See if You Can Spot AI Deepfakes in This Interactive Test

See if you can spot an AI deepfake with our interactive test. Psychologist Dr. Clare Sutherland stands before the camera holding two substantial photographs. While one captures a genuine portrait of an Australian scholar conducting global research, the other represents a synthetic creation produced by artificial intelligence. As machine learning technology advances, distinguishing authentic human likenesses from computer-generated imitations has grown increasingly complex. However, recent investigations suggest that individuals can indeed develop the ability to identify machine-made imagery through targeted instruction.

This inquiry is being spearheaded by Sutherland at the University of Aberdeen alongside her counterpart in Australia. Before unveiling their findings, readers are invited to participate in a brief assessment to gauge their own detection skills. See if you can spot the differences yourself. Many find this task challenging, which is entirely normal. Historically, identifying synthetic visuals was simpler because early AI models frequently committed glaring errors, such as rendering an additional finger or distorting facial features in obvious ways.

The Evolution of AI Detection Challenges

However, these systems evolve rapidly. “Training on visual artifacts, like looking for a sixth finger or odd earrings, has had limited success, partly because the AI is getting too good, and fraudsters may avoid using pictures with obvious flaws anyway,” explained Prof Amy Dawel. Dawel, recognizable in the photograph as the woman with shoulder-length hair, serves as the director of the Australian National University Emotions and Faces Lab. Her colleague, the man depicted in the image, represents the artificial construct.

Dawel leads an international consortium comprising scientists from the United Kingdom, Canada, and Australia. Their primary objective is determining whether human observers can be conditioned to recognize AI imposters. Current evidence indicates that the answer is affirmative, provided the training methodology emphasizes nuanced observation rather than simple rule-checking. Sutherland notes that humans naturally begin to sense the difference between real and synthetic faces simply through exposure. “So we thought, OK, it would be really interesting to see if we could teach other people this too,” she said.

To conduct their trials, the team utilized StyleGAN3, a sophisticated tool capable of generating highly realistic human portraits. Thousands of these synthetic faces were compiled for the experiment. Participants underwent testing both prior to and following a structured training session. The curriculum focused on six distinct perceptual attributes that differentiate human faces from AI creations.

First, researchers highlighted symmetry. Humans possess unique asymmetries, such as a slightly uneven eyelid or a crooked smile, which AI often smooths out. “If it’s too good to be true, it probably isn’t,” the team observed regarding proportionality. Large noses or prominent ears are less common in deepfakes due to this tendency toward perfection. Furthermore, attractiveness plays a role. “AI faces tend to look more attractive,” explains Sutherland. “That one is more subjective, an aesthetic judgement, but AI often creates faces that are pleasant looking.”

Distinctiveness is another critical factor. “That could be something like ‘what would make a face stand out in a crowd,'” Dawel noted. AI-generated faces often lack these memorable characteristics, appearing somewhat generic despite their realism. The training program taught participants to look beyond surface-level features and consider the overall coherence of facial elements.

See if you can spot these subtle differences in our comprehensive test. The results demonstrated that with proper guidance, people significantly improved their ability to distinguish between genuine and artificial portraits. This research has important implications for combating misinformation and verifying digital content in an era where AI-generated media becomes increasingly prevalent.