Hacker News Reader

Comment by why_at | Hacker News Reader

I agree with the ruling, but this makes me wonder if it will be possible to have any AI agent at all if it's consistently applied.

After all, if I can get ChatGPT or Claude to say something false that should count too, right?

View on HN

If it's consistently applied, any AI agent provider has to comply with cease and desist letters that tell the company to make specific false claims

The arguments of the ruling should generally apply even when the AI agent makes false statements it wasn't notified about. But in that case the defendant might have a stronger claim about not being able to reasonably ensure the correctness of all statements, and having taken reasonable measures to ensure correctness. Google couldn't really claim any of that after ignoring cease and desist letters about the false claims

Due to costs of running frontier models on every search request, Google simply does not: the failure rate is so high when you are just expecting an objective output.

Imagine a search for your name resulted in an AI summary saying you are involved with child-trafficking because low capability model linked your first name and perhaps a couple articles on supporting children non-profits to it — and then offering that in a convincing sounding summary right at the top!

And the more you protest the more your name will be associated with child trafficking. Streisand effect multiplied by LLMs being not good in dealing with negative information.

If you give a language model, empowered through an agent, the ability to publish information on your behalf, and it publishes false information which causes either direct or even indirect harm, and you fail to correct it, then yes.. by every conceivable definition already in law.. you are a criminal.

If you knew this was all possible and you did it anyways for personal gain then you are additionally negligent which may add aggravation to your charges.

I think it should be relatively possible for an LLM to know when it's talking about something in the risk zone of libel. And when it identifies such a case it links direct to source material rather than trying to fabricate an answer itself. This is much how a journalist works, refusing to make claims that aren't provable.

Someone needs to hold the liability at the end of the day. People are experiencing real harm from false claims LLMs are spitting out.

In normal flow yes, but likely not if you intentionally entrap it to say something wrong.

A disclaimer and couched language will probably fly through. And it's going to matter what expectations an user could reasonably have, too.

Reply on news.ycombinator.com