I don’t see the claims from Twitter matching the headline The Guardian is running. Twitter’s internal research is saying that certain tweets are organically more popular and engaging and more likely to be shown, not that the algorithm has an ideological bias. If you search for the word “bias” in Twitter’s post you’ll not find support for The Guardian’s editorialization. Here’s an important excerpt, which notes that more analysis is needed to understand whether the amplification is unnatural relative to user interactions:
> Algorithmic amplification is problematic if there is preferential treatment as a function of how the algorithm is constructed versus the interactions people have with it. Further root cause analysis is required in order to determine what, if any, changes are required to reduce adverse impacts by our Home timeline algorithm.
This study also leaves out a very important consideration, which is the impact of Twitter’s moderation along political lines. I suspect that’s where the true bias lies and given Twitter’s content policies reflect progressive ideology, it is very likely that bias leans left.
>Twitter’s internal research is saying that certain tweets are organically more popular and engaging and more likely to be shown, not that the algorithm has an ideological bias
No one is claiming that there's a piece of code that detects a Twitter user being a right-wing politician and then showing it to more users.
Whether the bias is deliberate or not, it's still a bias. Right wing gets their message amplified more than the left.
A platform promoting one side of the political spectrum over another is not bias? There's no deliberation needed to meet the definition of bias. Every person have unconscious biases, and any structural system, including algorithms that promote content on social media can have unconscious biases as well. Simply stating that it's not bias does not cut it for me.
OK, I admit it's not unequivocally not bias. It could be bias. I would just say there could be other explanations, such as a different ratio between non-politician activists to politicians on either side of the isle and how each slice of the pie tends to monopolize limited attention bandwidth.
If three political tweets by Lebron James are algorithmically boosted, and three political tweets by Ted Cruz are boosted, I wouldn't call that bias necessarily even though the metric being studied did.
Perhaps I didn’t explain my opinion clearly, but I think we might have different definitions for what constitutes bias. You’re suggesting that a difference in outcomes means something is wrong or unfair with the algorithm. But what if it is that right wing users simply engage more, or that the right wing news outlets put out better content, or other such factor? I wouldn’t view the increased algorithmic amplification as problematic if there were such root causes behind it. Put another way, I am taking an equality lens (equality of opportunity) rather than an equity lens (equality of outcomes). If we take an equity lens, then you can start questioning why any number of things don’t get the same amplification - it doesn’t have to stop at a binary of “left versus right”.
How would you enforce, e.g. hate speech, moderation if there are both left and right wing definitions for it?
Or Twitter's content policies could follow the law and reflect the political leaning of the governement, but then how do you enforce moderation on a free/liberal social network which by definition should be free from governement censorship?
> which notes that more analysis is needed to understand whether the amplification is unnatural relative to user interactions
Whether it’s natural or not is irrelevant IMHO.
If a post fanning the flames of a religious war or ethnic cleansing is “organically popular” should the algorithm amplified it?
It should also be noted that just because something is popular doesn’t mean it’s a factually correct piece of information or a sensible course of action.
Sometimes lies and bad ideas are more (emotionally) appealing and eye catching but that doesn’t mean one should propagate them.
Reddit and HN are guilty of this too given how their vote based system works.
> How does an algorithm determine what is a lie or bad idea? Please share.
I don't know. Perhaps there will never be an algorithm that can do it even "approximately" never mind accurately - OK maybe "never" is too strong a word.
It doesn't change the fact that the current way we are doing things are causing problems.
In the past we rely on fallible editors to filter the news. It wasn't perfect. However I think they did a better job than the algorithms of Twitter and Facebook - most editors had some common sense with regards to the effects of what they allow to go to print; the more responsible ones will do due diligence and fact check stuff before printing it (as imperfect as their ability to do so is, it's better than the no fact checking at all done by Twitter/Facebook before they amplify it to millions of people).
