I’m aware that I’m falling into the AI trap of personifying these results, writing as if Google’s AI imagines and presumes. Google the company, of course, wants this too–it wants me to see my search results as a conversation with a helpful problem-solving buddy. The company has aims down the line to introduce actual agents that can search for you, jumping on the latest Silicon Valley obsession to basically make a version of Google Alerts that are worse for the environment. Maybe its current version of search is something of a test balloon for an eventual service it can charge for.
But the same way it drives me crazy when Stripe’s AI customer service tells me it’s so sorry, omg that must be so hard to hear about whatever utterly mundane, thoroughly unemotional issue I’ve brought it that day, Google search’s hollow facsimile of the human desire to help lays bare how badly its inventors misunderstand what regular people want and how we feel when we’re searching for information. (Full disclosure: I wrote my library school Master’s thesis literally about this, so it really riles me up!) It exposes how Google sees its users: helpless and intellectually lazy, longing for machines to solve our problems for us, unable to sit in the sometimes uneasy space of unknowing that motivates information searching in the first place. Shortcutting yourself out of this space–or, worse, letting AI do it for you–robs you of the chance to explore and discover, to let your search take you to new places and learn new things.
Are websites still protected by Section 230 when the results are editorialized on the fly by an LLM?
This is undecided by courts.
From what I’ve found from some simple research, it’s possible that if it’s reproducing the same information as another user that the website is protected, possibly even if the LLM modified the exact presentation of that information.
If it’s creating new claims, then I think it’d be likely that whoever is running the service would be liable. Who knows though, since they might just claim that the new claims are from its training data and therefore they are protected.
In an ideal world, we’d just follow Germany’s example and hold the website liable for any new content generated by a LLM. Ideally, if you make shit up, you made the claim, and if you’re parroting a user, the user made the claim.
I’d imagine due to current circumstances, the US would want to shield AI companies though because politicians want money and the SCOTUS is useless.