Since the beginning of the year, I started a small experiment: I take Romanian science news, compress it with the help of AI and turn it into short clips for TikTok. The recipe is simple: an AI writes the text, another one suggests generic images ("laboratory researcher"), and a third one animates them.
I've already generated over 300 images. And among them, it always appears he.
The same man. A kind of perfect researcher, as the algorithms dream it.
He's in his 30s, brown hair, wears glasses, has a short, neat beard. Fair skin. European.
Here he is.
See him 7 more times below.
And here he is again, together with his clones.
Just like artificial intelligence that generates text, image-generating AI often relies on stereotypes. Researchers are usually seen holding a test tube, always wearing lab coats, etc.
Children already have stereotypes about researchers
And people think stereotypically about researchers, and that can actually be measured with a test called Draw a Scientist, created by David Chambers, who proposed a series of criteria for evaluating how clichéd the drawings are. For example, when children are asked to draw a scientist, both boys and girls are more likely to draw male scientists (but girls are less likely than boys). [1] [2] [3] [4] [5]. Researchers appear in lab coats, with glasses and messy hair, generally surrounded by chemistry experiment equipment. [2] [4] [6] [7] [8]They are alone and rarely in a group, and if they are not in the lab, they are surrounded by either books or technology. [2] [9] [5] [8]A recurring cliché in children's cartoons is that of the "nerd" and the "crazy researcher", as they appear in the media. [10] [11]. See scientific references below.
Reality is unbalanced, but to a lesser extent
In reality, the variation in the field is greater. Take the European Union. In higher education institutions, women represent 45% of the total number of employed researchers. Taking into account the private sector, the share of women drops to 34 percent.
"But why should we be upset with AI?" someone might say. "If in reality men really do predominate, isn't it normal for it to generate males more often?"
The problem is that stereotypes not only reflect reality, but also increases. Like a snowball, they grow bigger and bigger. If little kids see only "crazy" scientists everywhere, gray-haired men with wild hair, a whole generation will grow up thinking that's what scientists look like. Whole generations of little girls grow up feeling like there's no place for them, and their belief slowly turns into reality.
Why is it wrong that AI prefers stereotypes?
If the AI generates cliché images of male researchers in 80% from locations, it is actually distorting reality in favor of the cliché. It is true that men are the majority, but in a smaller proportion.
Moreover, new AI models are now mostly trained on “synthetic” data – that is, on text or images generated by older AI models. But that data already contains stereotypes. So the new models absorb them, repeat them, reinforce them. Until they come to seem like the reality of the AI.
The mysterious researcher is self-perpetuating on other sites
I wanted to see if the perfect image of the middle-aged white male researcher still appeared anywhere, so I turned to two search engines: Bing and Google. On both, when you enter an image, they return similar images.
The mysterious researcher also exists on stock image sites, those generic, for-sale photos that anyone can use in presentations or articles. But only in images that are visibly marked as being created by AI.
Once it has entered the "stock" world, the image of this generic researcher has the chance to circulate more and more. And to become, for the public, the "universal" face of research.
***********
- Scientific references were found using Consensus.
- Joy, A., Mathews, C., Hartstone-Rose, A., & Mulvey, K. What does a scientist look like? Children's perceptions of scientist gender and skin tone. School Science and Mathematics. 2024 https://doi.org/10.1111/ssm.18308
- Bozzato, P., Fabris, M., & Longobardi, C. Gender, stereotypes and grade level in the draw-a-scientist test in Italian schoolchildren. International Journal of Science Education. 2021; 43. https://doi.org/10.1080/09500693.2021.1982062
- Miller, D., Nolla, K., Eagly, A., & Uttal, D. The Development of Children's Gender-Science Stereotypes: A Meta-analysis of 5 Decades of US Draw-A-Scientist Studies.. Child development. 2018; 89 6. https://doi.org/10.1111/cdev.13039
- Blagdanić, S., Kadijevic, G., & Kovačević, Z. Gender stereotypes in preschoolers' image of scientists. European Early Childhood Education Research Journal. 2019; 27. https://doi.org/10.1080/1350293X.2019.1579551
- Bardullas, U., & Leyva-Figueroa, E. Unveiling Stereotypes: A Study on Science Perceptions Among Children in Northwest Mexico. Research in Science Education. 2024 https://doi.org/10.1007/s11165-024-10175-4
- Pekdoğan, S., & Bozgün, K. I Can Draw a Scientist Whom I Imagined. NeuroQuantology. 2019 https://doi.org/10.14704/NQ.2019.17.3.1932
- Buldu, M. Young children's perceptions of scientists: a preliminary study. Educational Research. 2006; 48. https://doi.org/10.1080/00131880500498602
- Newton, D., & Newton, L. Young children's perceptions of science and the scientist. International Journal of Science Education. 1992; 14. https://doi.org/10.1080/0950069920140309
- Savaş, E., & Gündüz, M. Primary and Elementary School Students' Profiles of Scientist Images. International Journal of Innovative Approaches in Education. 2024 https://doi.org/10.29329/ijiape.2024.662.3
- Brumovská, T., Carroll, S., Javornicky, M., & Grenon, M. Brainy, Crazy, Supernatural, Clumsy and Normal: Five profiles of children's stereotypical and non-stereotypical perceptions of scientists in the Draw-A-Scientist-Test. International Journal of Educational Research Open. 2022 https://doi.org/10.1016/j.ijedro.2022.100180
- 11. Tan, A., Jocz, J., & Zhai, J. Spiderman and science: How students' perceptions of scientists are shaped by popular media. Public Understanding of Science. 2017; 26. https://doi.org/10.1177/0963662515615086
English (UK) 
Română