Actually the guardian headline is perfectly accurate. It completely is accurate to say the algorithm has a bias, even if that bias isn't "if (right-wing) then amplify()"
> If you search for the word “bias” in Twitter’s post you’ll not find support for The Guardian’s editorialization
But if you search for "favour" you'll find "[our findings revealed that] algorithmic amplification favours right-leaning news source". In the abstract. Which means the exact same thing.
> Here’s an important excerpt, which notes that more analysis is needed to understand whether the amplification is unnatural relative to user interactions
The guardian article includes your quote so I don't think you can accuse it of omitting important context.
> Twitter’s internal research is saying that certain tweets are organically more popular and engaging and more likely to be shown
I'm not sure where you're reading this from. It is pretty explicit that further research is required in to the cause of the discrepancy.
yes, the twitter research showed that right-wing tweets were organically more popular than left wing, and due to that organic popularity were more likely to be shown, but that's true of all popular things on twitter, it's got nothing to do with the content. the headline and article are intentionally misleading.
This is also the second time this article was shared instead of just posting the twitter blog post or the original study, considering we like to use original source material it's odd that this article keeps getting shared.
The Guardian is one of the few good news sources left in the UK, after the Telegraph and others went full tabloid. Op-eds are often written by insane SJWs, but if you skip them the articles are often very good.
Also - who cares about Twitter's bias and who gives a damn whether it's an AI algotrithm or something simple like if/else.
It's (or should be) assumed they make decisions for their own benefit. Fix: easy - don't use Twitter.
Why would anyone want to read about that and especially at the statist Guardian were journos never see a problem for which there should be no government (or globalist)-enforced solution?
Externalities are a problem still. When you're a truck driver and your employer gets 100 angry emails and phone calls because someone tweeted a picture of you doing an "OK sign" to a passing motorist, then you lose your job, not being on Twitter wasn't an option.
If you know its low effort, I would encourage you to not post it. Or at least make an effort to add some substance to it, perhaps by enumerating the flaws in it?
For breaking their ToS literally hundreds of times? If anything it showed incredible bias they kept him on the platform just because of his job ( and the fear of retaliation). If he were a random person, he would have been banned much earlier.
This might be overly reductive, but I read this as "right" tweets and figures generate more collective user engagement than "left" tweets and figures, though the underlying algorithm(s) is blind to the substance - political or otherwise - of the message.
The proposed change would then introduce conscious political bias to ensure an even representation of the spectrum... but I wonder how we define even? Perhaps we can base it on user's constituencies?
I think this is the type of question we're left with, usually, after analysing the behaviour of any ranking/amplification algorithm.
The reductive conclusion is either "right wing tweets are more engaging" or "the algorithm amplifies right wing tweets." A reduction implies that either the algorithm or users are neutral. Without algorithmic bias, users will amplify both equally. Or conversely, results reflect user bias/preference/behaviour because the software is unbiased.
Like a lot of reductions, these are true but incomplete. We know damn well that software design will affect the popularity of various post types. Most forums have active mitigation strategies to avoid certain types of content from becoming overabundant. HN has the "middlebrow dismissal" rule, for example. We also have no reason to believe that users, on average, are "unbiased" or that any specific outcome represents and unbiased reflection collective user views.
There's no doubt that the character of HN's homepage, or reddit, FB & Twitter reflection of both moderation/software decisions and "organic" user preferences. There's no hard line between them.
OTOH, this doesn't mean "nothing to see here."
This is also not unique to social media. You can clearly see that specific traditional media stations/publications favour specific views. More importantly, they amplify certain stories, cover certain candidates, and have many other preferences and biases that have big political impacts.
Here, instead of "left vs right," they distinguish between "horse race coverage" and "substance." 91.3% of CNN's coverage was found to be "horse race coverage?" Is that a bias? Pattern? Editorial Decision." Reflection of user interest? It sure does affect politics though. Frames the whole debate.
It would not work as both "sides" claim the other to be over-represented. In truth they want to silence each other if we are honest. I don't like Twitter making content decision but I guess that ship has sailed long ago for any large social media platform.
