A team of Chinese scientists has found more evidence that most people cannot distinguish between real speech and AI-generated speech, even after receiving some training. The researchers, from Tianjin University and the Chinese University of Hong Kong, connected 30 people to brain-scanning devices while they listened to audio recordings, attempting to determine if the recordings were AI-generated or genuine human voices. In most cases, the answer was their inability to tell the difference, with the team describing the group as «weak at distinguishing between the two types». The team then sought to train the group that seemed unable to do so to improve their ability to detect fake voices, but the training «achieved only slight improvement». Meanwhile, the study showed that training opened the door to potential progress, with researchers involved in the study stating that «at the neural level, training made the brain's responses more distinct to human speech versus AI-generated speech». Xiang Binting, the team leader, said that «the auditory device in the brain seems to start picking up subtle sound differences, even if people cannot reliably convert that into a behavioral decision», adding that the «faint signals for voice recognition are encouraging». These tests came after a study published last September by Queen Mary University of London, warning that voices generated by «deepfake» technology using widely available programs «have now become indistinguishable from real human voices». It appears that people perform only slightly better when it comes to AI-generated images, with a study conducted by the University of New South Wales and the Australian National University and published last month finding that most people overestimate their ability to detect fake faces. Last year, Citibank issued a warning about these increasingly hard-to-detect AI-generated voice and image «deepfakes» that are spreading in areas like hiring, financial operations, and impersonation of executives. ..even in the area of images, most people overestimate their ability to detect fake faces.
