> For example, much of the right-wing US media audience primarily consumes Fox News, while non-right audiences are spread across outlets.
In the context of US Google News, why wouldn't Fox News be a legitimate news source? I mean, it's not a high quality news source, even before the News side was basically subordinated to the needs of the Opinion side post 2020 election, but Google News was not the legitimizer, the American public was. Google went where they went, and while I'm no fan of Google News either, to their credit no one news organization tends to dominate search unless that organization is literally the only one covering a story (and sometimes Fox is that organization, the left buries stories they don't like for a long time too).
> All experiments should be opt-in. We don't know the effects of algorithmic recommendations, especially because they constantly evolve.
All product decisions including the algorithm in-use are a matter of someone's professional judgement in the end. When you don't trust their judgement, you shouldn't use their services and people do make that choice all of the time. I think this is conceptually severable from the A/B psychological experimentation that SaaS companies engage in on their existing userbase which changes the product for a subset of users in order to evoke and examine a response.
> Because the government's job is to make it illegal to do things that are dangerous. Why does the government get to decide that cars have published safety ratings or that food has lists of ingredients?
To "make something illegal" is specifically in the purview of legislatures in our system of government, and the legislatures of both the States and the Federal government have a lot more other work to do besides going around looking for things to make illegal. If anything, the laws that make up our collective criminal codes could see a reduction rather than further expansion and we would be better for it. To "make something illegal", you also need to make the political case for it to be something the legislature takes up, and since that's what I'm asking you to do here, it's a bit circular to then say "because it's the government's job".
Also: websites and internet services are not automobiles, food or drugs. They don't share the same challenges, issues, and risks. To the extent we tolerate government regulation of websites so far, it's to suppress things that are illegal in meatspace too (child pornography, money laundering, conspiracies to engage in various criminal activities like selling drugs and human trafficking). The presentation of a website or internet service, manual or algorithmically-generated, is unique to that website or internet service.
> If (for example) Facebook wants to be responsible for its algorithm's recommendations, then sure, take the government out of it. It becomes Facebook's speech, which is protected.
> But Facebook is hiding behind Section 230, and that should come with additional oversight and restrictions.
They do, and they are, but their liability is limited by Section 230 and I'm not sure that it shouldn't be. I keep going back to re-read Section 230, and to be honest it's one of those rare statutes that I think is exactly what it needs to be in this time and nothing more, and nothing less. Considering how bad and unclear most statutes are, that's something of an accomplishment. There is just enough room for interpretation for the Judiciary to examine its interactions with other statutes and they'll get to it when they get to it, on narrow grounds. I thought Twitter vs. Taamneh had potential to ever so slightly narrowly constrain Section 230, but the way that case came out, it didn't even come close to addressing the possible interaction between Section 230 of Communications Decency Act and Section 2333 of the Anti-Terrorism Act because the holding was that the Plaintiff failed to state a Section 2333 claim in their allegations.
> Seeing how something worked doesn't tell you how to build it yourself. If TikTok tells me how it recommended a video to me, that information isn't nearly deterministic enough for me to reverse-engineer their algorithm. In fact, their algorithm is impossible to replicate without all of their training data (which remains proprietary).
I think it still tells you a lot, even if it is insufficient on its own to build it yourself, it is sufficient for actors on the network to aggregate how a service comes by its recommendations and gives them additional insight into how to exploit it for their own uses and it still comes up against the issue of where is it the government's place to insert itself into product decisions like this and take this choice out of the hands of the product designers of websites and internet services?