I completely agree with the article. There is a vast soup of user generated content that could be displayed and there's always an algorithm of some sort employed, even if that algorithm is "show me only content from people I explicitly follow and sort it chronologically".

If you wanted to give the social media companies the benefit of the doubt, you would say "they just want to have a feed that the user enjoys and the best metric they have to tune their algorithm is is engagement". If you didn't want to give them the benefit of the doubt, you might say "They want to surface a feed that keeps you hooked so they make the most money on ads, so they use the metric of engagement".

I also completely agree with Tim that a marketplace of algorithms is the best way to solve this problem. If users can choose their algorithm, they can choose between the "sugar high" content that is optimized to maximize engagement and time-on-platform if thats what they want. If they'd rather choose one that maximizes for other metrics, they are free to do so. Examples of metrics that a given user might prefer: "A monthly survey that the user fills out that asks how good the recommendations are" or "a daily survey asking questions related to mood/mental health/etc".

> And anyhow, those algorithms are just showing you what you want. Don’t try to deny it, if it wasn’t what you wanted you wouldn’t be doomscrolling so much, would you?

I believe this to be based on a faulty assumption: people do what they want. While it may be true that at the reptile-brain level we do want what the Algorithm feeds us, that statement is equivalent to saying that someone who is addicted to opiates wants to feed their addition. It's only true at the most shallow of levels.

The article at large seems to realize that, but I think what is needed is to distinguish "want" as what we as rats in the Skinner box want and "want" as what we humans who do not wish to be prey to that Skinner box. That is the promise of platforms without the Algorithm, just that it wont try to take advantage of users.

(As a side note, I am an advocate for using capital 'A' the Algorithm to refer to the content-aware, black-box recommendation engines that run social media sites. That let's us continue to talk about lowercase 'a' algorithm to refer to sorting and such)

An algorithm that I can tune by having private +1, -1, and "hell no" buttons would be helpful. If I follow too many people and/or hashtags, I will end up with so many posts/toots that I spend a bunch of time reading through things that really aren't that interesting. If I had a way to easily mark content as interesting or not, an algorithm could construct my feed such that the things that I find interesting are more likely to be read. There have been people that I followed because they had an interesting post or two, then the other side of them started posting that is not interesting to me.

As an example, one person posted that they were working on end-to-end encrypted messages. That was really cool, nearly instant follow. This person identifies as a furry and posts a lot more content related to their furry identity than E2E encrypted messages. I'm not really into the furry scene, even as a spectator. Nothing against this person - I'm happy they can be who they are. I just wish there was a way that I could tune into the once in a while updates about the tech project without the part that is of no interest to me. That person may be equally annoyed with my feed if I posted regular updates of my love for mac n cheese.

Importantly, the +1/-1 that I give should be used only for my feed with no feedback to the poser. My -1 does not mean I disapprove and I certainly don't want it to hurt anyone. I don't want it to influence others' feed. A -1 needn't mean that I never see something like that, but a "hell no" should.

Such a system would allow me to follow more people and be exposed to different ideas that come from more exposure. It's then up to me to tune my feed to match my interests and available time.

Most Twitter users, as best I can tell, aren't actually complaining about what they see. They are complaining about what others see.

In their view, they have the "correct opinions" and have not been biased by any algorithms or moderation. Meanwhile, the people that disagree with them are the ones being duped by algorithms, bad moderation, and bad actors.

That's why there's a huge censorship movement right now in the US. People aren't saying, "protect me from what I want" -- they are saying, "protect others from what they want, but leave me alone." Which is, of course, entirely hypocritical.

I read this as a problem with how the term "algorithm" is used by / when communicating to the layperson (basically, the general problem of "buzzwords"). It's just being simplified; possibly overmuch but that's not relevant to the layperson's complaints. I think it's fair to say that the layperson understands that they don't want an algorithm which manipulates them to spend more time "engaged" than they otherwise would.

I do think there is value to algorithms which incidentally increase engagement as argued by the article (e.g. capturing and isolating email spam), but there is understandable push-back against an algorithm that is, additionally or not, being optimized for pupils pointed at screen.

Instagram, youtube & tiktok are a reflection of what you click on.

