How to get revenge on phone scammers using artificial intelligence

A phone service called Jolly Roger uses ChatGPT and voice modulation software to create and read scripts to phone scammers. Acting as digital, fictional characters, Jolly Roger can keep phone scammers on the line indefinitely – wasting their time and giving great pleasure to the real owner of the phone, informs The Wall Street Journal.

The service, which launched seven years ago, introduced ChatGPT capabilities over the past few months. As a result, phone scammers remained on the line for up to 30 minutes, said company co-founder Roger Anderson.

After paying for the subscription, users get access to the service, which is a cloud phone system. There are also apps to view, listen to, and share recorded conversations between a bot and an annoyed phone scammer, as well as ways to create your own annoying characters.

Scam calls, which typically start with an automated robot and then connect the caller to a real person, have gotten even worse since 2016, when Anderson and Steve Berkson founded Jolly Roger.

In February 2016, 2.28 billion spam calls were made in the US. This number rose to 5.19 billion in January 2019, peaking at 5.66 billion in October 2019. There were 4.7 billion spam calls in November 2022 and 5.08 billion last May, according to YouMail’s robocall index.

The previous process was that Steve and I would sit in front of the microphone and dictate the required phrases to the actors,” Anderson recalls, implying that the work used professional voice actors to record various scenarios.

Now, thanks to ChatGPT, we have been able to automate and significantly enrich the whole process – add more characters and different situations in order to keep phone scammers on the line.

A powerful chatbot generates a script which is then passed to Play.HT − artificial intelligenceA that converts text to speech. Play.HT voices the scenario with the voice of the selected character.

To create these characters, the authors turned to the help of friends and relatives. For example, a character called Whitebeard speaks in the voice of Berkson’s father, Sid Berkson. It took only 45 seconds of audio recording to clone his voice in Play.HT.

Despite this, the first steps were not without trial and error. For example, when the system was asked to “take its time,” ChatGPT was initially uncooperative.

“I had to experiment with super hints because the system initially said, “As a useful AI model, it’s not in my nature to waste time,” Anderson said. “We ended up tweaking the super hints and making the bots be just the right amount of useful but also fun and interesting.”

“The Wall Street Journal” described the details of how the service works. The user has the opportunity to try on a certain role: an example is the naive pensioner Whitebeard, who confuses intruders with questions about how to set up the TV remote control. When asked about his credit cards, Whitebeard replies: “I have one with a picture of a cute kitten and one with a beautiful beach scenery.” Other characters include the housewife Salty Sally and the frivolous Whiskey Jack.

At the beginning of each phone conversation, the chatbot takes time to talk about problems with the TV remote or the bee sat on the arm, thus giving the GPT-4 a few minutes to process the introductory phrases of the telemarketer or scammer and generate appropriate responses. Once the text generated by artificial intelligence is ready, it is sent to the voice generator, which continues to carry on the conversation. Such conversations can drag on for up to 15 minutes, and the call is usually terminated by the initiators of the call.

Of course, Jolly Roger only keeps a fraction of active scammers on the line, but its recent integration with artificial intelligence comes at a time when many countries are looking to crack down on call centers that bombard American consumers with robocalls.

Until there is a world where there are no suspicious calls from unknown numbers, Jolly Roger will do everything possible to reduce irritation.