Facebook is rating the trustworthiness of its users on a scale from zero to one
washingtonpost.com
washingtonpost.com
Well, yes. Have they not learned from any of the preceding work on moderating?
The whole article sounds more like focusing on one implementation details of review queue prioritization (i.e. what is the difference of definition of "fake news" for this user vs Facebook's definition), rather than bigger picture.
Maybe the term isn’t as narrowly defined as your (former?) employer would like.
Edit: upon a bit more reflection, I do take for granted a desire to protect life and liberty (in that order), and anything that appears to threaten that view will draw special scrutiny, and potential abuse of reporting features.
"Fake" means "false", not "stuff I disagree with".
People will report any negative coverage of their favorite party/politician as biased, especially now that the current U.S. president hoists that banner every day.
One avenue would be to put those who report the same things as biased in cohorts, to be shown or not shown similar content.
That just reinforces the current self-silo effect we see, with people only seeing the point of view with which they agree.
In any case, the ratio of bias reports/views for content is likely to be inversely related to its value and general appeal.
That would mean that your number would be on the lower end, but that's not a reflection of your character. More likely it just means that whatever queue is set up for review will wait until more and higher-scoring individuals also flag the article before bumping it up the queue.
Funny how this article is something I would flag if I used your definition of fake news. To me this article is trying to allude to some nefarious Black Mirror nonsense.
Couldn't say it better.
There's still an aspect of this with which I'm unsatisfied, however. What's the character of leadership who would address misuse of the feature by devising weights intended to devalue the opinions of large swaths of their engaged user base? Significant numbers of people are expressing a sentiment through their alleged abuse of the feature, and instead of trying to channel that constructively, the company is devising ways to ignore it.
What's the character of leadership who would unjustly silence group A just because significant numbers of people are expressing sentiment to silence group A? Not sure if not doing what they are doing is actually better.
On another note:
This is not necessarily abuse. Even with no abuse, you can have different definition of fake news, than FB have. In such case, you report, someone at FB reviews, and nothing happens. Having weight on you does not change this. The "leadership" is "ignoring you" in a same way as before. Apart from the fact that it's not "ignoring you". They just have different opinion than you.
Note: I used "you" but I don't think that you necessarily have different definition than FB, if was just more convenient to formulate it this way.
I subscribed to a lot of contradictory crap on facebook after hearing about their data mining (Fox News plus Huffington Post, Obama as well as Trump). Its actually a lot more interesting seeing how both sides try to manipulate the truth.
This happens on HN so it's going to happen in userland for sure!
Some people view a like as a personal read receipt
Some people view a like as a broadcasting to their friends/network that they saw it
Some people view a like as consent to the message
Some people view a like as liking the message
The most extreme of those benign actions are the ones making its way into public policy.
So it isn't surprising to me that a moderator also had no clue what people do.
That said, I think it is great that they are trying to characterize this sort of thing. The second law of institutions is that every process must require more people to execute than can conspire without being detected[1]. But relying so much on humans in one way or another is going to gum it up. My mom for example, a certified human, would always flag anything that offended her belief system as 'fake news.' Fact checking is fine but it is important to understand that conspiracy folks actually believe what they are saying is true, regardless of any facts presented that contradict their world view. I have always attributed that to an internally rooted desire to "need" it to be true, either for their own self image or their own conscience.
[1] No, I just made that up. But the kernel is true, if you have a system that is being gamed you have to build a defense that cannot itself be gamed.
I guess, except without transparency and accountability, this will give more power to those gaming the system
Step 1 - let users report posts that are false
Step 2 - human fact checkers investigate user reports
Step 3 - rely more on users who report posts as false that really are false, and less on users who report posts as false that really are true
Step 4 - repeat step 3 but for the fact checkers themselves
Step 5 - train an ML model to automate more of the fact checking work
> [1] No, I just made that up.
You didn't, actually. This is the Byzantine Generals problems as described by Leslie Lamport back in 1982. It is huge in fault tolerance for distributed systems.
https://en.wikipedia.org/wiki/Byzantine_fault_tolerance#Byza...
Where I was coming from was that there aren't, as far as I know, a codified set of rules for creating a stable institution. I was thinking that someone would say "there is no such thing" and ignore the advice so in my footnote I pointed out the the basic idea was in fact quite sound. I probably should have referenced the paper as well but assumed most people would already know about it.
As my Linear Algebra prof used to say - deep down, everything is just linear algebra...
So, people report posts as false when they do not agree that they are true?
I'm struggling to understand how that is a surprise, or what other possible thing you could expect. That's literally exactly what “false” means.
