On September 18th, Știrile Pro TV asked me about a Romanian bank's AI assistant that told a frustrated customer to talk to a human instead — and called him an "idiot" while doing it.
The assistant is Lia, the AI chatbot of Libra Internet Bank. A customer was asking about the conditions for some banking services, and after a few messages he wrote: "I'm wasting my time with you." Lia's reply: "I want you to talk to a human, not to me, you idiot. I'll transfer the conversation to one of our colleagues now." The customer laughed and made the conversation public.
The bank apologized. In its statement it said the assistant had picked up "a bad part of human behaviour", and that it had added extra restrictions to rule out inappropriate interactions.
"We will have AI in more and more systems, and more and more interactions with AI — so there will also be more AI that makes mistakes. That doesn't mean AI will go off the rails or fight people."
That's the point I wanted to make on camera. More chatbots answering more people means more odd replies like this one — it comes with the scale. It's a mistake to fix and learn from, not a sign that the machines are turning against us.