I’m told, though, that it’s unpossible to tell facts from lies, so Google will surely fail.
Timex
4198
The number of directly monetized videos is much smaller than the overall content we have been talking about. It’s a different problem.
Additionally, it’s likely that this kind of system will be handled in ways similar to other ways that YouTube de-monetizes videos… namely, by manual reporting from other users, and then after-the-fact demonetization. Youtube has done such demonetization for a long time, and it’s worth noting that it’s got a history of not really working that well, resulting in many false positives/negatives.
I do not believe they have developed autonomous algorithms for determining if a video contains “climate change denial”.
This is wrong. YouTube sometimes shows you why it recommended a certain video. “Because you watched videos by xyz”
The algorithms aren’t magic. Content has a a score, based on certain criteria, and it’s promoted based on how that score maps to the persona they’ve built on you. Force transparency there. Show me the content scores. Let users tune their own personas. If I watched a video about weightlifting I do not want to get scored as an MRA, even if 75% of weightlifting fans cross over with MRAs.
They don’t do this because they don’t have to and it’s akin to KFC revealing their recipe or some shit.
Timex
4200
I don’t think that the representation of the users are necessarily going to be as clear cut as you imagine.
That being said, Google actually has published research papers that talk about how they do this. If you are interested in checking it out:
Heaven. God I hate Reddit’s curation. Just show me all the posts, In order
No, I imagine the categories and persona scores are all generated by ML and the labels are just guids or some shit. Well, they gotta do some fuckin work to make these transparent and human readable.
Maybe the approach to fixing this is strong regulation on data transparency. Firms need to show you all the data they have on you, on demand, in clear and concise formats and let you tune that data.
Timex
4203
This may not be possible to do, while still having systems that function. It may be, but achieving transparency into statistical machine learning models is a currently ongoing area of research. It’s not really a solved problem yet.
While this doesn’t mean that it shouldn’t be a goal, it does mean that if you legislate that it is a necessity, you can be causing problems because it legislating as a requirement that you do something that no one knows how to do.
In some cases, this may be ok, if you’re ok with the result being, “Well this capability would just go away,” like we’ve talked about in terms of automated suggestions.
I’m fine with digital marketing being less efficient for a few years if it means we get control of our data and Nazis fade away.
Seems like a win win honestly.
HIPAA blew up hospitals and HIS systems for a year or two but they all adapted. Created a lot of work and jobs for HIS vendors too. So it was a good thing.
Timex
4205
Ya, but the HIPAA/HIS thing wasn’t really some kind of rocket science. “Putting data into databases” is shit that people already understood for a long time before that.
Kicking an industry in the ass to simply embrace well understood, modern practices is totally reasonable. But it’s also different from regulations that would require the people being regulated to invent totally new technology.
We could also look too the GDPR as a model, maybe? That’s a massive regulation with rules about data, how the use of that data should be communicated to the user, and how users must be allowed to opt in and out. That seems to be like the closest starting place, but I’m envisioning something obviously broader. It’s certainly difficult, but it doesn’t seem unsolvable, and I would argue that it’s a problem that needs to be solved.
Facebook pushing right-wing radicalization for the sake of profit is scummy but not illegal, and any regulation that tries to make that specific action illegal is a non-starter. But with some imagination, you could come up with rules that don’t run afoul of the First Amendment:
- Technological regulation on algorithmic delivery, debated above (to target the tools used).
- A requirement to publish some kind of algorithm policy, with penalties for going against it (to target the unequal application of the tools).
- Introducing some kind of “after the fact” liability for providers that host content if that content can be linked to a crime. This might be a way to make providers more cautious about their moderation without outright regulating what they can host or holding them liable for everything.
No silver-bullet solutions here, but I don’t think we need to just roll over and accept that social media can exert this much societal impact without repercussions.
Timex
4207
This is essentially just a repeal of section 230, right?
I mean, you might as well hold them accountable for everything, the effect would be the same. There wouldn’t really be any feasible way for anyone on the internet to operate with large scale user created content, because since they cannot effectively guarantee that none of that content will be related to a crime (due to both the immense size of the data, plus the fact that it may be related to a crime that hasn’t even happened yet), they would be putting themselves at risk of criminal prosecution if they allow users to create content.
GDPR is great in principle, but the Facebook solution seems to mostly have come down to “agree to let us use your data if you want to carry on using Facebook”. And just like so many regulations it’s much more a barrier to entry than a problem to incumbents.
