In the highly AI-driven world, we have been finding it increasingly challenging to figure out whether a photograph is real or is created with generative AI. While a large number of people were unable to differentiate between the two, a new study claims that 70% of the Americans participating in the experiment were actually able to tell the difference. How accurate is this? Let’s find out.
According to a study conducted by Joe Youngblood, the survey had 4,016 Americans ranging between ages 18 to 65. The study was conducted using a random sample method, meaning that these participants could be from any social state, race, gender, or ethnicity and have varying intelligence levels. Using the Typeform-based survey, the participants were asked to choose between two images and point out which one was created using AI. The team also employed AI models such as ash Midjourney, Grok 3, Dall-E, and ChatGPT (only until version o3, since o4 hadn’t launched then).
The images ranged from Scarlett Johansson as Black Widow to a painting of George Washington and images of the Eiffel Tower. Since the subjects were so different, a large number of them were able to identify the AI images instantly. Here are a few examples of the images below:
From the images seen above, 88.78%of people could detect an AI image of Scarlett Johansson, and 87.97% could do the same with the peacock. Only 73.1% could detect the Gen AI image of the billboard, while only 18.05% could figure out the fake Eiffel Tower.
The gradual drop seen in the images only proves that as long as the model is sophisticated, one can figure out whether an image is real or AI-generated. However, with the newer model of Midjourney V7 and Chat-GPT 4, this survey would tell a rather different story. As you may have seen in our reportage, the newer models are far more sophisticated, with one, including the professional photographers, being unable to tell the difference. The only reason this survey showed positive results is because some images appear like 3D models instead of the real deal. For instance, anyone who has seen a real person can easily differentiate between Johansson and the animated character. Furthermore, Johansson is a far more famous figure in popular culture, which makes it easy to recognize her. The examples of billboards and the Eiffel Tower only prove that with sophisticated software, anyone can easily be fooled.
So, the question of whether Americans can distinguish between AI and real images entirely depends on how the images are created and what the subject is. Furthermore, per a study (via Fast Company) conducted in 2024, only 61% of the people were able to figure out the image of a real-person version or an AI-generated image. The newer models will only going to add to the confusion, which makes it imperative that we have better guadrails. If not, then more people will fall into varied traps.