There are too many left-wing activists since being right-leaning is stigmatized (case in point, big tech) while being left-leaning is seen as boosting one's social/career path (the new komsomol).
Moreover, the left discourse is not engaging to anybody outside of this circle. No memes.
Of course they have trouble getting legitimate attention. Way too many activists reposting boring non-news, too little audience.
Now, I admit that most right-wing activists are crackpots, but they seem to understand pain points of society and act on these.
Yes its strange how right-wing bias is "problematic".
I'm not aware of big tech ever finding left-wing bias a problem or even being prepared to admit it exists.
IMO, everyone comes into this discussion with some sort of anchoring, and build up from that.
IMO, it's way to cumbersome to think in terms of right, left and bias. What's the observable difference between a world where "right wing bias" exists and one where "right wing posts are just more compelling."
Bias is a strong word, but if we're using it, then I think it's near universal. Media is never neutral. Social software is never neutral. Software design always affects content on a social media. The business model of the media (eg cable news, subscriptions newspaper, free dailies) very obviously dictates content.
Liberalism has a history of falling into a pattern. Theory about fair/good systems, and the outcome they'll produce. If they don't produce the desired outcome, we fall into a mess. The pattern applies to economics, social stuff, domestic & international politics.
IE, start with a belief that liberal democracy (formulations vary) is the only system that's fair, functional and successful. Post-soviet russia implements free market capitalism, emphasising contracts, property rights and such. It quickly develops an oligarchy. That's attributed to "not doing capitalism right" or the moral failings of Russian culture.
Myanmar's despotic press monopoly opens up to FB. This a more free and open media diet. Rumours about minorities quickly spread, evolve into violence and genocide. Again, we have a hard time believing that freedom is a culprit. It's unfree media that causes genocide. Free media is the antidote. So, the only explanation is FB "bias" or some other corruption.
We start with a notion of fair systems and what their outcome will be. When reality disagree, we tend to double down. There are examples in every decade for centuries.
Shortly after Laissez Faire economic theories were adopted in the UK, a famine occured in Ireland. The government response was doubling down. A famine could only mean that markets were not free enough. Population numbers have still not recovered 180 years later.
I skimmed through the study [1], and it seems that the methodology is as follows:
They define a metric called "amplification" ratio for a set of tweets. Roughly, for a specific set of the tweets, it is the ratio of their "reach" in the sample of users with the chronological timeline (control, 1% of global users), to the sample of users with the ML timeline (treatment, 4% of global users). For a sample of users, reach of a set of tweets is defined as the share of the sample which encounter at least one of the tweets in the set. (the amplification metric is actually shifted so that 0% means a ratio of 1, i.e. equal reach)
Then, they took a sample of right-wing and left-wing politions and media and calculated these amplification ratios (for individual accounts, and for the left-team and right-team grouped together, etc.). Generally, this "amplification" metric was larger for right-wing accounts (or groups of accounts).
I think the use of that metric for measuring bias is misleading though, in the sense that they do not account for the fact that Twitter users are mostly left-wing [2], and this does significantly effect the metric they have chosen.
Assume that I'm left-winger who does not follow any right-wing politions. Then probably any sensible algorithm which includes tweets in my timeline from accounts that I do not follow will increase the right-wing "amplification" metric, as it is enough for it to show me just one single tweet from a right-winger. If my understanding is correct, their measure for amplification is way too sensitive. (and it is worse when applied to understand the reach of a larger group of the accounts, as an encounter with a single tweet from any of the members of the group is counted as reach)
44 comments
[ 3.3 ms ] story [ 99.1 ms ] thread> Algorithmic amplification is problematic if there is preferential treatment as a function of how the algorithm is constructed versus the interactions people have with it. Further root cause analysis is required in order to determine what, if any, changes are required to reduce adverse impacts by our Home timeline algorithm.
This study also leaves out a very important consideration, which is the impact of Twitter’s moderation along political lines. I suspect that’s where the true bias lies and given Twitter’s content policies reflect progressive ideology, it is very likely that bias leans left.
