Fraudsters Used AI to Mimic CEO’s Voice
thenextweb.com
thenextweb.com
The source of that fact is probably a Wall Street Journal reporter, who heard it from the victim's insurer, who heard it from the police.
If they didn't catch the criminals, how is it possible for the police to find out they used "deepfake" or "AI"?
Whilst it's funny these guys did this with AI I can't help but feel that they could have done this with someone who is halfway decent at accents and saved themselves a lot of effort.
Imagine getting a call from your wife. Do you doubt right now if it’s her or something that just sounds similar.
This man could have know the ceo quite well.
but the deepfake thing is a good story.
Would you tell your wife your credit card number if she called you and told you her card is blocked for some reason and she needs to refill the gas tank or buy something?
So I asked who sold him that Music Man Petrucci 7-string in 2009(or so) and after he answered correctly, I followed through with his request.
My point being: it helps to have such a question prepared for everyone - or alternatively - a question concerning some fact about you that only people who know you could answer.
On the contrary, it seems like they did save themselves a lot of effort by deploying the de-centralised 'deepfake' AI model. It would be significantly hard for the opportunistic scammers to go out talent-scouting locally, trying to match the exact vocal characteristics of a person. It would also create a vulnerability, notably the person hired to impersonate the CEO.
Do you really think that most employees would need more than that?
But yes, at my current employer, I believe no one would do that just based on an email and a phone call. But then I dare say that faking Swiss German is an order of magnitude harder than faking English.
Someone has compared the typical cadence of Schwyzerdütsch to riding a donkey over very rough mountainous terrain, and I'm inclined to agree.
Also while Bitcoin community is creating new standards for collaboration, like multisig transactions and partially signed transaction format (which would help here in creating a bank-independent solution), I don't see the same standardization happening in the banking system.
There's nothing in this scam that implies the AI is in any way responsible for the actual content of the conversation.
The Turing test is "determining whether or not a computer is capable of thinking like a human being"
Not whether it can mimic the voice and speech patterns of one.
A computer capable of thinking like a human is "strong ai", or "general AI", or any of a few other poorly defined terms.
The turing test is a poor approximation of that goal. Computers can and will pass the turing test long before they're actually capable of real human thought.
Specifically, the turing test is merely having a human judge evaluate a free-form text-based conversation with a partner, and them being unable to distinguish a computer from a human in that scenario.
The main thing this instance lacks is the free-form nature of the conversation. It also lacked the humans knowing that it was a test, so they didn't properly "evaluate" it regardless.