If the user can search like in Youtube then how do you rank the results? That's also an algorithm.
It isn't pretty easy to solve at all.
If you search on youtube then it can rank any way it wants, just not use e.g. anything from the viewing history. No "related videos" column. That's what YouTube used to be. But YouTube (unlike TikTok) worked well before it had rabbit holes.
For TikTok the situation is worse. Their whole app just doesn't exist unless you have the custom feeds. This would make YouTube be 2010 youtube, Instagram be 2010 Instagram (great!) but it would effectively be a ban of TikTok's whole functionality (again, great!).
But the thing about regulation is that it doesn’t need to be water tight. You can just target a small handful of large players and it will improve the situation in practice. It doesn’t matter if 998/1000 apps use addictive feeds if the largest two apps don’t and they have 90% of users/views.
If you regulated domestic companies out of existence, global options would pop up in their place. You could try to block them all in app stores but people would go to the web views.
Obviously we need the ability to regulate also global options. Typically if these actors truly become big, then they have a presence in their "target" countries, such as ad sales.
Then all of the people who have trouble with self-control on infinite feeds can enable this mode, and everyone who wants the recommendation algorithm can leave it on.
This is the optimal outcome that actually serves everyone’s personal goals for using these platforms. If we get into a conversation where some are demanding we don’t allow anyone to use a recommendation algorithm because they feel the need to control what other people see, that’s a different conversation. That conversation usually reveals other motives, like when people defend the algorithm sites they view (Hacker News, Reddit, whatever) but targets sites they don’t like TikTok.
I’d love for there to be an option to have them as default. It’s obvious ($$$) why they won’t do that unless forced to by regulators.
IMO they should not only be opt-in, but should actually be required to publicly list the parameters and weights they’re using and allow users to tune those weights.
> IMO they should not only be opt-in, but should actually be required to publicly list the parameters and weights they’re using and allow users to tune those weights.
I wonder how many people here know that many of the popular apps have rolled out finer controls for recommendation algorithms so you can do this. On Instagram you can go in and see the topics your recommendation algorithm picked up and modify them manually if you like.
I think the goalposts will just continue to move, though.
People here know that they have finer controls (which are still not actually that fine and also don’t really make the parameters auditable). The problem is these settings are hidden away in places most people will never look. And also, I stress again, none of this is actually auditable because they treat these as some kind of trade secret special sauce and there’s really no reason society should feel obligated to support or enable this business model.
That actually wasn’t the objection in the article we’re discussing at all.
The objection is that recommendation algorithms show people more content they want to view, which leads vulnerable people (kids in this case) to consume more content.
You can’t put a restriction on people under X years without gathering information about everyone’s age. You can’t confirm everyone’s age without some ID check. You can’t do an ID check based on anonymous tokens (too easily shared) so every age check mechanism has some ID revealing step, either to the company or to a 3rd party like a government entity (which will pinky swear they’re not looking at the data).
This is something EU regulation requires them. Earlier this year the Dutch courts ruled as such, all the way up to appeal. It's just a matter of time before other European courts repeat this ruling.
https://www.reuters.com/legal/litigation/dutch-court-upholds...
If they don't know what they want, perhaps a good use case for the newfangled LLM-search we have now would be "What's an interesting or popular topic I haven't searched for before?" to which the AI will respond with a list of newly searchable terms.
For new users? A search bar and a set of (human? AI?) curated seed recommendations that the platform is comfortable with being held liable for.
Whatever is latest posted across their followings/subscriptions?
You can (and should) argue that such a simple algorithm doesn’t “count”, but fundamentally the exact wording of the grandparent post never works, legislatively.
Lawyers will lawyer.
The problem always has been "(personalized) opaque algorithms". Time sorted by followers isn't really opaque, nor is "sorted by likes" or whatever. The problem is always pulling in parameters that a users either has no active control over or are so variable they effectively could be random.
users coarse geographic location? Fine
AI detected language of the content? Fine
global popularity of the video clip? fine
user's past behavior: number of videos with similar content they watched? Average number of seconds this particular user usually waits until scrolling further?
The pattern is obvious. Personalized algorithms is what's targeted. Let's keep the discussion intelligent.
Furthermore, bills have been brought to EU parliaments that have erroneously attempted to ban all forms of ranking, which would include even the most basic information retrieval algorithms. So it isn't obvious at all what is meant by 'algorithm'.
I don't consume any content from my friends on something like tiktok where I'm interested in discovering people that have good content under topics I'm interested in. I don't know who those people are and I want to discover new ones that come up not just follow some already popular accounts.
The whole idea here is to make content consumption more deliberate and mindful rather than just opening the app and veging out to an endless feed of slop.
Youtube won't show you anything at all if you have a new account with watch history turned off. It says something like "turn on watch history and watch videos so we can recommend some for you".
Any ordering is an algorithm technically, so yes just "banning algorithm" doesn't work.
A better alternative could be "the algorithm must be public and reproducible by the user".
"Sort the posts of the people I follow in chronological order" you're good
"Sort the posts by the output of a blackbox trained on user data" too bad you're a publisher and are responsible for what people post.
Algorithm in this context (and presumably in any proposed legal text) is about personalization and purpose.
No one worries about presenting content based on total popularity, coarse geography. user's browser language, or anything like that, regardless of whether the actual ranking algorithm (in the CS sense) is an algorithm. Yes it's a terrible name for what's being discussed, but let's not lose focus on the purpose because of that.
Rank them by best keyword match from their search query, if match is equal, order them newest posted to oldest posted.
Done.
"So the user opens the app - what is the first video you show them?"
You don't. How about that?
Like social media 1.0.
Its okay if they have some hard problems to solve.
You rely on unambigous, "physical" properties of the videos.
There is a physical property of all the videos: the time of publication.
There is a physical property of all the channels: did you subscribe to it, or not ?
So, you show, in (reverse) chronological order of publication, the list of videos published by the channels you subscribed to.
Now, of course, a brand new user would have no subscription - you show them a search box.
But then, now, your search algorithm has to weight the various channels that match - but your algo can be relatively transparent, relatively auditable, and the same for all users (unless given explicit preferences, and of course national laws, etc, etc...)
I'm sorry, but, I have a "subscriptions" page in youtube or substack, and they're chronological, and they show me what I want to watch. You keep that.
There is a "home" page in both service that is algoritmically built, and they show me crap that the algo want me to watch. You get rid of that.
Do this, and I can consider you a "neutral" actor, and accept that you shift the blame to content producer.
Or, keep the algo feed, but don't take money from advertiser when I watch yet another flat earther video because YOU decided it was trending.
If you want to decide what I watch, and make money from that decision - congrats, you are an editor. You get the earnings, and the responsibility.
Please don't tell me, with a straight face, that the people who build the algo don't "decide" what I watch. If they want to tweak the algo to downgrade the flamewars and outrage and conspiracy theories and violence and abuse, they can. They do not want to, for business reasons. [1]
That's fair, up to a point - we need publications with editors that agree on having "edgy" content. I'm not advocating for blanket censorship.
I did not like social network preventing me from _sharing_ articles about Biden's son laptop (this was actually beyond the law, but somehow they managed to find the resources and programmers to implement _that_, because, at the time, the execs where cozying with a different administration.)
I'm advocating for "accepting your responsibility as an editor".
[1] https://en.wikipedia.org/wiki/Frances_Haugen#October_5,_2021...