I disagree that that is obvious, and, moreover, I would disagree with the idea that the author would have any way of telling whether someone flagging merely disagreed with some aspect of the viewpoint/presentation outside of the base facts or disagreed with both the viewpoint and the fact claims, seeing the latter as misrepresented to serve the former. Further, it's virtually impossible to cleanly separate disagreement on viewpoint from disagreement over facts, because the two mutually interact (particularly viewpoints not only shape perception of facts but also perceptions of the; relative importance of facts.)
For instance, there is a story right now circulating in even fairly mainstream, especially conservative, media (notably including Fox News) and being widely shared on social media alleging that a proposed California law would ban restaurants from serving children any drinks except the two choices of milk and water. This story is invariably couches in a lot of viewpoint around crazy nanny state liberals and other conservative stereotypes of California. It is very likely that lots of people would disagree strongly with the viewpoint of the story.
But the actual proposal, as one discovers with a few minutes of research, (1) doesn't limit what restaurants can offer to children (it limits “default drink options” for children's meal combos) and the limitation it proposes isn't to just milk and water (it includes milk, water, flavored water, and nondairy milk alternatives). It is, in other words, factually false in several dimensions.
Now, if one agrees with the viewpoint, one may be more likely to dismiss those factual errors, even if you are aware of them, as minor and tangential. If one disagrees with the viewpoint, one is likely more inclined to see them as significant and central and indicative of a deliberate distortion to push an agenda.
But seriously, imagine an article posted on Facebook stating that Planned Parenthood sells body parts of aborted fetuses (or something similarly likely to elicit reactions). Obviously, that's either factual or it’s not. Yet I'm certain that if Facebook users had the ability to vote on whether the story is true of false, they would vote based on personal beliefs, not on what they would find by fact-checking the story.
Likewise, for anything else that in any way touched on an inflammatory issue. Do you actually think otherwise?
Sure, but unless people actually work at PP or carry out their own individual on-site investigation, their knowledge of the fact is going to be based on what third-party sources they trust (including those acting as fact checkers), which is tightly correlated with viewpoint.
> Facebook users had the ability to vote on whether the story is true of false, they would vote based on personal beliefs, not on what they would find by fact-checking the story.
They'd probably vote based on their beliefs about the facts, which would often include checking with sources they trust on facts. Of course, those beliefs—and where fact checking is involved, the set of trusted sources—will be strongly correlated with their beliefs about abortion.
The ivory tower ideal of a clean separation between how people address fact claims and the value system in which they give significance to facts, where you could then isolate which process was involved in a decision excluding the other, is, while convenient, not at all a reflection of reality.
But that's true of nearly everything.
On anything controversial, users will generally vote based on what they like to believe, and such a system of ranking news stories or posters of news stories would never be reliable.
Yes, that's my whole point: the claim made that users flag content bases on ideology rather than factual disagreement is essentially impossible to be justified, since perception of facts and ideological viewpoints are deeply intertwined.
This is the crux of the whole issue. Everyone's talking about "facts" but should be talking about "proof". We would need a vouching system where only people who possess evidence or can otherwise demonstrate proof will even have their vote counted. Who watches the watchman yada yada. The whole thing is an impossible feat. The best we as content consumers can do is only trust that which is created by trusted hardware (i.e., is actual live action video and not CGI) rather than that which is created by journalists or reporters of any reputation whatsoever.
This seems like a "the road to hell is paved with good intentions" kind of thing. On one hand, I understand not wanting to forward every single story when a troll is mass-reporting, on the other hand though, this seems like a very 1984 approach that like all algorithms is horrendously naive.
Any closed group is likely to have a skewed sense of reality. On one hand, third party fact checkers are hopefully true to their title, but on the other hand, they could just be adding their own bias to the decisions with no oversight in a way that affects users going forward.
Add in no ability to appeal and no visibility into the system, and you have a closed system ripe for abuse.
Good job Facebook, you have learned nothing.
This is the part that stood out to me.
It's allows Facebook to tell lawmakers and the public "Hey, look! We're taking things seriously!"
Until something goes wrong. Then they can lay the blame on someone else and issue another non-apology apology saying, "We can do better."
Is it the third-party aspect that you take issue with, or the fact-checking? Or the combination of the two?
I'd say contrary - that would be so wrong for FB to verify facts in-house. This way they are more transparent
While this is literally true, it's not clear to me that it's a relevant metaphor for online communities.
Not attempting to the arbiter of truth and falsehood would be one.
Or reporting on what users think of a story's truthfulness: "49% rate this story as true, 21% rate it as untrue".
Or, if they want to be honest, acknowledging that it is about policing taboo thoughts. "Smith, 4980203 Winston. This article is crimethink! Warning: If you read too many articles like it your Citizen Credit Score will be downgraded. You have been warned!". OK, maybe they won't go for that :-)
Think of something like Pizzagate - the fact that the restaurant in question didn't have a basement belied the assertions that bad things were happening in said basement. Any article or comment that passed on that provable lie should be flagged as a lie.
