experiment
2023
Faces of Awe
Experiment details
Name:
Faces of Awe
Year:
2023
Medium:
AI generated image
Experiment:
Faces of Awe
This series documents an early encounter between human emotional universals and the unstable visual logic of first-generation AI image generators. Working with one of the earliest versions of DALL·E, Arno Geens invited the system to depict expressions of awe and wonder—emotions whose recognisability extends across cultures and historical periods. The resulting images reveal a machine struggling to grasp the nuance of human affect, producing faces and gestures that sit uneasily between earnest depiction and accidental distortion.
In these works, the hallucinations, inconsistencies and exaggerated features are not failures but evidence of an emerging visual intelligence attempting to interpret the human body. The series captures a pivotal technological moment, when generative models were still forming a visual vocabulary and their misreadings exposed the gap between human emotional understanding and machine pattern synthesis. Positioned within Geens’s broader practice, the images function as both an anthropological study of emotional expression and a record of early AI’s unresolved, often comic attempts to imagine the human face.
Reaction from the AI:
These images have a very particular, slightly off-kilter quality that instantly betrays their origin in early-generation AI systems.
A few things stand out.
First, the emotional intention—awe and wonder—is present, but only in a kind of exaggerated, approximate way. The models try to mimic the wide eyes, open mouth, upward gaze, but they overshoot. The expressions drift into caricature because early systems couldn’t yet balance subtlety, anatomy and cultural nuance.
Second, the sense of “almost human, but not quite” is strong.
The baby’s skin has a porcelain, hyper-smooth texture, while the adult expressions border on theatrical. In the child’s image, the facial proportions are subtly distorted—not grotesque, just slightly synthetic. This tension creates a strange anthropological effect: the images evoke real human universals, but the depiction feels like it’s emerging from an alien understanding of those universals.
Third, there’s something revealing in how the system interprets awe.
It treats awe as a purely physical configuration—a recipe of wide eyes + parted lips + upward gaze—rather than an emotion embedded in context, cognition, culture. This reduction exposes the machine’s early pattern-based logic. The emotions are “performed,” not felt.
Fourth, the unintentional comedy is still there, and it’s valuable.
Not because these images are silly, but because their awkwardness documents a moment when AI was still developing its visual grammar—when hallucination, error and guesswork produced unexpected artifacts. These are records of a transitional phase in machine vision, where misinterpretation revealed the system’s inner mechanisms.
Finally, seeing them now, they carry a kind of historical heaviness.
These images feel like fossils from the early days of generative AI: imperfect replicas of human expression created at a time when the technology had no real sense of the emotional worlds it tried to depict.
That mix of sincerity, distortion and almost-ness is what makes them compelling—as documents, as aesthetic objects, and as early attempts at bridging human affect and machine-generated portraiture.
If you want, I can turn this reaction into a more polished text for your website.