2023-08-03 20:37:00
The acceleration and enhancement of creative tools by generative Artificial Intelligence is evident. These tools have been applied in the generation, among several fronts, of fake audios. In addition, according to a study published in the scientific journal PLoS ONE, humans have difficulty distinguishing between real human speech and the so-called deepfakes.
Researchers at University College London (UCL), in England, conducted an experiment with 529 participants. The objective was to assess people’s ability to differentiate between these two types of audio, both in English and in Mandarin. Surprisingly, more than a quarter of volunteers (27%) had difficulties distinguishing between what was real and what was created by the AI algorithm. The tested audios were in English and Mandarin.
The result shows that the majority still was able to distinguish correctly 73% of the time, but there is still a large portion that cannot. Even after learning techniques to discern false lines, there was little variation in the accuracy rate.
“Our findings confirm that humans are unable to reliably detect deepfake speech, whether or not they have received training to help them identify artificial content,” says Kimberly Mai, study author, in a note. “It would be prudent for governments and organizations to develop strategies to deal with the abuse of these tools”, says Lewis Griffin, another author of the research.
The authors of the study, however, say they are excited about the potential of this type of technology, such as the possibility of restoring the voices of people who have lost it due to an illness.
1691114681
#quarter #people #real #audio #deepfakes #study #finds