But, to make them nice, you need to give them both negative and positive feedback at the very early stage.

If you see something that you don't like? Don't ignore it, hit the "..." and select the "get tae fuk" option. Otherwise it'll keep trying that genre of what ever is popular with a similar persona to yours.

The problem is, the "I don't like this" button is often hidden. (in youtube I'm not sure how well weighted the downvote button is).

For mastodon its hard, whilst the data is out there for most people to grab, the models need to be stored and trained somewhere. Your model is a PII risk. Also, there isn't much info out there on how to make a good recommendation engine (for obvious reasons) Unlike object detection/text to speech/OCR and other freely available models, recommendation models are as valuable as the dataset. This means they are rarely made public.

Even if Mastodon doesn't want to implement or play with "algorithms" (ignoring as is correctly pointed out that everything is/has an algorithm), all Mastodon content is ActivityPub content, right? So someone can put together a Mastodon alternative that supports pluggable algorithms, no? No need to convince the Mastodon devs who don't want this sort of thing.

One other thought is that most algorithms don't need to be very sophisticated.

So DuckDuckGo is (mostly) predicated on the idea that if you type "mens running shoes" into a search box you can confidently sell ads for trainers all day, and possibly other sporty goods. The bet Gabriel Weinberg is making is that the accuracy / profit for selling those ads is not significantly worse than using the text plus a thousand data points about that persons pregnant daughter and bathroom habits. Inwoukd love to see some study on that (not sure how - maybe have google return duckduckgo results and see what they buy. Hell I would bet they have already done this !)

Anyway I suspect that I would get as much out of twitter if I just get what David Attenborough reads on twitter (ie a manual curation from someone interesting and well versed in the world)

Sometimes the best algorithm is a much more educated and experienced human

The shock people experience when you tell them any sorting, especially sorting chronologically, requires an "algorithm" is disappointing.

Yeah, this is pretty much what I'm hoping for on Mastodon. On Twitter I would occasionally turn on the "home screen" instead of the "latest" feed to see what's going on in a larger world. Including such a facility (or even multiple alternate implementations of such a facility) would be nice in mastodon.

The other part of the post - the abuse? That part is harder. Fundamentally, commercial social networks spend a crapload of money and traumatize working-class moderators all over the world trying to keep spammers, scammers, illegal pornography, hatespeech, and worse off their platforms. I don't know how sustainable it is for hobbyists to implement that.

Realistically, I'm expecting that if Mastodon truly takes off, it will gradually bifurcate: 1) Free instances that are hives of scum and villainy 2) Paid instances that are well-moderated and pleasant

Fortunately, Mastodon servers have the ability to federate with other servers in a "second class citizen" sort of way, so members of group 2 will be able to follow people on group 1, but the mayhem of free instances will be purely opt-in on a per-user basis.

I mean Mastodon has a lot of problems. Its tech stack is a nightmare to admin, and it is notoriously wastefully chatty when it comes to server-to-server messaging. But those faults can be fixed.

The social problem is much harder.

It seems to me that there is a "solution" that goes completely unspoken in these conversations.

There's the one side that says, "don't do anything opaque, just show me a chrono-timeline, I'll deal with the deluge myself". And another side that says, "you delusional to think you can deal with a deluge".

But maybe there is a third position: don't get into a deluge in the first place. Maybe the answer is actually that we shouldn't be trying to follow thousands of people on social media. Maybe there is no meaningful way to keep track of that many people and still be able to existentially understand them as, ahem, people anymore.

So far, my experience on Mastodon is bearing this out. I have almost exactly 10% of the followership on Mastodon than I do on Twitter. Yet I'm easily having 5x more conversations. The quality of those conversations is very significantly better, but I'll leave that hairy ball of unquantifiability by the wayside for now. So far, "engagement" on Mastodon is 50x better than on Twitter.

It would be nice if there was a way to talk about this subject without immediately devolving into partisan political comparisons. This is a far more fundamental issue with recommendation systems, a conflict between immediate gratification and addiction mechanisms versus deeper nourishment and satisfaction. You can see this in so many arenas of human consumption.