Twitter’s original blog post: https://blog.twitter.com/en_us/topics/company/2021/rml-polit...
No one is claiming that there's a piece of code that detects a Twitter user being a right-wing politician and then showing it to more users.
Whether the bias is deliberate or not, it's still a bias. Right wing gets their message amplified more than the left.
If three political tweets by Lebron James are algorithmically boosted, and three political tweets by Ted Cruz are boosted, I wouldn't call that bias necessarily even though the metric being studied did.
Or Twitter's content policies could follow the law and reflect the political leaning of the governement, but then how do you enforce moderation on a free/liberal social network which by definition should be free from governement censorship?
Whether it’s natural or not is irrelevant IMHO.
If a post fanning the flames of a religious war or ethnic cleansing is “organically popular” should the algorithm amplified it?
It should also be noted that just because something is popular doesn’t mean it’s a factually correct piece of information or a sensible course of action.
Sometimes lies and bad ideas are more (emotionally) appealing and eye catching but that doesn’t mean one should propagate them.
Reddit and HN are guilty of this too given how their vote based system works.
How does an algorithm determine what is a lie or bad idea? Please share.
I don't know. Perhaps there will never be an algorithm that can do it even "approximately" never mind accurately - OK maybe "never" is too strong a word.
It doesn't change the fact that the current way we are doing things are causing problems.
In the past we rely on fallible editors to filter the news. It wasn't perfect. However I think they did a better job than the algorithms of Twitter and Facebook - most editors had some common sense with regards to the effects of what they allow to go to print; the more responsible ones will do due diligence and fact check stuff before printing it (as imperfect as their ability to do so is, it's better than the no fact checking at all done by Twitter/Facebook before they amplify it to millions of people).
> If you search for the word “bias” in Twitter’s post you’ll not find support for The Guardian’s editorialization
But if you search for "favour" you'll find "[our findings revealed that] algorithmic amplification favours right-leaning news source". In the abstract. Which means the exact same thing.
> Here’s an important excerpt, which notes that more analysis is needed to understand whether the amplification is unnatural relative to user interactions
The guardian article includes your quote so I don't think you can accuse it of omitting important context.
> Twitter’s internal research is saying that certain tweets are organically more popular and engaging and more likely to be shown
I'm not sure where you're reading this from. It is pretty explicit that further research is required in to the cause of the discrepancy.
https://cdn.cms-twdigitalassets.com/content/dam/blog-twitter...
This is also the second time this article was shared instead of just posting the twitter blog post or the original study, considering we like to use original source material it's odd that this article keeps getting shared.
I'm also curious about what the parent's issues are with it.
[0] https://www.medialens.org/2020/the-people-have-spoken-bastar...
Also - who cares about Twitter's bias and who gives a damn whether it's an AI algotrithm or something simple like if/else.
It's (or should be) assumed they make decisions for their own benefit. Fix: easy - don't use Twitter.
Why would anyone want to read about that and especially at the statist Guardian were journos never see a problem for which there should be no government (or globalist)-enforced solution?
Didn't they totally ban Trump?
The proposed change would then introduce conscious political bias to ensure an even representation of the spectrum... but I wonder how we define even? Perhaps we can base it on user's constituencies?
The reductive conclusion is either "right wing tweets are more engaging" or "the algorithm amplifies right wing tweets." A reduction implies that either the algorithm or users are neutral. Without algorithmic bias, users will amplify both equally. Or conversely, results reflect user bias/preference/behaviour because the software is unbiased.
Like a lot of reductions, these are true but incomplete. We know damn well that software design will affect the popularity of various post types. Most forums have active mitigation strategies to avoid certain types of content from becoming overabundant. HN has the "middlebrow dismissal" rule, for example. We also have no reason to believe that users, on average, are "unbiased" or that any specific outcome represents and unbiased reflection collective user views.
There's no doubt that the character of HN's homepage, or reddit, FB & Twitter reflection of both moderation/software decisions and "organic" user preferences. There's no hard line between them.
