In the spirit of “The Player,” here’s my pitch, “Well, we’re building on the proven success of GPT-3, adding in deepfakes, to disrupt the business of sextortion.” For many of you, that’s enough. But if want the details, read on.
Yesterday, I learned that there are “legitimate” businesses arranged around financial sextortion. These companies are mostly based out of India, and the pictures I saw made them look like almost any other Indian call center.
The basic business model that that the employees troll Facebook for lonely guys, posing as young women. Once a target is found, they collect some details, like friends, family, and co-workers, and then initiate a Messenger chat. The conversations are surprisingly quick; it’s usually less than ten minutes before the target is in a video conference. Then it gets even “better.”
Most of the call center employees are male. In order to successfully pull off the sextortion scam, there is a program that stitches together pre-recorded snippets of video. The interface is seemingly built in Visual Basic circa 1997: it’s a bunch of buttons, with labels like “Wink,” “Blow Kiss,” “Smile,” and so forth. In addition, there is always a “technical problem” that keeps the audio from working.
It’s usually pretty easy for them to get screen shots or video clips of the target naked and aroused, and then it’s off to the races. The video chat will be ended, and the FinSex phase begins. This part is does not have any complexity: there’s a threat to release the images if the scammer is not paid.
This is a moral disaster, mostly because this entire thing is so easily automated. We will use a GPT3-type language model to generate the chat. ( As an aside, I’d make a small bet that GPT-3 turns out to be overkill, and a much simpler model converts more targets to victims.) Once we’ve moved into the “sex chat” phase, use deep-fake technology to create the image of the supposed “woman,” and continue to use the speech model to further entire the victim. The screen shots and video clips can be similarly automated. Finally, at the end, use a standard shopping cart solution.
The problem is, this is almost too easy. It means that it’s almost certainly been done. And that’s reeeeally scary. (Also, a “mockumentary” exploring doing this as a start-up would be amazing.)