Maybe social media algorithms don’t suck
1 Insulting my Readers
People keep complaining about how their twitter feed is infested with politics or relationship discourse, how Youtube keeps showing them Anime Girl Butts: A 5 Hour Review, or Facebook serves them up AI generated slop that clueless old-people upvote. They bemoan the tyranny of the Algorithm and talk about how social media companies are Out To Get You. That said Algorithm is fiendishly clever. That said Algorithm only cares about what grabs your attention, not what you want. That said Algorithm sucks.
To which I say: Maybe you suck. Have you ever thought about that, eh, Lesswrong? No, seriously, consider it. Maybe this is just a skill issue? Maybe you’ve put your locus of control into FAANG’s hands and are wondering where things went wrong? Maybe you don’t know how to interface with algorithms? Maybe you don’t focus on what you want to see more of?
The answer is yes.
[1]And you’re in luck, because I’ve invested literally any effort at all into using algorithms. Which, shockingly, is enough to not only beat back the trash infesting the rest of your feeds, but remove it all together.
I’ll give you an example. You know what was the last big issue I had with my feed? Too much art. I was literally inundated by a sea of beautiful things. That’s it.
You can outwit social media algorithms. They are little ML systems that generalize poorly, suffer from catastrophic forgetting, and only give results as good as the data they get. They’re dumber than my friend’s sister’s friend’s dog. And like a dog, you can train them if you try, literally at all.
OK, time to stop insulting you and get to what you should do.
2 Giving Useful Advice
Treat social media algorithms as if they’re a student or a dog. Literally view it as pavlovian conditioning[2].
You have two core options: show the algorithm what you want to see more of, and show it what you want to see less off. There are several, platform dependant ways to do this. Take Twitter as an example: Twitter treats the following as “show me more of this”, in roughly ascending order: what you spend time looking at, what you click on, what you reply to, what you like, who you follow and what you retweet/quote. Conversely, it treats the following as “show me less of this”: what you ignore, who/what words you mute, what you say you’re not interested in. These are your tools. Especially likes/retweets. Youtube is similar, but it lets you dislike things.[3]
On the margin, do not treat these as expressing social approval/disavowal. Humans will interpret a retweet or a like as a social act. The algorithm will not. This can suck for the people you’re interacting with. A dislike can weigh heavily on scrappy young Youtubers, and indeed result in Youtube deciding to stop promoting their video. But if you want control of your feed, you must be willing to negatively reinforce what you don’t want to see, and positively reinforce what you go. Not what you think is socially acceptable. Not what you should want. What you actually want. [4]
The algorithm will immediately show you more of what you engaged with. If you click like on a piece of art, it will start showing you some art ~ immediately. If you click on another, it will show you even more. And then if you like it on top of that, it will get extremely excited and flood your feed with art. The algorithm is dumb.
You are almost certainly far too focused on what you don’t want to see, and are by default telling the algorithm that you do want to see whatever engages you. This is how you get bait.
The algorithm doesn’t generalize well. If you click on a piece of art, it will just start showing you vaguely related art stuff. It doesn’t that know you are a fine connoisseur of AsukaxShinji mpreg pics. It just knows you like something vaguely anime related. Perhaps even NGE related. And that’s what it will show you. Take me: I clicked on a video about the economics of nuclear reactors, and watched the whole thing and clicked “like”. My Youtube feed then showed me 5 more videos about nuclear economics. The algorithm is really dumb.
Likes are super-weapons which can radically alter your feed: use them wisely.
The twitter algorithm will stop showing you things after a couple of days pretty quickly if you don’t engage with them. Youtube, somewhat less so, but it will still do it.
Use negative reinforcement signals to clean your feed. Do so with extreme prejudice. Do this as soon as you start seeing signs the algorithm is showing you new sorts of content you dislike.
You will regularly have to “clean-house” and re-teach the algorithm what you like because you slipped up and spent 10 minutes reading a Hegelian e-girl’s synthesis of the dialectic between Trump and Greenland out of morbid curiosity.
After cultivating a habit of telling the algorithm what you want to see more of, you may find it starts to become intuitive.
“Oh, I shouldn’t click on this politics thread, that will just show me more rage-bait.”
“Wait, I don’t actually like this guy’s tweets. In fact, they’re low signal. Muted.”
“Click on profile. Like 8⁄10 tweets. Follow.”
At which point, you too will feel a sense of distance when people complain about how terrible their “For You” feed is.
With thanks to norvid_studies, kit, imit, @Croissanthology, @Tomás B., @lsusr, tassilo and Taylor G. Lunt for giving feedback.
- ^
A friend notes that this post reminds them of a common “two-lens view of poverty, diet and weight, …, that is: for any person, ordinal personal factors always ‘cause’ why they’re somewhere in the distribution of people. but at the same time ‘the economy’ or ‘the food environment’ move the entire distribution. and in secular fashion will raise or lower everyone independent of the individual treading water. that’s how i’d think about ‘algorithms’ funneling and translating attention into semi stable loops. that said there is plenty of room to blame the user.” This is a good point, and as AI gets better, I expect the skills required to interface with SM algorithms to become too high for anyone unless we’re watched over by machines of loving grace. Even now for some poor souls, learning to train social media algorithms is too great a challenge. Frankly, we were dealt a bad hand with the social media landscape we got. Yet it is still possible to win with a bad hand. This article is not for the poor souls who can not. It is for you, dear reader. You are strong enough to win.