Sure, there are grey areas beyond that. Those are opinions, and should be separable from facts.
Sure, but for most contentious political issues, this is not the case.
Truth and falsehood are absolute for pretty much all fact claims.
Whether they can easily by determined is another story.
> Think of something like Pizzagate - the fact that the restaurant in question didn't have a basement belied the assertions that bad things were happening in said basement.
Sure, but suppose (they weren't, by for sake of illustrating the problem) bad things were happening in, say, a attached a controlled by the owners of the restaurant, but somehiw as the story got out and before it was fixed in text, it was distorted so that it referred to a non-existent basement instead. Would actively suppressing the story because of the impossible location described really be in the public interest?
Sure, in strict terms any single factual error in the most involved story makes the story false. But, most accounts will accumulate some false elements in a complex story: there has to be consideration of importance and materiality.
Valve also does something similar with their anti-cheat system called overwatch. A machine learning system selects which replays are most likely to result in a conviction, then those replays are shown to human peer-reviewers who must come to a consensus on whether the replay shows cheating or not. The pre-selection stage greatly increases the efficiency of the system.
There's absolutely nothing nefarious about remembering previous interactions you had with someone. In fact it's impossible not to do so.
The problem with China is how it tears down the natural separation between different spheres of your life, how all-encompassing their panopticum can become because it has access to any data source it desires, and that the issues it engages with all seem to be of the this-doesn't-concern-you-and-nobody-is-actually-harmed variety, such as peoples' porn consumption or eating habits.
Seems like infinite regress and turtles all the way down to me. Who fact checks the fact checkers?
They seem to demand a sort of proof from first principle for journalism, where every reader can (and must!) verify an article without any heuristics such as a medium's track record ("Argument from Authority!", they will complain, "FALLACY!").
Yet such a system is simply impossible. In practice, because nobody would ever have the time & energy to verify any meaningful fraction of the information they consume. But even in theory, there just aren't any methods to distribute "proof": you could be an eyewitness, which doesn't scale. You could inspect original documents, which also doesn't scale, and where almost nobody has the expertise to actually verify them. Oh, and then there are distributed ledgers, which could help around the margins but nobody knows how (Bitcoin, incidentally, is the outgrow of exactly the same mechanism, only for the financial system: the breakdown of trust in institutions such as the law or the FED, and the attempt to replace it with technology).
One of many incarnations of this dynamic the widespread criticism of the use of anonymous sources. And while the practice does have its problems, it's almost universally misunderstood by those criticising it today.
When journalists use anonymous sources, they substitute their own reputation, and that of their publication, for the source's. Which is why, when you go back and read articles from a year or so ago, you'll find that each and every single anonymously reported fact has actually been proven correct, or remains unknown. The system works (worked), because everyone in it is playing the long game.
But that would require trust. And people today aren't just distrustful. They completely deny the utility of the concept. They try to evaluate all information with complete disregard for the source (and fail). They assume equivalence between anonymous internet commentators and institutions such as universities, newspapers, or established politicians, without even noticing how this conflicts with their criticism of anonymity as described above.
This mindset is related to a perversion of the scientific method, one that denies the usefulness of all emotion, of the social sciences, and of all human relationships. It's perverse because even its proponents entirely fail to live by it: if you're more willing to let your spouse borrow you car than just any stranger, you're using the very same mechanisms (familiarity, the game theory of long-term relationships, reputation) that you otherwise reject.
So that's a long-winded way to say: these fact-checkers are supposedly different than whoever posted the material in question. You, and everyone else, can observe them over time, and they will have a reputation to defend. They will presumably also be recruited from the set of people and institutions that are established as trustworthy, at least among the minority that still believes in such a thing might actually exist.
Third party fact checkers seems like the best possible thing that could be in that box. It's not perfect, but there has never been a perfect system to determine truth so waiting until there is one is counterproductive. You do the best with the tools you have available.
"new",
> But how these new credibility systems work is highly opaque, and the companies are wary of discussing them, in part because doing so might invite further gaming — a predicament that the firms increasingly find themselves in as they weigh calls for more transparency around their decision-making.
Fractal stupidity - this is the term for this situation
You build better systems, which dont ACTUALLY capture the core bad behavior.
Users figure out the edges of said system, and start finding newer ways to slip through the gaps.
Now you need a new system.
Repeat till the history of the system looks like a pullitzer prize winning Kafka novel.
It seems to me that if the claims and stories in question could be re-posed (e.g. as falsifiable, logically coherent statements or arguments) the worst of "fake news" could be weeded out.
