80 votes

Landmark German ruling declares Google's AI Overviews are Google's own words and makes it liable for false answers

16 comments

  1. [4]
    zoroa
    Link
    I thought this article was worth sharing, because it highlights ways that the push to Gen AI challenges the foundations of how the Internet operates. Most of the legal questions I've seen for LLMs...

    A German court has ruled that Google is directly liable for what its AI search overviews say. Previous case law shielding search engine operators from liability doesn't apply to AI overviews.

    I thought this article was worth sharing, because it highlights ways that the push to Gen AI challenges the foundations of how the Internet operates.

    Most of the legal questions I've seen for LLMs have touched on whether it was legal for companies to scrape the entire internet as input to train their models. But the article touches on the other end, "are companies liable for the output of their models?"

    In many jurisdictions, like Germany, companies are shielded from being liable for 3rd party content that they host and serve on their platforms. That shield is foundational to the operation of sites, like Google or Tildes, that primarily surface content they did not create. The German Court's decision marks the first move I've seen to strip that protection from content synthesized by an LLM, which could jeopardize many company's pushes towards "AI Everything".

    The United States' version of that shield is Section 230, "the law that built the internet". I'm curious whether there'd be a move to follow the German court's lead, and clarify whether Section 230 protects LLM content too.

    Further Reading

    34 votes
    1. [3]
      slade
      Link Parent
      This seems like extremely common sense. You're either quoting someone else's writing or writing something yourself. If they want the freedom that comes with "AI isn't stealing, it's creating...

      This seems like extremely common sense. You're either quoting someone else's writing or writing something yourself. If they want the freedom that comes with "AI isn't stealing, it's creating something new" necessarily includes "my business is creating and publishing novel content", and of course a business should be responsible for the content it publishes.

      35 votes
      1. [2]
        d32
        Link Parent
        Right. Of course. But they (along with facebook) have been trying to push a narrative "It's not us, it's The Algorithm" for quite some time - long before LLMs - so it is refreshing to see them...

        Right. Of course. But they (along with facebook) have been trying to push a narrative "It's not us, it's The Algorithm" for quite some time - long before LLMs - so it is refreshing to see them getting slapped.

        12 votes
        1. slade
          Link Parent
          For sure. I hope it wakes us up to a whole new way of thinking about accountability. I've long ranted about the dangers and societal harms of The Algorithm.

          For sure. I hope it wakes us up to a whole new way of thinking about accountability. I've long ranted about the dangers and societal harms of The Algorithm.

          2 votes
  2. [5]
    Eric_the_Cerise
    Link
    Ars Technica article (I'm intentionally not linking it) on the same court case used Google translate, without a trace of irony, to translate the German court's ruling.

    Ars Technica article (I'm intentionally not linking it) on the same court case used Google translate, without a trace of irony, to translate the German court's ruling.

    8 votes
    1. [4]
      Ganymede
      Link Parent
      I don't get it?

      I don't get it?

      5 votes
      1. [3]
        updawg
        Link Parent
        Google Translate uses LLMs now.

        Google Translate uses LLMs now.

        8 votes
        1. Eric_the_Cerise
          Link Parent
          I didn't even know this. My focus was simply, "using Google AI to translate German court ruling that concludes Google AI can't be trusted to be accurate".

          I didn't even know this.

          My focus was simply, "using Google AI to translate German court ruling that concludes Google AI can't be trusted to be accurate".

          5 votes
        2. Deely
          Link Parent
          And (personal opinion) it's really bad. I have better results directly asking chatGpt to translate phrases between languages.

          And (personal opinion) it's really bad. I have better results directly asking chatGpt to translate phrases between languages.

          1 vote
  3. [3]
    Minori
    (edited )
    Link
    Would it be any different if the user requested the AI overview? I'm curious whether the user requesting something changes the liability in any way. Because of how LLMs work, users can request...

    Would it be any different if the user requested the AI overview? I'm curious whether the user requesting something changes the liability in any way. Because of how LLMs work, users can request "wrong" answers. I don't really want any government too involved in deciding what's true or false.

    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. The publishers sent Google a cease-and-desist letter, but Google didn't respond appropriately.

    This could happen with any LLM. If upheld, this could significantly affect how LLMs are offered in Germany.

    7 votes
    1. Greg
      (edited )
      Link Parent
      Yeah, I think the difference in push vs pull here is probably significant (or at least should be IMO). AI overviews strike me as being conceptually closer to broadcast media, to be treated as a...

      Yeah, I think the difference in push vs pull here is probably significant (or at least should be IMO). AI overviews strike me as being conceptually closer to broadcast media, to be treated as a statement that the company or their spokesperson is actively choosing to put out into the world. If a company runs a headline that <corporation> are scammers, however that headline came to exist, it seems very reasonable to hold them responsible for it.

      Direct LLM chats are more like talking to customer service - the company should still definitely bear some responsibility for who (or what model) they choose to employ, but ultimately a reasonable person should probably also understand that if Clive is having a bad day and goes off the rails, he still doesn’t actually have the power to fire the CEO and give the job to you instead, and you probably shouldn’t listen to him if he tells you the lizard people are in the walls watching you. If Clive does something less egregious, like wrongly saying that <corporation> are scammers, he needs additional training not to say things like that. That’s true whether Clive is a person or an LLM.

      If a thousand people all hear from Clive and all his colleagues that <corporation> are scammers, even in direct chats to each individual, then we’re back into holding the company as a whole more strictly liable for allowing a systematic problem like that to happen.

      [Edit] Grammar

      10 votes
    2. updawg
      Link Parent
      AI overviews aren't the same as just talking to an LLM, and if the LLM gives you a wrong answer when you ask for one...that's actually a correct answer. I suppose it would be possible to use some...

      AI overviews aren't the same as just talking to an LLM, and if the LLM gives you a wrong answer when you ask for one...that's actually a correct answer. I suppose it would be possible to use some sort of prompt engineering to create conditions where you could sue, but that seems very risky.

      7 votes
  4. [2]
    CrypticCuriosity629
    Link
    This sounds like common sense. Companies are generally liable for the content they put on their websites, which is partly why social media removed harmful or illegal content. Why would AI be any...

    This sounds like common sense.

    Companies are generally liable for the content they put on their websites, which is partly why social media removed harmful or illegal content.

    Why would AI be any different?

    Plus, ultimately this should be considered a great thing because it forces innovation into accuracy as opposed to complacency with mediocre inaccuracy.

    3 votes
    1. Minori
      Link Parent
      There are better and more accurate models, but they're far too expensive to run on every single query. They're also slower than users expect from Google.

      There are better and more accurate models, but they're far too expensive to run on every single query. They're also slower than users expect from Google.

      1 vote
  5. [2]
    unkz
    Link
    Sounds like Google would have to basically pull out of that market or be faced with absurd liability? Or at least all AI responses.

    Sounds like Google would have to basically pull out of that market or be faced with absurd liability? Or at least all AI responses.

    2 votes
    1. Minori
      Link Parent
      They'd sooner disable the responses than pull out of the market (unless they really want to make a point and anger Brussels for little benefit).

      They'd sooner disable the responses than pull out of the market (unless they really want to make a point and anger Brussels for little benefit).

      10 votes