Consider diet. If you log everything I ever consume and then measure the probability I consume the same thing again in the next 30 minutes, what are you going to find? You'll be recommending I eat nothing but donuts, hard liquor, and potato chips.

Consider video. If you log everything I ever stream and watch and then measure the probability I stream and watch something similar in the next 30 minutes, you'll be recommending nothing but ChiveTV-style fail videos and short-form hot takes.

But are these the things I actually most want? Humans are complex creatures with desires and preferences that don't always express themselves in terms of quick re-consumption of similar items. If you ask me my favorite film, I'm going to say Apocalypse Now, not an epic fail compilation. But if you measure what I'm more likely to watch on repeat for 8 hours, it's going to be the fail compilation. If you ask my preferred foods, I'm going to say lean meats and vegetables, but if you measure what I'm most likely to repeatedly eat for hours without stopping, you're going to find a bunch of dessert foods and snacks.

The types of things people consume that are most nourishing and satisfying, since they actually provide nourishment and satisfaction, are not things that lead to immediate re-consumption of the same thing. The types of things that lead to immediate re-consumption are things that are insubstantial, don't require thought or reflection, don't lead to satiety, and things that are addictive.

I have no idea what sort of solution to this scales and will satisfy people that don't like gatekeepers, because the reality is, I've found what films and television shows and music albums I've most liked from top 100 lists curated by experts, and I've found what foods best nourish me and lead to the long-term health and physique outcomes I'm trying to achieve by the same method, expert recommendations from people well-versed in science.

You can't automate this, but to work, the public has to trust a curator, and it seems most of the public doesn't trust anyone to do this, or if they do, they'd rather trust the Critical Drinker and Liver King instead of the American Film Institute and FDA.

Once again I come to sing the praises of browser add-on Tweak New Twitter.

I see just tweets from people I follow, with their retweets available on a separate tab. No trending, no random tweets. Twitter the way I originally imagined it was supposed to be.

No need for any algorithm beyond what this offers, unless you're really keen on discovering random people to follow, and the people you already follow do not retweet. In which case ... you have my sympathy.

If one was to be a purist, you'd remove the boost option.

With a boost you want to increase the visibility of a post by somebody else. Probably because you agree with it. Something has to happen with that boost signal, otherwise boosting is pointless. Hence, in some way a boosted or often boosted post is promoted over ones that are not boosted.

As the boosted post gets more eye balls, it will get even more boosts. Not necessarily because it's so awesome, simply because it's the post that is shown more prominently.

This simple snowball effect has viral potential, hence it's just as corruptible as Twitter. It can and will be used agenda-driven. Power always consolidates, hence soon you'll have the elitist layer with large followers boosting each other's posts. The other 95% gets zero traction or engagement.

The other downside of boosts, retweets, quote tweets is that it discourages writing your own original posts. More than 80% of what Twitter calls tweet activity is simply retweets, not new tweets.

Finally, if you want organic social media, people to follow should never ever be automatically recommended. This triggers the exact same snowball effect.

> Don’t try to deny it, if it wasn’t what you wanted you wouldn’t be doomscrolling so much, would you?

This is the most ignorant and idiotic[1] take I've seen on something like this.

Don’t try to deny it, if it wasn’t what you wanted you wouldn’t be smoking so much, would you?

Don’t try to deny it, if it wasn’t what you wanted you wouldn’t be drinking alcohol so much, would you?

Don’t try to deny it, if it wasn’t what you wanted you wouldn’t be taking oxy so much, would you?

Don’t try to deny it, if it wasn’t what you wanted you wouldn’t be seeing ads so much, would you?

> These ML models know what you want and that’s what they show you.

No, they don't. This is 100% wrong, and coupled with the previous comment, shows a complete lack of knowledge in this area.

These ML models don't know what you want.

They know what you engage with. What you engage with isn't necessarily what you want. And ANY suggestion otherwise is wrong. Simply put, engaging with something doesn't necessarily mean you want to see more of it.

[1] Yes, that's a hard stance to take, but I stand by what I said. But you read this comment, which means you wanted it. Don’t try to deny it, if it wasn’t what you wanted you wouldn’t read any part of it, would you?

What criteria are being used to determine this is what we want?

