>
Why does an AI need a human to feed it what to say? That's not a very good AI.I'd say it's less about what, and more about how. AI can't simulate the emotional layer of the conversation in real-time just yet. Selecting the right words, tone, cadence, knowing when to pause and when to interrupt - it's all intuitive to us, and it's the core tool for the scammers[0] on the phone.
The companies are still trying to save on labor costs, hence outsourcing. The voice modification allows them to have their cake and eat it too - to move their call centers to countries with low labor costs, far away from their targets, while defeating the main heuristic people have to detect scammers: the caller sounding like a foreigner from a far-away country, despite claiming to represent a local company.
Until someone develops an AI for the emotional/manipulation layer, there's no point in adding AI for conversation here: the scripts the scam callers use are cheap, scalable, and work well enough.
> Have the AI hear the human voice, decide what to say that is relevant and useful, then generate a voice to deliver that information, then I'd almost be impressed.
That's actually what's been done for years now. Try calling the helpdesk of any large company, you'll most likely be greeted by a bot that tries to converse with you. It works as well as you can expect a chatbot attached to voice recognition to work - about the only thing it's good at is delaying and denying requests to connect you with a human.
In this case, the company can actually replace most of the support workers with a shitty chatbot because it's you who is calling them - they have no reason to care or make it easy for you to achieve anything.
----
[0] - I classify first-party and third-party telemarketing, cold calls and tech support scams as just different flavors of the same thing: preying on the vulnerable.