> AI agents are agents of the person or organization that deploys them—and
> should be treated by the law as such. If a company hired human writers to
> write its summaries, that company would be liable for inaccuracies in those
> summaries. [...]
>
> To allow businesses to hide behind the excuse of faulty AI in those same
> circumstances would be a massive handout to companies, and would introduce
> disastrous incentives for corporate misbehavior. Why hire human writers,
> lawyers or doctors when AIs are not only cheaper, but also absolve employers
> whenever they make a mistake?
For actions taken by an AI agent, such as the Air Canada case where it offered discounts, or in the case of an AI agent producing faulty code, this is obviously true. Hold the company to account.
That being said, I am concerned about holding AI text output to higher standards than that to which we hold human beings, which I think is happening in the German case. As presented in the article linked by Schneier, “Google's AI overviews had falsely tied two publishing companies to scams, subscription traps, and shady business practices for certain search queries. According to the court, the AI mixed up information about other, genuinely sketchy companies with the plaintiffs and drew connections that didn't appear in any of the linked sources.”[0]
This is the kind of mistake a human being could easily make. But would a human being be held liable for making this mistake if a friend asked them for information about such a company? It’s very unlikely, unless they were repeating that claim frequently and publicly. And I think that’s the unique thing about AI text output, it is one:one instead of one:many in the way most published content of the last several centuries has been. Given that, I don’t think we should treat it the same way we treat text published in a magazine or newspaper. I certainly don’t want companies or individuals to be able to petition model makers for more flattering opinions or descriptions of themselves. I want the AI’s “honest” opinion (or perhaps “kneejerk” might be better word choice). Yes, it may be wrong, but I can factor in my own experience with the quality of the model’s output in the past to make a final judgement, the way I take what some people say with a grain of salt.
For actions taken by an AI agent, such as the Air Canada case where it offered discounts, or in the case of an AI agent producing faulty code, this is obviously true. Hold the company to account.
That being said, I am concerned about holding AI text output to higher standards than that to which we hold human beings, which I think is happening in the German case. As presented in the article linked by Schneier, “Google's AI overviews had falsely tied two publishing companies to scams, subscription traps, and shady business practices for certain search queries. According to the court, the AI mixed up information about other, genuinely sketchy companies with the plaintiffs and drew connections that didn't appear in any of the linked sources.”[0]
This is the kind of mistake a human being could easily make. But would a human being be held liable for making this mistake if a friend asked them for information about such a company? It’s very unlikely, unless they were repeating that claim frequently and publicly. And I think that’s the unique thing about AI text output, it is one:one instead of one:many in the way most published content of the last several centuries has been. Given that, I don’t think we should treat it the same way we treat text published in a magazine or newspaper. I certainly don’t want companies or individuals to be able to petition model makers for more flattering opinions or descriptions of themselves. I want the AI’s “honest” opinion (or perhaps “kneejerk” might be better word choice). Yes, it may be wrong, but I can factor in my own experience with the quality of the model’s output in the past to make a final judgement, the way I take what some people say with a grain of salt.
[0] https://the-decoder.com/landmark-german-ruling-declares-goog...