People also want heroin and we use literal impact on society to curtail use. Freedom of reach is what we’re talking about limiting; right and left seem to both demand they have access to my beliefs, and leverage big corp as a tool to hide their political agenda to that effect as government social programs are accountable to public oversight.

I mean I’m pretty much convinced at this point human languages lack the nuance to make any conclusions of sound logic at this point except in contexts they came up in; Anglo social norms. The correctness of a colloquialism comes along with a command to use it in a certain emotional way.

We’re just engaged in spiraling emotional pettiness; how dare you offend thy sensibilities, good sir!

> Mastodon introduces a feature where you can download and install algorithms, which can be posted by anyone. They are given the raw unsorted list of posts from people you follow and use that produce a coherent feed. You might have to pay for them. They could be free. They could involve elaborate ML, or not. They might sometimes pull in posts from people you don’t follow. They could be open-source, or not.

That, or just use a separate ActivityPub client developed with this feature in mind. After all the point of it all is open and standardized APIs.

It bugs me that people call these "algorithms" when they really are "heuristics".

An algorithm for sorting an array sorts it correctly 100% of the time, at least it is supposed to

http://envisage-project.eu/proving-android-java-and-python-s...

Something that ranks your feed is a heuristic, a rule of thumb that works right frequently.

I have to protest the assumption that if people are spending time on it is what they want.

It can be and often is what people fear-- what they DON'T want. It's a pretty good survival technique, in general, to pay attention to things that threaten you.

Companies monetize this human tendency, same as any other.

Playing on people's wants may waste their time or make them dull, but playing on their fears also harms them by producing constant stress.

I think a happier person produces, does not consume. So a ideal algorithm would be shoving the user towards being productive, creative. To accomplish that, the feed would have to redirected to people who "educate" people to become creators.

Take that nice little austrian post-card creator living in bavaria i follow on twitter. Every now and then a political rant, but nothing.. oh No... oh No No No..

Am I missing something? I have Twitter set to show the timeline of tweets as they happen, which although I accept it requires a deduplication algorithm is broadly non algorithmic.

I realize there is also another 'curated' timeline setting, but you can always switch it off.

There are all kinds of problems with Twitter, but I'm surprised to see so many people saying 'the algorithm' is one of them.

If we had a succinct way to explain or logic to point to, it would help clear the confusion about the massive difference between algos built into Mastodon SW and the ML type algos deployed on platforms.

We end up obfuscating too much of the subject every time we say "algorithm".

We are talking about filtering and sorting. There is no need to be more abstract than that.

It's not what I want - it's what an AI model thinks I want and I feel like that's modelled after the average 'want'.

No, it isn't showing you what you want. It is showing you what it thinks will get you to spend more time on the service, that's all

What you are describing is similar to how my hobby project works: https://linklonk.com

When you upvote content (+1) you make your connection stronger to those who posted that content and also those who also upvoted it. When you downvote (-1), your connection to the poster and other upvoters becomes weaker. The content in your feed is ranked base on how strongly you are connected to those who upvoted the content.

For example, I upvoted the OP post on LinkLonk: https://linklonk.com/item/3406294698351951872. If you upvote that item then you will get connected to me and to two RSS feeds that posted that content: the blog's feed (https://www.tbray.org/ongoing/ongoing.atom) and the the HN's feed of newest items (https://news.ycombinator.com/rss). As a result you will start to see other content from those RSS feeds and from my "main" channel more prominently in your recommendations.

Users can post/submit links under different user channels. So if your hypothetical E2E person were to post E2E stuff under their "E2E" channel while keeping the furry stuff under their "furry" channel then you would only get connected to their E2E channels if you only +1'd content from that channel.

When RSS was hot stuff circa 2005, a friend on IRC and I were working on a bayesian classifier for RSS entries and it worked just like this. +1 trained it as "ham" and -1 trained it as "spam". It failed (among other reasons) because we were both really young and didn't have enough mathematial/computational maturity. With modern ML techniques, this doesn't seem hard to do at all. But you'd need access to the actual data stream which is something that the big social networks don't seem willing to provide.

I think it's actually important for you to see all the sick shit. I think it's worth rubbing your nose in it.

It's amazing how many technologists have a strong preference for truth in so many areas, but choose not understanding things they find disgusting and why they find them disgusting.