OTOH, this doesn't mean "nothing to see here."
This is also not unique to social media. You can clearly see that specific traditional media stations/publications favour specific views. More importantly, they amplify certain stories, cover certain candidates, and have many other preferences and biases that have big political impacts.
Take this example on a recent US bill's media coverage: https://thecolumn.substack.com/p/on-reconciliation-bill-cnn-...
Here, instead of "left vs right," they distinguish between "horse race coverage" and "substance." 91.3% of CNN's coverage was found to be "horse race coverage?" Is that a bias? Pattern? Editorial Decision." Reflection of user interest? It sure does affect politics though. Frames the whole debate.
Moreover, the left discourse is not engaging to anybody outside of this circle. No memes.
Of course they have trouble getting legitimate attention. Way too many activists reposting boring non-news, too little audience.
Now, I admit that most right-wing activists are crackpots, but they seem to understand pain points of society and act on these.
Both have some political points but their head maialse absolutely dwarf these.
IMO, it's way to cumbersome to think in terms of right, left and bias. What's the observable difference between a world where "right wing bias" exists and one where "right wing posts are just more compelling."
Bias is a strong word, but if we're using it, then I think it's near universal. Media is never neutral. Social software is never neutral. Software design always affects content on a social media. The business model of the media (eg cable news, subscriptions newspaper, free dailies) very obviously dictates content.
Liberalism has a history of falling into a pattern. Theory about fair/good systems, and the outcome they'll produce. If they don't produce the desired outcome, we fall into a mess. The pattern applies to economics, social stuff, domestic & international politics.
IE, start with a belief that liberal democracy (formulations vary) is the only system that's fair, functional and successful. Post-soviet russia implements free market capitalism, emphasising contracts, property rights and such. It quickly develops an oligarchy. That's attributed to "not doing capitalism right" or the moral failings of Russian culture.
Myanmar's despotic press monopoly opens up to FB. This a more free and open media diet. Rumours about minorities quickly spread, evolve into violence and genocide. Again, we have a hard time believing that freedom is a culprit. It's unfree media that causes genocide. Free media is the antidote. So, the only explanation is FB "bias" or some other corruption.
We start with a notion of fair systems and what their outcome will be. When reality disagree, we tend to double down. There are examples in every decade for centuries.
Shortly after Laissez Faire economic theories were adopted in the UK, a famine occured in Ireland. The government response was doubling down. A famine could only mean that markets were not free enough. Population numbers have still not recovered 180 years later.
[1] https://en.m.wikipedia.org/wiki/Emergence
They define a metric called "amplification" ratio for a set of tweets. Roughly, for a specific set of the tweets, it is the ratio of their "reach" in the sample of users with the chronological timeline (control, 1% of global users), to the sample of users with the ML timeline (treatment, 4% of global users). For a sample of users, reach of a set of tweets is defined as the share of the sample which encounter at least one of the tweets in the set. (the amplification metric is actually shifted so that 0% means a ratio of 1, i.e. equal reach)
Then, they took a sample of right-wing and left-wing politions and media and calculated these amplification ratios (for individual accounts, and for the left-team and right-team grouped together, etc.). Generally, this "amplification" metric was larger for right-wing accounts (or groups of accounts).
I think the use of that metric for measuring bias is misleading though, in the sense that they do not account for the fact that Twitter users are mostly left-wing [2], and this does significantly effect the metric they have chosen.
Assume that I'm left-winger who does not follow any right-wing politions. Then probably any sensible algorithm which includes tweets in my timeline from accounts that I do not follow will increase the right-wing "amplification" metric, as it is enough for it to show me just one single tweet from a right-winger. If my understanding is correct, their measure for amplification is way too sensitive. (and it is worse when applied to understand the reach of a larger group of the accounts, as an encounter with a single tweet from any of the members of the group is counted as reach)
[1]: https://cdn.cms-twdigitalassets.com/content/dam/blog-twitter...
[2]: There are many studies on this, e.g. this one by Pew Research Center: https://archive.md/iEJaq