Through a Gendered Lens: Artificial Intelligence, Photographic Representation, and the Politics of Vision in World Cinema
This mixed-methods study reveals that artificial intelligence systematically reproduces patriarchal photographic conventions in world cinema through a quantifiable "Gendered Photographic Bias Index" and a theoretical "AI-Gendered Gaze Framework," while empirical data confirms that female film practitioners experience these tools as significantly more constraining than their male counterparts.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the movie camera as a magical mirror that doesn't just reflect the world, but decides how we see it. For decades, film theorist Laura Mulvey argued that this mirror was held by a "male gaze," turning women into objects to be looked at rather than people with stories. But now, a new kind of mirror has arrived: Artificial Intelligence.
This paper asks a scary but fascinating question: If we teach an AI to make movies by showing it millions of old photos and films, will it just copy the bad habits of the past?
The short answer is a loud, statistical yes.
The "Bad Habits" Machine
Think of AI like a super-fast student who learns by copying. If you show this student a library full of old movies where women are always shown in tight close-ups (focusing only on their lips or eyes) and men are shown from far away (showing their whole body and power), the student learns that this is how the world works.
The researchers tested three popular AI image generators (Stable Diffusion XL, DALL-E 3, and Midjourney v6) to see what they produced when asked to create scenes with male and female characters. They invented a special ruler called the Gendered Photographic Bias Index (GPBI) to measure the "objectification" of the images.
- The Benchmark: A perfect, fair score is 1.00.
- The Results: The AI tools scored between 1.62 and 1.94.
This means the AI didn't just make a few mistakes; it systematically made women look more like objects than men. Specifically:
- The Zoom: When the AI drew a woman, it zoomed in tight (like a close-up of a face), but when it drew a man, it stepped back to show the whole scene. The women were framed at roughly half the distance of the men.
- The Lighting: The women were lit with harsh, high-contrast shadows (making them look dramatic but distant), while men got softer, more natural light.
- The Fragmentation: The AI was more likely to cut off parts of a woman's body in the frame, focusing on specific body parts rather than the whole person.
The "Democracy" Myth
You might have heard people say, "AI is great! It lets anyone make movies and breaks down the gates!" The paper argues that this is a trap.
Imagine a video game where the default settings are rigged so that playing as a female character is harder and looks worse than playing as a male character. If you want to fix it, you have to spend three times as much time tweaking the settings. The paper found that 74.1% of women in the film industry feel that AI tools are "encoding patriarchal photographic conventions," compared to only 40.8% of men.
For women filmmakers, using these tools isn't a free pass to creativity; it's an extra job. They have to fight the AI's default settings just to get a picture that doesn't look like it was taken through the eyes of someone who doesn't understand them. The paper suggests that instead of democratizing filmmaking, AI is just making it cheaper to reproduce the same old, biased ways of seeing.
The "Invisible Hand"
Why does this happen? The paper explains that the AI isn't "evil"; it's just a statistician. It learned from a "corpus" (a giant library) of images that were mostly created by men, for a male-dominated industry.
- The Training: The AI absorbed the "worst habits" of the past—like the way women were often filmed as objects—and made them the "default" setting.
- The Illusion: Because the AI creates images that look so real (photorealistic), we forget that they aren't real. There was no actual person there to be photographed. The AI just calculated the most likely way to draw a "woman" based on its biased training data. This makes the bias invisible, like a ghost in the machine.
What the Paper Says (and Doesn't Say)
The authors are very sure about their findings because they didn't just guess; they measured it.
- They measured: They surveyed 186 film professionals from India, the UK, and the US. They generated 600 images and had experts rate them.
- They proved: The bias is real, measurable, and consistent across different AI systems.
- They ruled out: They explicitly reject the idea that AI is a neutral tool that will naturally fix inequality. They also note that their study didn't cover every country in the world (like Korea or Nigeria), so we don't know if the numbers are exactly the same everywhere, but the pattern is strong where they looked.
The Fix?
The paper suggests that we can't just wait for the AI to "learn better" on its own. We need to change the rules:
- Audit the Library: Check the training data to make sure it includes more women directors and photographers.
- New Standards: Film studios should refuse to buy AI tools that have high bias scores (above 1.2).
- Transparency: AI images should carry a "label" telling us they were made by a machine and what biases might be in them.
- Women in the Driver's Seat: Women cinematographers and feminist scholars need to be the ones designing these tools, not just using them.
In the end, the paper concludes that the camera has always been political. Now, the "gaze" isn't just human; it's algorithmic. And unless we fix the code, the algorithm will keep showing us a world where women are objects, and men are the heroes. The question isn't if the AI has a gendered gaze, but whether we have the courage to see it and change it.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.