> Most Twitter users, as best I can tell, aren't actually complaining about what they see. They are complaining about what others see.

FWIW this doesn't match my experience at all, but I've hardly made a study of it. What I mean by that is that the limited complaints I have heard are mostly about a) bad quality advertisement targeting and b) political targeting, also mostly bad quality.

"quality" here meaning accuracy of the targeting, not a comment on the content.

> I dunno. Mostly I don't want to see ads. I liked Facebook when it just told me what my friends were doing. I don't want a "news feed" from a social network.

But the news media suffer from a huge punditry to hard facts ratio problem.

Today's first-screen news stories on major outlets that didn't begin as a press release or punditry:

* Fox: "Remains of missing toddler Quinton Simon found in landfill, mother charged"

* Washington Post: "Covid deaths skew older, reviving questions about ‘acceptable loss’"

* New York Times: "Antiwar Activists Who Flee Russia Find Detention, Not Freedom, in the U.S." "Exhumed Grave Near Kherson Shows Occupation’s Brutality"

* New York Post: nothing.

* Reuters: "Hawaii's Mauna Loa volcano erupts for first time in nearly 40 years".

* The Times (of London): nothing, because they have such large ads and banners that the content is hidden.

* South China Morning Post: nothing, again mostly because of oversized banners.

* Le Monde: "The unflinching gaze of Ukrainian drones in Bakhmut"

* CNN: multiple stories, because the first screen has many headline links and small banners.

Some of this is optimizing the use of screen space for revenue, not information.

For me, it was never about what I did see, it was always about what I didn't see. Want to show me extra stuff that you think matches? OK, but FFS let me see the stuff that I have purposefully, manually subscribed to/followed.

That's true.

My timeline is pretty chill.

But when I look what's trending in my country, Twitter is a garbage fire.

Hmm, I see a lot of trash that I don't want to see "suggested" by twitter in my feed on the time. The thing is...I don't care that others see it, like it, retweet it. It just, personally, I don't want to see content from accounts I'm not subscribed - I don't care how many of my accounts, that I follow, follow those accounts. Like, 2/3 of my feed is not from accounts I follow: promoted tweets and suggestions.

That's why I quit. I still log in from time to time to keep my handle alive and check out some crazies...

^ and this sentiment is repeated by everyone who thinks they're smarter than everyone else.

So now where are we?

Isn't it odd that we are both always perfect and always getting better with time?

I don't know anyone who consistently looks at their own thoughts and opinions from a self-critical eye with the goal of probing their mental models for flaws in the structure for the express purpose of fixing their logical or factual errors, but surely we do gradually over time.

That is probably a very difficult thing to do all at once, although I suppose it does happen little by little. Surely, no matter how rarely or in what brilliant glimpses we get time chisels away our ignorance so long as our ignorance is more pliable than the steel of the chisel of our self-education.

I doubt many adults feel as though they have suddenly sprung fully formed from the womb of childhood, fearfully and wonderfully made in the image of perfection, and yet most of us feel that as of right now, our small flaws aside, we are as close to perfection as we have ever been or possibly ever will be.

Eh, I think many of them are fully aware that it is the platform that pushes views that agree with them, hence any notion of removing "moderation" seems like helping "the enemy" get a hold.

I don't think that's correct. I think most people are complaining about what they see, and how they feel when using Twitter.

I'm basing my opinion on:

- The experience of seeing people move to Mastodon and comment on how much more pleasant they find it (I do think this might be fleeting - more because they're seeing only early adopters than because there's intrinsic 'bitterness' missing on Mastodon, at least compared to 'old Twitter').

- My own experience of Twitter. It just makes me anxious and angry now. Every time. Yet I keep going back! It's transformed from a pleasant place to converse and browse to somewhere where everyone is angry at everyone else all the time. I'm fairly liberal in my opinions, so usually people I see are angry at either right wing folks or other left wing folks being liberal in slightly different ways - but I do also see the retweets of angry right wing tweets with the "hey, isn't this opinion SO WRONG" opinion. This is with the 'chronological' feed. With the 'algorithmic' feed, it's _even worse_: I now get more angry people I don't know on both sides of the spectrum, and I don't see the non-angry normal conversations or life updates from people I do follow (presumably because they're low-engagement).

