A German regional court has ruled that Google is directly liable for the content of its AI search overviews. According to the court, previous limited liability protections for search engine operators don't apply to AI overviews. In this case, Google's AI had falsely linked two publishers to fraud and made claims that didn't appear in any of the linked sources. The ruling could set a precedent for AI-generated content liability worldwide.
In the not so distant future just about every site will have AI summarization or QnA as a core part.
Instead of searching through endless documentation you ask AI to trawl and give you the answer. This is undeniably useful. But if they give the wrong answer once and suddenly become liable, that’s a potential risk.
How is it “undeniably useful” if it has the potential of giving wrong answers?
Also and perhaps more importantly, are these the lengths people go to avoid reading? If so, we are doomed.
Not everyone enjoys reading documentation. We don’t need to be defensive about this. We already have search that can trawl through a well maintained site.
AI can not only go through the documentation but also translate it to layman and point to the sources.
If it gives the wrong answer 1 in every thousand results, it is still undeniably useful. You shouldn’t blindly trust AI is common place knowledge. And it’s no different than doing a Google search for something and some times clicking into a result that is bad. The fact that that possibility exists doesn’t change the fact google is "undeniably " useful.
I’m going to be straightforward with you and say that if someone doesn’t want to read documentation, they shouldn’t be doing the job the documentation is for.
I’ve been bitten by AI summarizing documentation so many times, these days I refuse to use it for that purpose anymore. It’s just not worth it. It creates a loop where it wants to try things that don’t work, walk back, try something else, repeat, and spend $10 worth of tokens in the process.
You say that I shouldn’t blindly trust AI like I shouldn’t blindly trust Google results. The difference is that AI is presented as an authoritative source in itself. Hell, most of the time LLMs don’t link sources unless explicitly asked for. And here’s the thing, if I have to go and read the actual sources, it isn’t doing anything significantly more time efficient than just text search, but it is doing it at ten times the cost.
A potential risk that any company implementing an AI for something as simple as a Q&A should be aware of prior to doing that.
If they don’t want the liability, then just don’t use AI for public facing functions. Its not difficult.
Hopefully not, and this ruling goes some way to ensuring sense prevails. It’s a little different if the LLM providing the “AI” summarization has been trained exclusively on the contents of the site; that ensures that only the work of the site authors is used in generating the summary, which means it’s their words, and also probably less likely to hallucinate.
I deny it. The results of an LLM being used to answer a question are far too often wrong to ever be trusted. Sometimes the errors are obvious, much more often they are subtle and harder to spot, but delivered with certainty none-the-less. This ruling ensures that the ones providing the LLM summary are held liable, in the same way they would be if a human wrote the same summary.
Correct, and that is as it should be. Apply the same logic to a human written piece and you will see that.
It’s very obviously not even the tiniest bit useful and, in fact, is simply a huge liability that could be done safer and cheaper by a person.
There’s a difference between you making use of a tool and you publishing the results of that tool.
Why should a vendor be able to make false claims about a product with impunity?
In the not so distant future you use your own personal AI to do the trawling and if that thing gets it wrong that’s on you or the company that made it.