The US should absolutely have GDPR-equivalent privacy and data protection laws for all sorts of reasons. But I don’t think they’re going to help with this problem.
Mission fucking accomplished, then!
And I do see it as different than a straight repeal of Section 230. I imagine it’s only a small subset of activity they could actually be held liable for, but the threat of “oh, well, what if this Proud Boys group actually does plan some violence here” would be enough for them to want to really step up their moderation game against groups likely to cause violence.
Definitely! And I think this is what we’re seeing with most tech regulation - companies are able to functionally just Terms of Service their way out of them and carry right back on doing what they’re doing. So I think the GDPR is a fine place to start, but now that we’ve had several years of seeing how tech companies respond to it, we ought to go a little further.
Timex
4211
But you’re just joking, right? You don’t in fact want to prevent users from creating content?
But that’s not how it could work.
If you are criminally liable for the content users create, then that means you really can’t let anyone create content, because even if it’s a small chance that those randos will create content related to crimes compared to the overall amount of content created, the liability would still be devastating, and there would essentially be no way to protect yourself from it.
Simply targeting moderation against certain groups “more likely” to commit crimes doesn’t work, because there’s always going to be a chance that some random user, from your literal billions of users, will commit some crime, post about it (even posting about it BEFORE the crime), and then you are liable.
I can certainly imagine a law written to differentiate between 1) a group where people share memes over several months about shooting Democrats whose members then go on to shoot Democrats, and 2) someone who posts to their individual page that they’re going to go on a shooting spree and then immediately goes on a shooting spree.
I know you like to think of things as very black-and-white, particularly when it comes to lawmaking, but we already have laws that allow for some judgment. Laws that govern data breaches often have some kind of “you must notify users without undue delay” language without going into specificity about what “undue delay” exactly means. Because like you say, there are cases where you don’t want to hold the company criminally liable for something they legit can’t control.
The answer is moderation. And I know you say that moderation over billions of users is impossible, to which I respond, “you shouldn’t have billions of users if you can’t moderate them.” So when I say “mission accomplished,” I am not joking but am explicitly saying that part of the knock-on effect of these proposals is to shrink these companies and their outsized influence in society unless they can figure out a way to moderate their billions of users.
This isn’t either and don’t let these fuckers tell you otherwise. It’s just them trying to protect their revenue stream. Yeah, they gotta slow down the models and let us see what they’re doing. Boo fuckin hoo. We don’t need Hyper Personalized Experiences. fuck off I don’t need to buy more dumb shit like an IoT salt shaker.
I love it. That’s what I had in mind with an opt-in restraining order.
In Denmark we already have similar rules for advertising. You can register your address and phone number with lists kept by the government, and if anyone sends you advertising in the mail or calls you with the purpose of selling you something (NGOs and pollsters are exempt) then they’re slapped with a fine.
Advertisers are required to check the lists before they mail out anything, or call anybody. Most of these advertisers have relatively low profit margins, so it’s an effective deterrent, and it gives each citizen a lot of power against businesses who don’t care. If they aren’t serious about compliance, they don’t have a business.
The idea shouldn’t be to reform Facebook, and you’ll never be able to stop people if they want to give away their data.
The thing that really bothers me is that even if you don’t use Facebook, they’re still getting your data from other sources. GDPR is a great first step, but they should keep improving on it.
They should block companies like Facebook from ever owning your data, if you don’t want them to have it. If you’re using Facebook, then you accept them spying on you.
They should really also get around to making a standardized “Reject All” button, that truly means Reject All.
Right now a lot of companies are trying to mislead, or drown you in paperwork when you visit their sites, because they know you don’t have 5 hours to go over everything, which is a huge middle finger to the idea of GDPR.
Timex
4215
I mean, I actually work with the kinds of AI we’re talking about here, as part of my job. It’s a lot more complex than putting healthcare records into shared databases.
This is stuff that is a current focus of major research initiatives.
That’s fair. I work in digital marketing so I have an understanding of the basic concepts. Obviously when you get to the FAANG levels of orgs, their entire business is content recommendation, and their algorithms are going to be more complex.
Tough shit. Regulate it, make them show us what they have on us and let us change our persona.
One of those A’s doesn’t care about all this, actually, and that’s Apple. They sell hardware.
