One hour of training can dramatically improve people's ability to detect AI-generated faces
A new study suggests that just one hour of structured training can significantly improve people's ability to distinguish between real human faces and those generated by artificial intelligence, offering a practical defense against the growing threat of deepfake technology.
The research, conducted by scientists from the University of Aberdeen in collaboration with the Australian National University, found that participants substantially increased their detection accuracy after receiving focused instruction on identifying subtle visual patterns commonly found in AI-generated portraits.
As modern image-generation models have become increasingly sophisticated, traditional signs of manipulation—such as distorted fingers, unrealistic backgrounds, or mismatched accessories—have largely disappeared. This has made fake images far more convincing and more difficult to identify with the naked eye.
Rather than teaching participants to search for obvious technical flaws, researchers encouraged them to recognize more subtle characteristics that current AI systems still struggle to reproduce consistently. These include excessive facial symmetry, unusually perfect proportions, overly idealized appearances, generic facial features lacking individuality, limited emotional expression, and faces that are difficult to remember after viewing.
The results showed a remarkable improvement. Before the training session, participants correctly identified AI-generated faces only about 40% of the time. After approximately one hour of practice using both authentic and synthetic images, detection accuracy rose to around 80%, with some participants achieving nearly perfect scores.
Researchers also observed that participants became more confident in their judgments while maintaining greater accuracy, suggesting that experience helps develop reliable visual intuition.
The findings arrive as concerns over deepfake technology continue to grow worldwide. AI-generated images and videos are increasingly being used in online fraud, identity theft, misinformation campaigns, political manipulation, and social engineering attacks. Cybersecurity experts have warned that highly realistic synthetic media presents significant challenges for governments, businesses, and individuals alike.
Recent incidents have demonstrated how convincing AI-generated content can be, with fraudsters using fabricated images, voices, and video calls to impersonate executives, public officials, and trusted individuals in attempts to steal money or sensitive information.
The study also found that AI systems remain less consistent when generating realistic images of elderly people, young children, and demographic groups that are underrepresented in training datasets. These limitations may provide additional clues for trained observers.
Researchers believe the human brain gradually develops pattern-recognition skills similar to those used by machine learning systems. Instead of relying on a single visual clue, people learn to recognize combinations of subtle features through repeated exposure.
The authors conclude that while automated deepfake detection software will remain an essential tool, human judgment supported by targeted training will continue to play a critical role in identifying manipulated digital content as generative AI technology becomes increasingly advanced.
-
19:05
-
18:44
-
18:25
-
18:10
-
17:48
-
17:33
-
17:15
-
17:00
-
16:41
-
16:20
-
16:05
-
15:44
-
15:25
-
15:10
-
14:50
-
14:34
-
14:22
-
14:15
-
14:05
-
14:00
-
14:00
-
13:53
-
13:49
-
13:42
-
13:21
-
13:21
-
13:21
-
13:05
-
13:02
-
12:45
-
12:28
-
12:12
-
11:47
-
11:30
-
11:20
-
11:16
-
11:15
-
11:00
-
10:52
-
10:43
-
10:26
-
10:25
-
10:19
-
10:12
-
10:10
-
10:07
-
09:58
-
09:45
-
09:45
-
09:25
-
09:10
-
08:57
-
08:48
-
08:44
-
08:33
-
08:25
-
08:24
-
08:17
-
08:16
-
08:15
-
08:13
-
08:11
-
08:08
-
08:06
-
08:03
-
08:00
-
07:59
-
07:54
-
07:43
-
07:21
