Man scammed after AI told him fake Facebook customer support number was real

Re:Man scammed after AI told him fake Facebook customer supp... by maxbond on Friday May 31, 2024 @01:36PM (#40538029)

I think what you've gotta do is say, "You can't call, but here is the number anyway," because customers aren't necessarily interacting with your page anymore. They're interacting with AI summaries of your page. Those AIs might be in house, or might be provided by a search engine. What is tenable or untenable will have to shift to the realities of how users are interacting with the information you present.

If you can't provide their AI with text answering their direct question (eg, "what is the support number for Facebook"), they'll find a document which does provide such text. If it's not you then it's a scammer or competitor. UX for these customers means presenting information in a way that sorts high in a semantic search and is robust to transformation.

If you provide text indirectly answering the question ("that number doesn't exist" rather than a literal number), you're liable to be scored as less relevant than a wrong but direct answer ("the number is 1555 SCAMMER"). You're also less robust to transformations, because you can't pull a valid phone number out of the text.

Or maybe I'm wrong, take any certainty implied by my language as rhetorical. That's just the pattern I'm seeing in these tea leaves.

[ Parent | on HN ]

Re:Man scammed after AI told him fake Facebook customer supp... by maxbond on Friday May 31, 2024 @03:46PM (#40539439)

Also, realistically, I don't imagine the phone number literally just telling you that the service wasn't available and hanging up. I imagine it would offer you options to get various pieces of information (the URL of the website, the legal address of Meta, how to navigate to the support knowledge base on the website, ...) and let you draw your own conclusion about how useful it was. Maybe it's occasionally handy to someone. At worst it's harmless.

I think in an ideal world, you could use speech recognition to let people leave a message, and open a ticket, as if they had emailed support@. When someone responds, the system gives them a call them back and delivers it using text to speech.

[ Parent | on HN ]

Re:Man scammed after AI told him fake Facebook customer supp... by squigz on Sunday June 02, 2024 @11:46AM (#40555050)

I like how we've suddenly accepted "AI" "summaries" as simply the way of the future, despite the inherent problems and repeatedly botched rollouts.

[ Parent | on HN ]

Re:Man scammed after AI told him fake Facebook customer supp... by maxbond on Sunday June 02, 2024 @07:15PM (#40557884)

It could be rejected for sure. I personally don't think it's working well. I don't support it really.

[ Parent | on HN ]