(Montreal) Although we have recently heard more about its pitfalls and potential dangers, artificial intelligence can also contribute to securing individuals’ data.
Some financial institutions use speaker recognition, that is, they authenticate the customer using telephone services through the distinct characteristics of their voice.
Desjardins implemented this technology in June 2021. Since then, 1.5 million members, who have previously consented to the process, have been identified by their voice in six million calls.
“The authentication process [traditionnel] was named as an irritant by our members when they called, says Annie-Claude Jutras, Senior Director, Transformation of Customer Relations Centers at Desjardins. It took several tens of seconds at the start of each call to complete all the steps in the process. With voice authentication, it’s faster, more natural and more pleasant. »
This method is also safer, says Mr.me Jutras, since Desjardins uses a type of passive authentication, where the computer system recognizes the voice during a simple conversation, and this, from a hundred parameters. “We model the voice using these parameters which give us a unique key for each member, in the form of an alphanumeric code,” she explains. It’s like a fingerprint. »
By doing so, the customer does not need to say a passphrase to be identified, a type of authentication that has already been used by fraudsters, who recorded the voice of a person without their knowledge to then impersonate it and have access to personal information.
“This kind of authentication takes place in an automated system, nuance Mme Jutras. Training a synthetic voice or using an artificial intelligence engine to calibrate the voice to perfection is much more difficult in the context of a fluid discussion where there is an exchange of questions and answers between the member and the Desjardins employee. »
Laurent Charlin, senior academic member at Mila, associate professor at HEC Montréal and holder of a Canada-CIFAR Chair in Artificial Intelligence, agrees.
“We humans can noticeably recognize the voice of an interlocutor, but the system can be trained to distinguish very specific intonations,” he says.
“If some financial institutions have deployed this system, they must have made sure that it was better than the technology they were using before,” he adds.
Faking is not as easy as you think
According to Patrick Cardinal, professor and director of the Department of Software Engineering at the École de technologie supérieure (ETS), technology does not yet make it possible to simulate an improvised conversation with the voice of others from a simple speech sample.
“The fact that the authentication is done during a conversation, it leads to uncertainty about the questions or the exchanges to come, so it is difficult to predict the answers. Moreover, to get something reliable, it would take a huge amount of data to train [l’intelligence artificielle] : most people don’t have this data online,” he comments, referring to voice clips available on the internet.
It is different for television, radio or cinema professionals, whose voice excerpts abound on the Web.
“We have already seen videos of actors or politicians who said things they had never said, thanks to hyperfaking (deep fake), underlines Laurent Charlin for his part. There are also applications that, from a small extract, manage to generate your voice by making you say what they want. »
In case of doubt, it is always possible to fall back on traditional methods, such as identification questions, recalls Mme Jutras.
Verification made with the Canadian Anti-Fraud Center, no report of fraud related to artificial intelligence with voice cloning has been made, in particular to access a victim’s bank accounts. “That said, this is something the CAFC will continue to monitor closely,” said Jeff Horncastle, Acting Client Outreach and Communications Officer, via email.
This dispatch was produced with financial assistance from the Meta Exchange and The Canadian Press for News.