The article repeatedly makes effort to assure the reader that the left/right dichotomy wasn't hardcoded but instead "emergently" appeared, then never goes into any detail about how that happened exactly...
It's weird to me that the left right divide happened to appear, rather than a faith based/non faith based, or an authoritarian/liberal, etc dichotomy. Is there no public data on this we can use to prove it's true, just trusting the org at it's word?
I think this is easy to explain by considering that the meaning of left/right constantly changes over time. For example, how would you characterize somebody who is pro free speech, anti war, anti hegemony, pro equality of opportunity, in support of a color-blind society, and so on? 20 years ago that would have been an absolutely textbook liberal, today that is no longer the case.
So when we say left or right we're not really referring to any sort of static values, or even fundamental values. It all just keeps shifting and often in self-contradictory ways. So the terms just become a proxy for the ever-shifting divides in society. This makes it essentially a tautology that any fair sampling of society will end up divided on those terms.
The changing of left/right over time is largely because much of American society views politics through the lens of Democrat/Republican. On a broader level, economic left/right and authoritarian/libertarian haven't moved, but both parties kinda rotate around them over time. At the moment I'd say both democrats are moving slightly further left, and both parties are moving more authoritarian.
To me, it appears clear that the divide is not really left/right; my understanding is that the algorithm basically reduces a high-dimensional "alignment" of a user/note to a 1-dimensional "polarity", and this simply happens to somewhat coincide with the US party split.
I think this is almost to be expected if you have a 2 party political system, because each party is trying to align with as many voters as possible; so you can fully expect that the republican/democrat split is a "good" cut through alignment space.
Basically, the algorithm explicitly looks for the "best" polarity axis (i.e. the one that contains the most information about whether a user is going to find something agreeable or not), while a 2 party political system implicitly alignes across such an axis (as an emergent property of party-politics-shifting).
> this simply happens to somewhat coincide with the US party split
That's not my reading. The negative polarity is clearly the US Democrats, but as Vitalik notes, the opposite polarity isn't anything obviously specific to the Republicans. His posts for negative were simply the first three posts and they're all of a very consistent political ideology and region. But he had to cherry-pick posts for positive polarity to try and sustain the claim that this is the opposite of the negative polarity, as Brazilian politics and Tesla people isn't something anyone would have picked if asked to guess what the posts would be about ahead of time.
Then there's the clear content differences between the positive vs negative polarity posts. The negative polarity posts are all highly subjective opinions, usually about the tone of what some famous people said. Their references are simply left-leaning US media articles which are themselves opinion. They could be easily disputed and frequently are. The positive polarity posts are cherry picked for the purpose of making a point, yet are mostly specific factual claims with hard data and references, except for the second, which is apparently disputing the claim that child trafficking doesn't even exist? I don't see any way to dispute any of the claims in the positive polarity notes. You'd have to fall back on "well that may be true but in wider context..." type responses.
So it's not really clear exactly what this algorithm is picking up on. The differences may just reflect the inherent nature of politics in which left and right are often presented as polar opposites, but which in reality is more like the left being relatively homogenous and consistent at any given point in time, whereas the "right" is more like a coalition of people who aren't on the left than a specific set of policy or cultural concerns.
> which is apparently disputing the claim that child trafficking doesn't even exist?
The troublesome part of that note is that "the movie accurately depicts" the issue. The following note has "these books are obscene." The first is a little bit different in that it is straightforward and factual, but with the (missing) context of the tweet it was probably thought to be an irrelevant insult i.e. I bet the tweet wasn't about the proportion of black children in single-parent households.
It's obviously a party split in a way, but to my eyes it isn't about assigning people to a party - it's finding people who hate current parties rather than people who love them. That is to say: polarized individuals. I think it's an accident of history that people who despise Republicans currently have their opinions fairly well-represented by the Democratic Party, but on the other side, people who hate Democrats aren't very well represented by the Republican Party. The Democrat-hating base is unruly, and I think it contains far more people who also dislike or are neutral toward Republican politics and politicians.
That could be. It's not clear from the note what movie is being referred to. The lack of context makes it harder to interpret.
The last post reflects ideological values, but appears to be a factually unambiguous claim about US law. The images show cartoons of men sucking each other's dicks. In law "obscene" is used to mean something like "overtly sexual", which at least the first book clearly meets.
No, obscene doesn't mean that. Obscene material is illegal material, not adults-only material.
edit: so saying that the books depicted explicit sexual acts would be indisputable, but the determination that the books were obscene would happen in court. And if drawings of two people sucking each other's dicks were judged obscene, a lot of things would instantly become illegal in that jurisdiction.
That's a circular definition (the law is defining what's illegal, so the category it defines cannot itself be defined as that which is illegal). Though the actual legal definition in the USA isn't much better.
I think the claim in question is about the first amendment issue. If the material was judged obscene then it could still be allowed, or disallowed, or disallowed in some contexts (i.e. schools, what's in dispute here) but such laws wouldn't get tripped up by the bill of rights.