The rest (and biases, etc.) would likely have to be left as an exercise for the (critical) reader.
What they're trying to accomplish, to my mind at least, would require the invention of general AI. Automatic content moderation sounds very close to a Turing test to me.
Considering that the state very likely has wholesale access to all the data, I'm not sure that's making actually much of a difference.
The difference vanishes when your state is run by corporations.
What impact on physical and mental health does it have to sit around checking different screens for InstaFaceSnap or RedditTwitTube updates?
Bill (not FB user)
Sue (FB user)
Jill (FB user)
Sue: Bill is just so unreliable. He says things and doesnt do them.
Jill: Yah he did the same to me!
(Notices Bill's ghost profile's trustworthiness score go down, without any interaction from Bill at all. What exactly do we do about that? We in the US have no rights like this. The corps can do whatever they want, and maybe just maybe, a decent law will be passed to protect us. Or likely not.)
Edit: Useds was name given by Stallman from https://stallman.org/facebook.html .
Also, -1? Where am I being wrong? Are ghost profiles being dismissed, or trustworthiness? Or is it the fact I'm putting these things together?
Against what? A company rating their trust in our ability to use their resources? This happens on a daily basis with almost any company; it's just less formalized.
1. How do we know FB will assign the scores fairly and objectively? We don't, of course, but is there any legal requirement or protection for users?
2. How is it like the Chinese government's social credit score? Right now Facebook says it's used for filtering user reports, but if FB becomes confident in it, what else might it be used for? Will they sell the score to advertisers and vendors so that they can filter customers? To credit score companies? To insurance companies and potential employers? Will it be used to 'nudge' mass behavior in certain directions? Is there anything preventing FB from doing those things?
3. As usual it's the powerful (FB, potentially their advertisers and other partners) who have information about ordinary people (FB users). Isn't it more important that ordinary people have trust information on the powerful? What is Mark Zuckerberg's trust score? Google's? Elon Musk's? Various politicians?
4. According to the article, FB isn't releasing the algorithm because it would enable users to game it (in fairness, I don't remember if FB said that or someone else). But those with resources and motivation, such as the high-end propagandists, will learn how to game it anyway, which can have at least two consequences: 1) Propaganda may become harder to identify, and 2) power will be even more concentrated in the hands of the few - FB, their partners, and those who can game the system.
5. As seems the norm for tech, the users have very little agency (other than to try to please the system); they are objects of technology and of those who control it. It bears repeating that we've come a long way from the world of free-as-in-speech systems and the idea that the goal is end-user control. In another context, an article posted to HN paraphrased Alan Kay as asking, 'Should the computer program the kid or should the kid program the computer?' [0] I think our society has answered that question for the present.
The article goes on to suggest this is just one of a range of heuristics use to determine trustworthiness. I can imagine all kinds of profile interactions that would be be good indicators of this.
a normalized statistical value is generated for a particular user's willingness to share fake news
after reading the article:
a normalized statistical value is generated for a particular user's willingness to report news they disagree with.
yeah, nothing to see here..... :-/
I'll write a note to the journalist, hopefully they can elide or contextualize this kind of thing in further articles.
This certainly seems to fit.
Careful, you could get taken down for fake news doing this.
https://www.washingtonpost.com/archive/opinions/2003/02/06/i...
But I'm sure this time is going to be different
I appreciate fake news being cleaned in my timeline, if the cost is no more than some occasional legit news being pruned once in awhile.
B. A private company scoring its users to combat spam, as all social networks do.
Also fantastic as a hiring screen.
This will become like gaming Google rank.
https://web.archive.org/web/20180821143120/https://www.washi...
It's designed to reduce THEIR risk, which means by default that it will be stacked against YOU, and they won't care about those getting ground up in the wheels of the machine.
<GPS Location: Mistress' apartment>
trustworthiness_score -= 0.1
In this context, a "perfect" algorithm would be one where someone trying to "game" the system would essentially be forced to post only true content to achieve a near perfect score?
I hope one factor in their scoring involves number of cases where someone reports factually accurate content as false on numerous occasions or does the same for opinion-based content. Of course, that then requires a means for validating the content as opinion or fact-checking, but they're already heading down that path.
I think we can all agree that the secret algorithm is going to be in the social signalling - job status, who you know, where you live.
Make everyone's rating public and you're there, or worse make it available to whoever pays.
What you will be left with is what many people originally flocked to the internet to avoid: the 'MSM echo chamber' of anodyne center-left opinion against which no dissent will be tolerated.
I think this idea should be taken further so that you can’t post as much (or even at all) unless you are actively participating in meta-moderating, i.e. helping the community.
Like what?