December 29, 2025 | 02:36 pm
TEMPO.CO, Jakarta - Artificial intelligence (AI) has reached the stage where it can generate realistic faces both in image and video format, raising global concerns about the spread of disinformation and identity fraud.
A recent study brings some relief, showing that social media users can still be trained to recognize such falsity. Researchers believe that the negative impact of AI can be reduced if users can distinguish between real and fake visuals.
But more cases demonstrate that the general public is likelier to perceive AI-generated faces as original compared to real human faces. This phenomenon occurs because imitation algorithms often produce faces that are too symmetrical, proportional, and free from minor imperfections that are actually distinctive features of natural human faces.
The lead author of this study, Katie Gray, an associate professor of psychology at the University of Reading, United Kingdom, explained that the visual perception problem among the general public can be anticipated despite the emergence of advanced AI technology. "Quite short training procedure increased performance in both groups quite a lot," Gray said in a report by Live Science on Sunday, December 28, 2025.
How to Spot Fake Hyper-realistic AI Faces
The researchers explained that a person's ability to distinguish between real and fake AI-generated faces significantly improves after receiving direct feedback. When someone is informed that their guess is wrong and shown which specific parts to pay attention to, their brain begins to learn to recognize subtle patterns left by the machine.
The key to this training is to shift the focus of observation. Instead of looking at the face as a whole, the observer is taught to look for specific anomalies that are often failed to be imitated by AI. Some visual tells that are AI's weak points include non-perfectly round pupil shapes, inconsistent light reflections in the eyes, or blurred details on the ears and hair.
In addition, elements surrounding the AI-generated person also serve as important clues. Artificial Intelligence often struggles to render a coherent background or accessories like glasses and jewelry with proper physical logic.
By training the eyes to scan these small details, human success rates in identifying deepfakes can be drastically improved, providing new hope in the fight against visual hoaxes.
Read: Google's Parent Alphabet Secures $4.75bn AI Power Supply
Click here to get the latest news updates from Tempo on Google News
Fact Check: AI-Generated Video Falsely Shows Tornado Under Red Sky in Pandeglang
6 jam lalu
Fact check - The red sky phenomenon did occur in Pandeglang Regency, but there was no tornado.
IBM Sees Growing AI Adoption in Indonesia's Manufacturing
1 hari lalu
According to IBM's data, over the past 3-5 years, many industry players have surpassed the initial experimental phase. They have started AI pilot projects.
7 Mysterious Languages That Have Yet to Be Deciphered
4 hari lalu
Indus, Rongorongo, Linear A: Some ancient writing systems cannot be deciphered. Can artificial intelligence help crack the codes of the past?
Google's Parent Alphabet Secures $4.75bn AI Power Supply
5 hari lalu
A number of tech giants, including Google, are increasingly developing self-sufficient energy supplies to power their data centers.
Can AI Make English Learning More Accurate?
6 hari lalu
Will English skills remain relevant as AI technology becomes better at translating languages?
Meta to Launch 'Mango' and 'Avocado' AI Models in 2026
6 hari lalu
This project is managed under Meta's superintelligence laboratory led by Alexandr Wang, a founder of Scale AI.
Fact Check: False Narrative in Bandung and Cirebon Flood Video
7 hari lalu
Fact check - The content features a video showing flash floods in different contexts, including those created using artificial intelligence (AI).
Fact Check: AI Videos Falsely Show Tigers Saved in Sumatra Floods
8 hari lalu
Fact check - The videos in circulation were made using artificial intelligence (AI).
Gemini Users Can Now Build AI-Powered Mini Apps with Opal
9 hari lalu
Google has once again made a breakthrough in artificial intelligence (AI) by integrating the experimental "Opal" project into Gemini.
Fact Check: Ronaldo Aid Video to Aceh Is False
11 hari lalu
Fact check - Some of the videos circulating are from different events and are the result of AI.