It's not a circular definition. "Obscene" is a term used for content that has been deemed illegal. It is not a general term for sexual content. It is not a term for content that is illegal sometimes but not other times. It is not a euphemism for material that is restricted to adult consumption, or for explicit pornography.
more charitably, but still repetitious: when one says that something is obscene, that's saying that it should be illegal in any context; that it has no value. A drawing of two people sucking each others dicks has surely met that bar in the past - information about birth control has met that bar in the past. But I do not think the suggestion was that drawings of gay men having sex should be illegal, what was being suggested was that it is not appropriate for children. That's not a question of obscenity.
Obscene in the context of the first amendment protections is defined by the Miller test regardless of how any laws define obscenity. The third prong alone is a high bar to clear in this case and its what makes the claim far from factually unambiguous imo.
> Whether a reasonable person finds that the matter, taken as a whole, lacks serious literary, artistic, political, or scientific value.
They are not categories, they are contexts, and educational texts can have value within any or all of those contexts. In fact being educational or informative could inherently be considered to be of literary value.
Many people rightfully complain about the lack of specificity and clarity in the third prong, but that is even more evidence for why it is a bad community note because there is obvious room for interpretation and the note leaves none. Thus, even if you agree with its conclusion, it lacks sufficient context which is one of the voting criteria.
I'd be very careful though about drawing conclusions about what exactly polarity means here because 3 examples are not nearly enough-- especially for conclusions like which polarity is more "fact based".
My viewpoint is that the polarity in the algorithm should be akin to a principal component in alignment space (because of how the algorithm works), and you would expect the political parties to be split along a very similar axis in a 2 party system (an emergent property-- if they were not, one party could gain voters or improve cohesiveness of its voter pool by shifting its alignment).
I think on re-reading, I'm not strongly disagreeing with you indeed. I think it's just that the article says the positive polarity posts don't map cleanly to Republicans, whereas the negative polarity posts do map cleanly to the Democrats, so how much does it really coincide?
Not a direct answer, but years ago a friend of mine tried a similar experiment by taking a bunch of polarizing questions and then doing principal component analysis on the results. He came up with two axes (the maths would spit out a number of axes, but the rest were low-salience). The top one was very similar to "left/right".
Not sure about proof, but the left/right dichotomy also “emerged” in the real world in most countries. It’s not the only axis of variation that we see, but it’s a dominant one. Since Twitter is part of the real world, the users may have arrived with a pre-existing propensity to group themselves into left/right tribes, and then the mechanics of social media bubble formation accelerated the polarization process.
Once a "tribe" gets big enough it starts to force others to pick a side. Inevitably the system reduces down to two.
In politics voting methods can defuse this somewhat by allowing the smaller tribes stick to their own values and not loose out too heavily in the power grab. First-past-the-post amplifies the larger tribes, proportional representation (or similar systems) can partially reduce their effect.
No, not really. There are different types of centrists, religious vs liberterian right, and even some socialist religionists. The world is a tat bit more complex than the loud minorities want to make you believe.
It's also one that is insufficient, you just have to look how often fascist-like parties are classified "right wing" (while they are centrists), or the likes of Gandhi and Stalin get put in the same "left" box, while they are polar opposites.
Unless I have misunderstood it, this algorithm sounds particularly bad, because it's either only able to see the world through a single dimension, or worse, it's even forcing it into the same dimension (criticizing from competent mathematicians/staticians would be welcome).
But in the end, I guess it doesn't change much : it's worse than useless, just like Twitter and the people that use it, for the last decade and counting.
Feel free to run the matrix factorization code on the data, and then try to interpret the resulting latent dimension it finds for yourself! And also, you can read the code to verify that it really is running a matrix factorization rather than hardcoding a particular left/right split.
One could argue that a particular question will always have a left/right-ish divide of some kind, even if people may disagree which is which. In many cases on social media discourse, it comes down to pro-establishment vs. anti-establishment (which previously would usually have been seen as right vs. left but now is generally seen as the reverse).
Anecdotally, most community notes I've seen appear to take the form of "anti-establishment claim -> community note refuting the anti-establishment claim". But what's nice is that when it's the opposite, they still seem to work well.
Both faith based/non-faith based and authoritarian/liberal views will often, if not exclusively, be interpreted as falling along the left/right axis by many people. It's simply a matter of interpretation and bias.
What exactly would it look like to be an authoritarian liberal?
Americans have a particularly confusing form of "liberal" today, but at least my understanding of the two terms has them pretty specifically at odds with one another. I'm not sure how an authoritarian would have centralized so much power when the focus is on individual rights and freedoms.
It's just a language issue. "Liberal" in the US doesn't mean the same thing it means in other English-speaking countries. Americans coined the term "neoliberal" to talk about what other people just call "liberal." In the US, "liberal" just means that you vote for Democrats, and that you're concerned. It's not a distinct ideological stance.
That really makes the distinction of am "authoritarian liberal" pointless at that point. One is a ideology and the other is a political party, the combination of the two is completely irrelevant as any party could align with one specific ideology.
There's the classic right wing economic liberal who is pro free market, but you also have the bottom wing civic freedoms anarchist liberal (which is indeed anti-authoritarian/fascist top wing). Both anarchist and fascist are centrists in the usual left vs right sense.
P.S. But you can add more dimensions, it's just that you cannot go below two without having nonsense as a result.
It's weird to me that the left right divide happened to appear, rather than a faith based/non faith based, or an authoritarian/liberal, etc dichotomy. Is there no public data on this we can use to prove it's true, just trusting the org at it's word?