Artificial intelligence avatars can be identified by blink rates that are half of humans
Digital human avatars can be distinguished from real people by their blink rates, which typically occur at half the frequency of a biological human's natural rhythm.
Artificial intelligence avatars created using Generative Adversarial Networks (GANs) often fail to replicate the involuntary physiological rhythms of the human eye, which typically blinks 400 times per minute. In contrast, synthetic personas like those developed from Tencent's Xiaoice framework often exhibit blink rates of only 200 per minute. This discrepancy serves as a primary forensic marker for identifying deepfakes, even when those avatars utilize Recurrent Neural Networks to time lip-sync phonemes to within 20 milliseconds of accuracy.
There's more to this story — open the app to keep reading.