It used to be easy to tell when a face was generated using artificial intelligence (AI). Whether it was a distinctive eerie glow, incredibly smooth skin, eyes that made no sense, or a noticeable third ear, the human face replicas of older AI models were easy to spot and easy to dismiss. That’s simply not true anymore.
Now AI image generators can create portraits so convincing that even careful observers have difficulty distinguishing fact from fiction. That’s why apps like Zoom and Tinder allow their users to submit biometric identification, such as retinal scans, to prove that a real person exists behind a profile picture. But a new study suggests you can train your brain to better detect fakes.
Previous attempts to teach humans to recognize AI faces have focused on teaching viewers to look for visual glitches or statistical fingerprints left by a particular image generator, such as a crooked ear or an eye with two pupils. The problem is that these notices can disappear with a software update or simply by using a different command prompt. “AI is getting too good,” said Amy Dawel, an associate professor at the Australian National University and lead author of the study, in a press release. “And fraudsters may avoid using images with obvious flaws anyway.” The result is an endless technological arms race that humanity appears to be losing.
Table of Contents
To support science journalism
If you enjoy this article, please consider supporting our award-winning journalism by subscribing. By purchasing a subscription, you are helping to ensure the future of impactful stories about the discoveries and ideas that shape our world today.
Instead, the researchers taught participants to recognize broader patterns in the way AI systems produce images. “Our training draws people’s attention to global characteristics that differ between AI and human faces,” Dawel said.
Current AI image generators are, in turn, trained using data sets consisting of millions of images. When asked to generate a face, they do not copy specific faces, but instead create a new face based in part on the mathematical patterns common to all faces in that data set – these allow the AI to construct a “typical” human face.
The result is that AI-generated faces often trend toward statistical averages. They’re not overly unrealistic, just a little too balanced, a little too general, and a little too conventional. By themselves, none of these features are necessarily suspicious. But taken together, the whole is more boring than the sum of its parts – a subtle banality that people can often sense implicitly.
“Even relatively short training sessions helped participants improve their accuracy,” says Tanya George, a student researcher at the Australian National University who trained the study’s participants. “Research like this can help people navigate increasingly complex online environments.”
Compared to real faces, AI-generated faces tend to be more symmetrical, proportional and attractive – but at the same time they are less expressive, less distinctive and significantly less memorable. When the researchers trained participants to look for these six markers instead of fleeting artifacts like misshapen ears or mismatched jewelry, their ability to recognize the AI’s face nearly doubled.
In other words, the AI focuses on the center. Real people don’t do that. Countless small deviations from the norm shape our face – our subtle asymmetries, peculiarities and expressions make us unforgettable. These imperfections are not defects. They are our trademark.
It’s time to stand up for science
If you enjoyed this article, I would like to ask for your support. Scientific American has been an advocate for science and industry for 180 years, and right now may be the most critical moment in this two-hundred-year history.
I was a Scientific American I’ve been a subscriber since I was 12 years old and it has shaped the way I see the world. SciAm It always educates and delights me and instills in me a sense of awe at our vast, beautiful universe. I hope it does that for you too.
If you subscribe Scientific Americanhelp ensure our reporting focuses on meaningful research and discovery; that we have the resources to report on the decisions that threaten laboratories across the United States; and that we support both aspiring and working scientists at a time when the value of science itself all too often goes unrecognized.
In return, you get important news, compelling podcasts, brilliant infographics, must-see newsletters, must-watch videos, challenging games, and the best writing and reports from the science world. You can even gift someone a subscription.
There has never been a more important time for us to stand up and show why science matters. I hope you will support us in this mission.
https://www.scientificamerican.com/article/how-to-spot-an-ai-generated-face-according-to-science/
