AI posing as a dim-witted grandma entangles scammers
Artificial intelligence, or, to be more precise, Large Language Models (LLMs), have been used under the hoods of many infrastructural systems for a good number of years now, with the public being only vaguely aware of the fact. In medicine, AI helps with diagnosing; in research, it cracks combinational computations that would take years without it; in city management, LLMs help adjust traffic lights operation and thus reduce the amount and severity of traffic jams. O2, a major telecommunications services provider in the United Kingdom, came up with another way to put AI to use that us, the general folk, benefit from. They invented a grandma that speaks to telephone scammers for as long as they have the patience to, effectively wasting their time and thus keeping people safe from them.
Phone scamming: a major problem
According to Truecaller, one of the providers of antispam solutions for phones, an average American receives up to 8 spam calls a month. A piece published on Comparitech sheds light on the severity of the problem: in 2022, Americans gave scammers around $39.5 billion. And it’s not only about the U.S., fraudulent call centers established with the sole purpose of cheating people out of their money operate all over the world.
An original take by O2
Many telecom companies offer tools designed to shield users from malevolent callers. There are also independent services like the aforementioned Truecaller that guarantee a respectful degree of protection. However, until recently, the methods of countering the risk revolved around blocking numbers and not letting the wrongdoers reach the actual person. O2 came up with a different idea: stall them, keep them talking, trying to get hold of the info needed to steal the money, or convince the respondent to wire a certain sum to an unknown account. The trick is, the respondent isn’t a human being, nor do they have anything that can be stolen. It’s an AI.
Elderly people are some of the primary targets for phone scammers, so AI engineers at O2 didn’t think long when deciding who to shape the artificial person after. Moreover, Daisy – this is how they named the model – was trained on real anonymized scammer data, so she really can waste a lot of time and resources of scammers by pretending to take the bait and playing by their rules, but not quite to the end.
So far, the effort pays off: Daisy the scambaiter can hold up a conversation with a scammer for up to 40 minutes, draining whatever strength there is in that ill-meaning individual. Sure, it’s only a matter of time before this sword loses its edge, but then some indigenous folk in the AI field will definitely come up with something new and even more effective.