- I _do_ think about how this affects others, you're right - and I imagine that those with more right wing opinions are in much the same state as me, but more 'opposite', and _even more angry_ because right wing accounts seem to be more likely to be aggressive than left wing ones. So, yes, I am concerned about what others see - but only in as much as I'm concerned about what everyone sees and how the current political climate is descending into an angry selfish madness where no-one is prepared to actually listen to anyone else. I do think this angry selfish madness is more prevalent on the right, especially in mainstream politics.

Maybe my experience is not the majority one, but I think you should take it into account.

Sounds just like Hacker News.

Read Marcuse's "Repressive Tolerance" to see why this is the case:

https://www.marcuse.org/herbert/publications/1960s/1965-repr...

The tl;dr is that if you actually want to move toward a society of free and equal human beings and eliminate oppression, "free speech for all" is not how you go about it. One side of the political spectrum supports equality and emancipation, the other is opposed to it. Therefore the speech of one side must be tolerated and supported; the speech of the other must be squelched and restricted.

The internet has had its time to experiment with free speech for all, and now we have Nazis. Social media companies saw greater value in muzzling the Nazis than they did in hewing to outdated libertarian internet values.

> Most Twitter users, as best I can tell, aren't actually complaining about what they see. They are complaining about what others see.

I don't see any evidence for that

> In their view, they have the "correct opinions" ... the people that disagree with them are the ones being duped

This is getting very straw-man

> That's why there's a huge censorship movement right now in the US

Oh really, is there?

It makes perfect sense considering the wider public understands "algorithm" as "big,opaque,unknowable way of doing something probably tilted towards interests of the corporation/government", which is how the word has been used in mass media for years.

When people complain about the "the algorithm" they're not talking about bubble sort. It's shorthand for something - "biased feed algorithms" maybe.

You may as well tell them tomatoes are actually a fruit.

"Algorithm" is similar to "cloud" and (much earlier) "online" in that it's acquired a popular/marketing meaning distinct from its original technical meaning. It's an example of an improper noun[1].

[1] https://siderea.dreamwidth.org/1773806.html

I can't recall where I saw it, but a few years ago I saw an interview that involved some techie. The host asked them about Twitter and said, "what if we get rid of the algorithm, and, say, put everything chronoligically." The interviewee responded with "Well if you get rid of the current algorithm and replace it with that, you'll have an algorithm that sorts in chronological order." to which the interviewer's response was, "no, I'm saying we get rid of the algorithm."

I think most people don't regard something they could easily—if tediously—do by hand without ever so much as thinking of a mathematical symbol or operation (I mean, I guess they do a bit of you're sorting by numbers, but "is this number/date higher or lower?" only barely qualifies as math for most people, if at all) as algorithmic, even if it entirely is, by the strict definition.

They aren't complaining about algorithms. They are complaining about an algorithm that determines what people should see for the sole purpose of increasing some engagement metric. This does not necessarily mean giving the person what they want to see.

I mean, it is mostly just people dont know what the word means. Its so vauge you could replace it with "do something" and be correct 99% of the time.

Sorting requires "doing something", doesn't sound too shocking.

There's absolutely nothing hypocritical about e.g. "defending peace". It's realistic. between Karl Popper's well-known writings on the subject, and one accusatory sentence from someone on the internet, I'll take Karl Popper.

You do realize that both "paradox" and "hypocrisy" have the same basic definition, don't you? They both mean "self-contradictory", the biggest difference is in connotation, not meaning.

"...for it may easily turn out that they are not prepared to meet us on the level of rational argument, but begin by denouncing all argument; they may forbid their followers to listen to rational argument..."

That's Popper.

Does that sound like "we can't let people speak"?

Or would that actually be more in the lines of "we should expose their followers to the arguments their leaders don't want them to see"?

I've read him, thank you, quite a few years ago. Wasn't impressed.

Some of people here who take free speech for granted do not know its value. I live in a very authoritarian country, that once had free speech, and I have seen how restrictions on it creep in. And what I see happen in US and Europe follows a far too familiar path.