For years, AI-generated images and AI-assisted artworks have been mocked for one thing above all else: the hands. Fingers twisted into knots, palms with too many digits, phantom thumbs floating on wrists. While much of the discourse treats these malformed hands as punchlines or evidence of AI’s incompetence, I want to propose something more generous, and perhaps more revealing.
First, I think I have a fair technological justification for why AI hands tend to appear as they do. But more importantly, I believe these flawed hands are unexpected mirrors. They offer us a chance to rethink perfection, beauty, and even what it means to be human in an increasingly digitized world.
Let’s begin with the technical side. Why are AI-generated hands often so... “AI-handsy”?
The simple answer is: AI hands aren’t perfect because all human hands aren’t perfect.
Machine learning systems like diffusion models learn patterns from enormous datasets. These datasets include millions, if not billions, of visual representations of hands: photographs, paintings, anatomical diagrams, cartoons, sculptures, and more. But crucially, not all of these hands conform to an idealized anatomical standard.
Some very real hands in the data have very real differences.
Some hands have birth differences. Others are missing fingers due to accidents or medical conditions. Some are stylized, abstracted, or distorted by artists intentionally. Many are partially obscured, blurred by motion, or captured from strange angles. The AI learns from all of it.
When a generative model creates a hand, it doesn’t retrieve a template or copy from memory. It constructs a new image from learned representations: internal mappings of patterns, shapes, and associations drawn from millions of visual examples. The result is not a direct reproduction of any one hand, but an original synthesis that reflects the statistical relationships it has inferred across a vast, messy, and diverse visual archive.
In that sense, the malformed AI hand is not a failure of logic, but a mirror to the variability of how hands appear across human experience.
This broader view of data should also inform how we understand the ongoing debates around AI art and training data ethics. It’s popular to claim that AI models are built solely on “stolen art,” but the reality is more complicated. These datasets contain everything from medical illustrations and photojournalism to comics and classical paintings, much of it publicly available or legally obtained through third parties.
By acknowledging the diversity of the data, not just in authorship, but in subject matter, we begin to understand why AI hands appear the way they do. More importantly, we can also challenge the aesthetic and cultural expectations that lead us to label these outputs as defective.
We are conditioned to expect a certain kind of hand: five fingers (four digits and a thumb), symmetrical, realistic. But what does it say about us if we dismiss anything outside that template as broken?
In our rush to correct the AI’s visual “mistakes,” are we also reinforcing a narrow and ableist vision of what bodies and art should look like?
We often assume a hand has five fingers—but this is not a biological law, it’s a cultural default. In reality, humans have always exhibited a wide range of anatomical variation. One such condition is polydactyly: being born with more than five fingers on a hand.
Polydactyly affects approximately 1 in every 500 to 1,000 births, and the extra digits can range from small, underdeveloped nubs to fully formed and functional fingers. Sometimes the additional fingers appear on the pinky side (postaxial), the thumb side (preaxial), or even in the center (central polydactyly, which is rare). It can run in families, appear as a spontaneous mutation, or be associated with other syndromes.
Importantly, these hands which have been documented across cultures, eras, and medical archives, are part of the dataset. They exist in medical textbooks, family photo albums, and documentary photography. Their presence in training data shouldn’t surprise us, but our discomfort with them might. If an AI model generates a hand with six fingers, is that truly an error, or is it an echo of a statistically rare, but very real, human experience?
By insisting that AI hands must conform to a narrow five-digit norm, we risk perpetuating ableist assumptions that only certain bodies are worthy of depiction. We forget that the history of art and anatomy is full of variation, distortion, and deviation. Including polydactyl hands in our conception of “normal” challenges the rigid standards we often apply, both to art and to each other.
Surrealism, Expressionism, and modern art movements have long distorted anatomy to evoke emotion, psychological depth, or political tension. Egon Schiele’s jagged hands, Käthe Kollwitz’s swollen fists, and Picasso’s fractured figures all rejected anatomical realism in favor of expressive truth.
AI’s flawed hands, though unintentional, often fall into a similar aesthetic territory. They are strange, alien, and sometimes unsettling, but they are also evocative. They point to the limits of representation and the unpredictability of machine vision. And they remind us that deviation from the norm is not just acceptable—it’s essential to the evolution of art.
To demand flawless hands from AI is not just a critique of software: it’s a veiled attack on real bodies that don’t meet sanitized standards of “normal.”
When people mock AI-generated hands for having too many fingers or “getting it wrong,” they’re not usually thinking about the humans who live with such differences. But they are participating, however unwittingly. in a broader act of erasure. These critiques expose an instinct to erase visible difference. To pathologize it. To call it broken. But if a six-fingered hand is an “error,” what does that say about those born with polydactyly? What does it say about those who’ve lost fingers, or never had them? What does it say about every imperfect hand that has painted, built, comforted, or survived?
The AI doesn’t know better—but we should.
If we truly value humanism, then we must extend that value to the outputs of our tools, technologies, and perhaps most importantly to our art, irrespective of how it is made. The malformed AI hand is not an error to be debugged, it’s a mirror held up to the fractured image of ourselves, shaped by billions of messy, partial, and imperfect representations.
In AI’s flawed hands, we can see the history of art, the diversity of human bodies, and the ongoing tension between order and chaos, control and creativity. We can see the mess of life itself.
Rather than striving to fix these outputs into "realism," we might instead ask: What do these hands reveal about how we define normalcy, beauty, and worth? What do they say about how we treat people whose bodies deviate from the expected?
To grasp imperfection is not to surrender to error. It is to recognize in it a deeper kind of truth.
Josh Davila
