Study reveals that speech recognition systems are more racist than previously thought

A Stanford University study reveals that voice recognition systems from Amazon, Apple, Google, IBM and Microsoft are not prepared to correctly identify what individuals of different races say. The researchers found that the systems recognize words spoken by users much less accurately than non-Caucasians.

To understand how the systems recognize different voices, the researchers decided to have them transcribe almost 20 hours of interviews with 73 participants of African origin and 42 white participants. The study published in the Proceedings of the National Academy of Sciences shows that the percentage of errors given by the systems when listening to people with black skin tone is 35%, with 20% of the audio heard considered unreadable. As far as white users are concerned, the figure is around 19%, and they cannot understand just 2% of the audio samples collected.

Stanford University researchers conclude that racial prejudice exists in systems as a result of the way they are trained. According to John Rickford, one of the study's authors, in an interview with The New York Times, the five tech giants are all making the same kind of mistake. It would be expected that all ethnic groups were well represented by these companies, but they are not, said the researcher.

In response to the study, Justin Burr, a Google spokesman indicated that the company is committed to improving the accuracy of its speech recognition system. Over the years, we have been trying to find a solution to the challenge of accurately recognizing different speech variations and we will continue to do so.

An IBM spokesman said the company continues to develop and improve the voice processing capabilities of its system. To date, none of the other technological giants mentioned in the study have yet commented on what happened.

The study returns to the public the issue of prejudice or even discrimination in artificial intelligence systems. Previously, an MIT study found that Amazon's facial recognition technology made many more mistakes when trying to identify women and black people. Remember also the example of Microsoft's chatbot, which the Internet managed to teach to be racist.