TL;DR:
- Financial institutions are utilizing artificial intelligence (AI) to enhance data security.
- Speaker recognition technology is being employed to authenticate customers based on their voice characteristics.
- Desjardins, a Quebec-based institution, has successfully identified 1.5 million members through voice authentication in six million calls.
- Voice authentication offers a faster, more natural, and more secure alternative to traditional authentication methods.
- AI-powered systems can discern specific intonations and provide superior performance compared to previous technologies.
- Challenges remain in simulating improvised conversations and the need for extensive training data for AI reliability.
- Fallback methods, such as identification verification, can be utilized in situations of uncertainty.
- The Canadian Anti-Fraud Centre is closely monitoring the potential risks of AI fraud and voice cloning.
Main AI News:
In the face of growing concerns surrounding data security, some financial institutions are championing the use of artificial intelligence (AI) as a means to protect individuals’ personal information. While recent reports have shed light on the potential dangers of AI, these institutions argue that it can also serve as a valuable tool in safeguarding customer data.
Speaker recognition is one AI application that is gaining traction among financial institutions. By leveraging the unique characteristics of an individual’s voice, institutions are using this technology to authenticate customers during telephone interactions. Quebec-based Desjardins, for instance, introduced this technology in June 2021, and since then, approximately 1.5 million members who previously consented to the process have been successfully identified in over six million calls.
Annie-Claude Jutras, Senior Manager at Desjardins’ Customer Relations Centre Transformation, explained the motivation behind adopting voice authentication: “The traditional authentication process was seen as an irritant by our members when they called. We had to spend several dozen 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.”
In addition to offering a more streamlined customer experience, Jutras emphasized the enhanced security provided by voice authentication. Desjardins employs a passive form of authentication, where the computer system recognizes the unique voice patterns of individuals during regular conversations, relying on approximately a hundred distinct parameters. By modeling these parameters, Desjardins creates a unique alphanumeric code for each member, akin to a fingerprint.
This method eliminates the need for customers to provide a secret phrase, a type of authentication that fraudsters have exploited in the past by surreptitiously recording individuals’ voices. “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,” explained Jutras.
Laurent Charlin, a Senior Academic Member at Mila and an Associate Professor at HEC Montréal specializing in Artificial Intelligence, supports the implementation of voice authentication systems. Charlin noted, “While humans are adept at recognizing a speaker’s voice, the system can be trained to discern highly specific intonations. Financial institutions that have deployed this system must have ensured its superiority over previous technologies.“
Overcoming Challenges: The Complexity of Voice Cloning
According to Patrick Cardinal, Professor and Head of the Software Engineering Department at the École de technologie supérieure (ETS), current technology does not yet allow for the simulation of improvised conversations using someone else’s voice based on a simple speech sample. Cardinal explained, “Given that authentication occurs during a conversation, there is inherent uncertainty regarding the questions or exchanges that will transpire, making it difficult to predict the answers. Furthermore, training artificial intelligence to achieve reliable results requires an extensive amount of data, which most individuals do not have readily available online.”
Cardinal’s concerns pertain to the scarcity of voice samples accessible on the internet. However, this limitation does not apply to professionals in television, radio, or film whose voice recordings are abundant online. “Thanks to deep faking, we have already witnessed videos in which actors or politicians appear to say things they have never said before,” pointed out Charlin. “Certain applications can generate a person’s voice and manipulate it to utter any desired content based on a short voice sample.”
Despite these challenges, industry experts like Jutras emphasize the importance of fallback methods. “When in doubt, you can always fall back on traditional methods, such as asking for identification,” she suggested. To address concerns surrounding AI fraud and voice cloning, the Canadian Anti-Fraud Centre (CAFC) remains vigilant. Jeff Horncastle, Acting Customer Awareness and Communications Officer, assured, “To date, the CAFC has not received any reports of artificial intelligence fraud involving voice cloning to gain unauthorized access to victims’ bank accounts. Nonetheless, this is an area that the CAFC will continue to closely monitor.”
Conclusion:
The adoption of AI-based voice authentication by financial institutions presents a significant advancement in data security. By leveraging unique voice characteristics, institutions can streamline authentication processes while mitigating the risks of traditional methods. Although challenges exist, such as voice cloning and data scarcity, the application of AI technology offers promise for enhanced customer protection. Ongoing monitoring and proactive measures by regulatory bodies like the Canadian Anti-Fraud Centre are crucial to ensuring the integrity of the market.
