Against telephone scams, a British operator creates an AI-generated “grandmother” to waste scammers’ time

Operator O2 has unveiled “Daisy”, an artificial intelligence that imitates a chatty grandmother to keep scammers on the phone for as long as possible, and thus prevent them from trapping other people.

What is the “worst nightmare” of scammers who try to extract money from you over the phone by pretending to be your bank, technical support, a delivery person or something else? Maybe a grandmother who likes to make conversation a little too much.

It's the idea of ​​the British telephone operator O2, which unveiled on Thursday November 14 a program called “Daisy”: a generative artificial intelligence (AI) designed to imitate a talkative grandmother and keep scammers at bay. wire as long as possible.

In a video (obviously illustrated using another generative AI), the operator gives selected examples of the use of Daisy, in which the program with its slow and fairly monotonous voice makes its interlocutors go crazy. Daisy pretends not to understand their instructions, makes them repeat them, explains she wants “just a little chat”…

Et “while they are busy talking to me they can't scam you”explains the program. The operator states in its press release that Daisy has already “kept numerous scammers on the phone for 40 minutes”.

How are scammers identified and redirected to Daisy? Contacted by franceinfo, O2 specifies that it has worked with professionals who trap the scammers themselves. The operator explains having used “several tactics including 'number seeding'” : deliberately letting Daisy's number be included in the number lists used by scammers (“mug lists”), so that they try to call her… and find themselves caught in their own trap.

How does AI work? The operator claims that “Daisy” “combines multiple AI models that work together to listen to the caller and transcribe their voice into text. Appropriate responses are then generated using a large language model”a program like ChatGPT, “endowed with a layer of 'personality'” – probably thanks to a written instruction (a “prompt”) which asks him to adopt the perspective of a grandmother focused on the conversation.

The written responses are then “returned to a custom AI text-to-speech model to generate a voice response”and this “in real time”explains O2. However, AI is not a panacea: to minimize the risk of scams, the operator encourages you to transfer suspicious calls and messages to a local number, and to call your loved ones directly if an unknown number is impersonated. for them.