- ^
For a case study in pavlovian conditioning, see @lsusr’s post Training My Friend To Cook. (His friend was glad of it.)
- ^
A fellow traveller on Twitter tells me that my advice basically doesn’t work for Instagram as its algorithm is much harder to train. I’m going to stick my neck out and say “skill issue”, but he may well be right. It wouldn’t shock me if some algorithms are more inclined to slop.
- ^
Of course, the optimal amount of control of your feed is not 100%. You can, and should, somtimes use the tools available to you to engage in digital social life at the expense of confusing the algorithm.
This is plausibly good advice, but none of it makes the case for social media algorithms not sucking in the way I care about (and which I think is what most people intuitively will read the title as saying): having a net societal effect towards social media addiction, polarization, and outrage.
Edited this response. Previous version didn’t convey all of what I meant and felt vague to me upon re-reading it.
Yeah, the main thing people care about at a far/large scale isn’t what their feeds are like. And that’s a big part of any discourse about social media. But wrt. the near/small scale, I think people mainly care about “why am I getting so much slop/engagement bait?” That latter perspective is what I was focused on.
Nonetheless, many people must’ve read the title and (reasonably) assumed I meant the far/large scale. So the title was misleading. My bad.
my antidote for this is to consume a lot of media from the opposite side. and spend a lot of time in “enemy” territory trying to find the wisdom of people i disagree with. consider the trained up ideologies to be a form of compression over true people’s desires. what you call polarization, i call specialization.
i also think this is a very uh news thinkpeice eternal-september way of looking at social media. by and large interaction on social media is wholesome entertainment and commerce. that people get in vitriolic fights is just the nature of the agora. i really don’t take “disinformation” seriously either. in a state of nature everyone is wrong about everything and only on modern internet do people regularly encounter whole other lives and worldviews.
it is good and right for you lesswrongers to continue doin research and longposting here. this place is something special. i feel like twitter is kind of the street epistemology of rationalism. could be good. could get you hurt. not for everyone. not every place should be like it.
the addiction is real tho.
Seems like a lot of work and a lot of side effects just to induce it to go way overboard and do a crappy job of showing me something maybe a bit closer to what I want. How about if I just don’t visit the site unless I want something specific, and then I use search to find it?
I spend very little effort optimizing my feed and I get to see a bunch of interesting stuff, have fun conversations with friends, and make new ones.
As an addendum to the post, I note that I unlocked a lot of fun on twitter when I realized I can treat it as a messaging platform for my friends, or basically anyone I want to talk to. This realization happened after meeting @Croissanthology and seeing his (its?) fearless use of @ and total rejection of algorithmically curated feeds.
still way worse than just giving me a vector embedding ui that shows me everything recent and lets me browse it. I used to do this, it’s how I got the videos that I used to share here, but a while ago I switched to only using youtube in incognito mode and restarting incognito if the recommendations seem to know who I am particularly much. seems to result in a lot less time spent on youtube.
wait can you explain what you mean by vector embedding UI in this context? How did you do that?
Maybe it’s not the algorithm that sucks, but the interface—specifically that it conflates algorithm training with content consumption. Perhaps the main page should not update on your clicks, just show content. A separate interface should be used to pick content you want to see more or less of.
The interfaces do suck, yeah. You could design something more conducive to flourishing in a weekend.
I have been thinking in the ‘Pavlov-ing my Algorithm’ mindset for years, and there is a failure state I would like to warn about.
It is possible for an algorithm to pick up on you trying to train it, then purposely show you some bad things, so that you feel the need to stick around longer, so that you can train it properly, all the while, you see incremental progress in what the algorithm is showing you.
I have failed in this way, the training becomes a meta game atop the algorithm, and for a certain type of person, that meta game can be more engaging than the content itself.
facebook knows that some people hide posts to “archive” them, not because they don’t like them. i wonder if the platform you are using thinks this, and maybe you have to ignore bad content with your mind and scroll past rather than ignoring it with your hands.
This advice is correct, and is of a similar valence (though lesser magnitude) of the observation that you could safely use heroin
A thing I’d like in most of my feeds is “different modes”, i.e. “thoughtful highbrow mode” and “give me the embrassing dopamine hits” mode.
Hell yeah. I loved reading this.
I plan to continue basically not consuming content from any of these platforms, tho.
Hell yeah. I loved reading this.
I plan to continue not consuming content from any of these platforms, tho.
A reminder for people who are unhappy with the current state of the Internet: you have the option of just using it less.
Nice I like it.
A random thing here as well is to have specific accounts focused on different algorithms. (The only annoying part is when you watch a gaming video on your well-trained research youtube but that’s a skill issue.)
You can delete Youtube videos from your watch history if you don’t want it to be used for recommendations. I do this. It would be nice to have an easier way to switch through preference profiles than switching accounts though, that seems like a hassle.