Hacker News Reader

Comment by CoastalCoder | Hacker News Reader

This reminds me of that recent issue with a Canadian airline, where (IIRC) a court ruled that their chatbot made a wrong, but binding, commitment to a customer.

I'm curious if a Canadian court would hold Meta liable for the man's losses in this case as well.

View on HN

That was a very interesting case. The chatbot in question was not LLM based (the incident was pre-chatGPT in any case) and was simply parroting an out of date or incorrect policy that it had been explicitly programmed to do. It seemed to gain a lot more traction in the press because of LLMs. "Air Canada forced to honor terms and conditions on their website" is a whole lot less interesting.

This FB thing is a case of an LLM simply hallucinating without direct human intervention.

Very different cases from a computer science perspective. My hope is that legally, they don't get viewed differently.

If you outsource functions of your business to a third party contractor you are still responsible for what they do and say. I don't think we should allow companies to weasel out of their obligations because they were dumb enough to let a sentence generator loose in a way that it could make commitments.

Yea, it’s certainly a reasonable argument if the wrong information comes from the company itself.

Yeah, headline is overly broad by just saying 'AI'. From just the headline itself, it'd be easy to write this off as "duh, this guy's a fool", but the AI in question here is from Meta, itself. And, not only is it from Meta, but it's the AI they've put in charge of support.

> And, not only is it from Meta, but it's the AI they've put in charge of support.

Does it say that in the article or somewhere else? I didn’t see that in the article.

You can see it in the screenshots.

It says “Meta AI”, but I don’t see an indicator that it’s labeled as providing support. On my device, it doesn’t say so, and is labeled as possibly “inaccurate or inappropriate “. (It still provides a bogus phone number.)

That's an excellent point. That court decided that an AI agent was an agent in the legal sense. "Agent" is a legal concept - someone acting for someone else.[1] It's what allows employees to act for a company. Otherwise nobody could do anything without signoff from the top. There are limits to agency, but it's a rule of reason thing - you can assume a store clerk has the authority to sell you stuff, and someone whose job is to answer questions has the authority to answer questions. The company has responsibility for the agent's actions within the scope of their authority.

[1] https://en.wikipedia.org/wiki/Law_of_agency

[2] https://www.upcounsel.com/lectl-what-the-california-civil-co...

The situation here is slightly different, though. Meta's AI in their various products is explicitly marketed as an LLM chatbot, not as a customer support channel.

Whether they've been diligent enough in making that distinction (and whether that's even possible) will very likely be determined in court at some point.

Reply on news.ycombinator.com