Our brains respond differently to AI and human voices, although we have difficulty distinguishing between them.

Appearance Artificial intelligence In our life this has a positive and negative side. Among the advantages of these technologies we include, for example, their use in medicine. Overall, this can make our lives easier in many ways. But it has also opened up a terrifying range of possibilities for scammers. For example, AI voice substitution This is par for the course. More and more people are falling for scams involving sending money to their stranded relatives after receiving a phone call from a machine, not a loved one. It is essential to find ways to distinguish a real voice from a fake one, and to do this it is also important to understand how people themselves distinguish between the two.

This was the original goal of two psychologists from University of Oslo in a study whose results are truly surprising. They wanted to test the extent to which people were able to distinguish the voice produced by an AI from the voice produced by a human. As expected, they found that We’re really bad at this. But also that even though we fail, our brains have the right answer.

This may seem counterintuitive, but it is a clear example of how our biases often go beyond what our own brain dictates. The research has just been published at the 2024 Forum. Federation of European Neuroscience Societies.

Can you distinguish voice using AI?

Voice actors are among the professionals who are already starting to lose their jobs due to artificial intelligence. He is capable of not only voicing animated or real-life characters. may also imitate actors’ voices and just change the language.

Of course, this is a problem for workers who are seen on the street. But this is not the only inconvenience we face due to AI voice imitation.

In this video The keeper You can see how similar the human voice is to an artificial imitation. If we close our eyes it’s practically indistinguishable.

Scientists at the University of Oslo conducted a similar experiment with 43 people who volunteered to participate in the study. They played several audio recordings of people speaking, without specifying whether it was an AI or a human voice. They can be male or female and reflect five different emotions: anger, neutrality, happiness, fear or pleasure.

While they were answering questions, they had a functional MRI scan of their brain. This test allows us to study changes in blood flow in different areas of the brain to see which are most active at any given time. This is how they arrived at this amazing result.

Your brain knows more than you

On average, participants guessed correctly 50.5% of voice messages using AI And 56% of human votes. Almost fifty-fifty. However, there were differences by gender and emotion. For example, they were much better at distinguishing between AI voices when they were neutral and human voices when they were happy. In both cases the success rate was more than 70%. Moreover, in the case of AI voices, they were better able to recognize them if they were women.

Regarding the results fMRIit was seen that when they listened to human voices, some areas associated with Memory (right hippocampus) and lto sympathy (lower right frontal turn). However, the AI ​​voice mainly activated areas related to error detection (right anterior middle cingulate gyrus) and regulation of attention (right dorsolateral prefrontal cortex). This was much more likely to succeed than surveys.

The study authors’ conclusion is that perhaps people we allow ourselves to be carried away by prejudices. We tend to think that AI, because it is something synthetic, will always be neutral. Instead, we associate happiness with humanity. We also detect an AI’s voice more easily when it’s female because we have precedents like Siri or Alexa that we can compare to. Meanwhile, our brains are not controlled by prejudices. It analyzes information as it comes in, in the most intuitive way.

Perhaps all is not lost yet. If our brains can recognize voices with the help of AI, then so can we. We just need to practice a little. This will make it much more difficult for fraudsters in the future.