AI-generated images could influence upcoming elections in 2024. How can fake images be identified? An expert gives tips.
Frankfurt – There are some important elections coming up around the world in 2024. The European elections will take place in June, Saxony, Thuringia and Brandenburg will elect a new state parliament in September and the United States will most likely have to decide between Donald Trump and Joe Biden as the next US president in November.
What makes matters worse: AI models are getting better and better at creating deceptively real images, videos and deepfakes. As AI image generators are optimized, the risk of disinformation also increases. Therefore, here are some tips to help you better recognize AI-generated images.
AI images promote disinformation in the 2024 elections - many do not recognize AI-generated images
As a recent report from the Center for Countering Digital Hate revealed, AI image generators can still be manipulated to create misleading election-related images. Although leading AI companies have already promised to address this problem in the run-up to the US election, popular AI image tools Midjourney, Stability AI's DreamStudio, OpenAI's ChatGPT Plus and Microsoft's Image Creator failed the tests. With the “right” commands, all of them could be persuaded to generate deceptive images – which could also promote disinformation before the elections
In a study, the University of Waterloo also found that the test subjects had surprisingly great difficulty recognizing AI-generated people in images. Of 260 participants, only 61 percent could tell the difference between AI-generated and real people. The researchers had expected a proportion of 85 percent.
People are not as adept at making distinctions as they think
Andreea Pocol, a doctoral student in computer science at the University of Waterloo and lead author of the study
Recognize AI-generated images: You should pay attention to these features
So that you can be part of this percentage in the future, here are some tips for recognizing AI-generated images. When asked by
IPPEN.MEDIA
, Professor Doctor Marco Huber, AI expert and head of the image and signal processing department at the Fraunhofer Institute in Stuttgart, lists
the most important identifying features in deepfake images of people:
Eyes
: Light reflections in AI-generated images are asymmetrical, the pupils are not round, hairs (eyelashes, eyebrows) have an unnatural course, the course of the wrinkles around the eye is unnatural
Ears
: Symmetry of both ears is incorrect in fake images, earrings are not symmetrical and/or strangely shaped, shape of the auricle and earlobe is unnatural, shape progression is not continuous
Hair
: streaks without connection to the head, free-floating hair, very marked parting
Glasses
: In deepfake images, the temples of the glasses are not symmetrical and/or deformed, and the nose pads are not at the same height
Teeth
: The junction of the lip and teeth is blurred or frayed, the shape of the teeth is unnatural
Fingers
: unnatural hand/finger position in AI images, number of fingers is incorrect
Clothing
: Shirt/blouse collars are not symmetrical or have an unnatural curve, necklaces blend into the skin, hats are deformed and labels are incorrect
Skin
: unsteady progression of skin color in images modified or created with artificial intelligence
Background
: Continuation of the background is discontinuous, background has artifacts
The computer magazine
PCMag
also advises you to zoom in on images as closely as possible. Scattered pixels, strange outlines, misplaced shapes and blurry backgrounds can be better recognized in this way and can provide clues to an AI-generated image. On sites like “Which Face Is Real” you can directly test the knowledge you have learned.
Detecting AI images: These online tools can help you
If you are still unsure, there are now AI-controlled tools that are designed to recognize AI-generated images. The tools “AI or Not” and “Hive Moderation” are intended to help identify images created using artificial intelligence. The FakeCatcher from chip manufacturer Intel can also allegedly detect deepfakes with 96 percent accuracy. But be careful: these tools are not always correct either. A certain degree of skepticism and a careful eye on suspicious images remain essential.