning based facial analysis algorithms, which attempt to categorize the faces according to the emotions apparent in the subjects’ facial expressions. Despite the marketing of such tools, reading emotions solely by analyzing a person’s face is a feat that neither humans nor “AI’s” can reliably do. Further, these deep learning algorithms are themselves trained on data categorized by humans — so they reflect human biases.

Sequence of line drawings of a depicting a late 19th century woman in various poses stereotypically associated with stress or anxiety.
Sequence of drawings from 1893 depicting a woman with “hysteria.” By Albert Londe – La photographie médicale : application aux sciences médicales et physiologiques, Public Domain.
Google Image search results for "anxiety" depicting women in various emotional poses.
Google Image search results for “anxiety,” 2023

For centuries, “hysteria” was a medical and mental diagnosis that assumed females had an innate predisposition toward an anxious and nervous emotional state. Although the diagnosis has been retired, stereotypes of women as nervous, fearful, and uncertain continue to impact how women are perceived and treated. And while more women than men are diagnosed with anxiety, a Google image search for “anxiety” returns a far disproportionate number of images of women — who tend to be depicted in stereotypical “female hysteria” poses. The stereotype is amplified by the cultural expectation that smiling is a woman’s default facial expression. Consider the phenomena of “Resting Bitch Face” and “telling women to smile.” A neutral facial expression on a women is read as disgust, distress, or unhappiness: “What’s wrong?” At the same time, stereotypes around masculinity may prevent men who experience anxiety from seeking or receiving help.

The artworks in Deep Hysteria redeploy the bias embedded in facial analysis algorithms in the service of probing this deeply entrenched social bias.

Complete info on Deep Hysteria here.

A paper, Deep Hysteria: What algorithmic bias tells us about our emotional perceptions of women, is available here.

High resolution images are available for exhibition or publication.

Nine algorithmically gender-varied versions of an AI-generated face.