AI Secrets is your new weekly update on how AI is changing our everyday lives. Our experts will keep it clear and simple, so you can stay ahead of the game. This week we’re looking at deepfakes of the audio variety.Share this with anyone you want to keep up to date.
Your voice is NOT your password đź”’
The speed of AI development is wreaking havoc in all sorts of industries, and the latest to be affected is consumer banking. Many retail banks use Voice ID technology for telephone security checks, but last week Vice reporters succeeded in using a commercial AI program to fool a bank’s Voice ID check, and gain access to a customer’s telephone banking service.
The incredible part of this is not that it happened, but the ease with which it was achieved. Voice ID has never been a foolproof system, and soon after banks began to roll it out in 2016 hackers and journalists were finding elaborate ways to beat it, but until recently the effort required was far greater than that needed for more conventional hacking methods.
AI voice tech has changed all that in just a few short months. Vice used online software - created by ElevenLabs - to generate fake audio of their reporter saying his bank’s audio passphrase. This time last year, a similar outcome would have required several weeks, enormous computing power and many hours of recorded audio to train the AI’s output, but with ElevenLabs’ software, only five minutes of training audio was required, and the output was near instantaneous.
This new software is not just a concern for banking security professionals; for example, trolls have been using the ElevenLabs software to create faked recordings of celebrities, including one of Emma Watson quoting Hitler. Elsewhere, voice actors and audiobook narrators are up in arms over their voices being used to train AIs without their consent, and privacy campaigners are concerned about the increasing number of companies who are harvesting customer biometric data, either through telephone calls or other means: for example, TikTok came under fire recently after changing their terms and conditions to allow collection of “faceprint and voiceprint” data.
The degree to which these developments should alarm you in the short term depends on how widely available your voice recordings are. For YouTubers, podcasters, actors and politicians there is a real danger that their voices could be used to train an AI to impersonate them, to a standard that would fool a bank’s automated security. For the rest of us, given the speed with which these technologies are emerging, that threat seems more of a medium-term than a long-term worry. Much like with video deepfakes, the question is no longer if, but when; to help you stay prepared, don’t miss our in-depth look at deepfakes!
It's not too late for the perfect date!
NYC's #1 rated museum & experience is the perfect surprise date venue - and right now you can get two tickets for just $70 - use the code: DATES23
Microsoft has added a toggle for the all-new AI-centric Bing to allow you (if you want!) to bring back some of the zany, madcap antics of its early days - but hopefully not all of them!
The latest example of facial recognition software resulting in misidentification by algorithm has emerged, and once again it’s a black male who’s been wrongly accused of a crime they couldn't have committed.
The Guardian takes Spotify’s new AI-powered DJ for a test… dance. Can it get the party started with its trivia tidbits, or will it clear the dancefloor with its OpenAI-driven transitions?