An emerging means of online security could prove to be one of the safest means of controlling fast-developing artificial intelligence applications, assuaging some of the deeper worries over AI’s potential threat but raising new dangers of its own.
Voiceprint technology may offer a safe way of controlling access to AI. Already in use in the military, it can ensure that only designated controllers can issue spoken instructions to AI-operated tech.
It works by converting the human voice into data used as a unique signature for the person concerned, which can then verify who is giving the orders.
Al Majalla looks at how voiceprint technology works, how it has already been used and what might lie ahead for what could be an important part of our AI-driven future.
As well as running live on the battlefield, voiceprint is also being used to improve the quality of military intelligence. It has been deployed in the Gaza war, analysing material to help assess its quality via accurate material analysis.
Its biometric blend of security and accuracy means voiceprint offers a way around some of the well-documented concerns over AI’s potential existential-level risk to humanity.
65 million voices and counting
In the meantime, it is already proving itself in the commercial world, with around 65 million voices covered so far. Companies, including major banks and airlines, run voiceprint to authenticate customer identities.
Its biometrics already offer customers faster and more secure access to sensitive data or applications than traditional methods, such as passwords or PIN codes.
Both have long been vulnerable to cybercriminals running phishing campaigns to steal passwords or crack them using so-called brute force attacks.
How does it work?
There are currently two main types of voiceprint technology. One is text-dependent, and the other is text-independent.
Text-dependent voiceprint tech identity is verified when a person reads a pre-prepared text. This is the same technique used to teach speech recognition AI-based apps.
The voiceprint is not recognised if the AI detects a difference in how the person reads the text. Non-text-dependent tech is more flexible and screens for a voiceprint without using specific words.