I don't know how the author came to the conclusion that they reached. As a person whose livelihood depends on the pound being strong against the yen, I'm not a fan of Brexit by any stretch of the imagination, but I don't understand what the difference is between pro and anti Brexit videos in this context.
IIUC, the author is saying that divisive (i.e. controversial?) videos lead to longer watch times. Longer watch times lead to more ad breaks. More ad breaks leads to Youtube making more money. Videos with longer watch times are recommended over videos with shorter watch times (because Youtube makes more money that way). Thus, divisive topics are recommended over non-divisive topics.
But pro-Brexit and anti-Brexit are two sides of the same coin. Given that Brexit won we could even conclude that the anti-Brexit stance was more divisive (controversial?). So I'm not sure what the author is saying.
Again, to avoid a flame war, I'm not trying to make a political point. I'm just trying to follow the author's logic. What makes pro-Brexit videos more likely to be selected that couldn't also be utilised in an anti-Brexit stance?
Because obviously all moral, right-thinking people know that Remain is the right thing to do. The fact that Remain didn't win just proves that some people were tricked by divisive videos in favour of Brexit.
The videos in favour of Remain weren't divisive, because being in favour of Remain is the moral, right-thinking stance.
Anti-Brexit is the status quo. It's by definition not controversial since they can't do anything other than advocate for how things currently are.
Pro-Brexit is by definition controversial since they were advocating for a complete upheaval in the relationship with the EU constructed over decades. And it was always going to be tinged with xenophobia and nationalism since that was the entire premise for the Brexit vote to begin with. And it doesn't get any more divisive than xenophobia.
It's not status quo in an active vote situation. Your not showing up to vote did not cast a vote in favor of the status quo. Divisiveness is bidirectional by its nature.
The rest of your post is political fervor on your view of the vote and not related to whether YouTube's algorithms promoted divisiveness in all forms instead of just the one you don't like.
As the turnout implies, the anti-brexit camp had it harder to mobilize, since the status que seemed quite safe and unlikely to be challenged by many, especially, being already in a pro EU-region. So a lot of people did not come out to vote in favor since they seemed their vote not neccessary. Asymmetric mobilisation and de-mobilization is a thing.
The election had higher voter turnout than any national election in the UK in decades, and most of all polls showed the election was going to be very close. Blaming voter complacency does not seem logical.
It's quite reasonable to suppose that there are many reasons why one person would or would not mobilize to vote, and, correct me if I'm wrong, it's common knowledge that the "swing voter" effect is not due to undecided people becoming decided, rather that people are marginally motivated or unmotivated to vote.
So, in the spirit of democracy, where the desires of all those who /can/ vote (not just those who /do/ vote) are respected, it is probably interesting to examine why voters do and do not find the motivation to vote. Hypothetically, if we had mind-reading devices, and motivation was not a prerequisite for voting, which side of the vote would have benefited?
Asymmetrical motivation is the hypothesis that in referenda where Action/No-Action are the choices, there will be more voter motivation for Action rather than No-Action. This very reasonably applies to Brexit.
A swing vote just refers to a voter whose decision cannot be accurately predicted. Not voting is as much a part of democracy as voting. Even in extremely formal and small settings ranging from national congresses to the EU, abstaining from casting a vote one way or the other is very much a viable decision in and of itself.
I would argue that abstaining from voting in a formal setting is significantly different than not "voting" in modern elections. There's significant historical evidence of voter suppression, including suffragist movements. Voter motivation is not something that I think we should brush away lightly, because it has an impact on democracy just like voter manipulation and voter suppression.
Not to put too fine a point on it, if the author's intent is to say that their algorithm intentionally promotes nationalism and xenophobia, I think they missed the headline ;-)
I think it's perfectly fair to say that videos from both sides evoked similar passion. We can argue whether it's "good" passion or "bad" passion, but I can't see what difference it makes. Why would I watch a pro-Brexit video for a longer time than an anti-Brexit video, based on topic alone (i.e. assuming that both videos were equally effective in addressing their audience)?
I may be misreading, but I think he's talking about pro-Brexit videos being heavily promoted relative to all other videos, as opposed to relative to anti-Brexit videos. Which still means they were heavily promoted.
> But pro-Brexit and anti-Brexit are two sides of the same coin.
Not necessarily.
For example, recommending videos that are somehow categorized "pro-evolutionary theory" would almost certainly be less divisive than recommending Creationist videos that attempt to "debunk" evolutionary theory. The pro-evolutionary theory videos would typically be substantive scientific talks or lectures that require patience and critical thinking in order to understand. The Creationist videos, on the other hand, typically use lots of underhanded tactics. They will attempt to mislead the viewer and typically throw out so many different arguments from so many disparate fields that it's impossible for any one person to debunk them all out of hand.
If you recommend the "pro-evolution" videos you probably end up with a lot of bored viewers who don't even make it through a single video. If you recommend the "anti-evolution" videos then you'll get a biologist debunking point 1 of 100 and a troglodyte responding "what about point 2-100?"
I have no clue whether the pro/anti Brexit videos fit that bill. But it's certainly a possibility.
Pro and anti brexit are not questions of what materially is, but a question of what ought, but no, perhaps not even what ought, but what choice to make, possibly one which doesn't have one option as morally obligatory over the other.
So, no, it isn't really like one group claiming a falsehood to be material fact while another claims the negation of that falsehood to be a material fact. (Material fact as opposed to moral fact)
If only that was the case! One of the conversations I had before the referendum went like this:
Them: “We should leave because then they’ll give us a great deal.”
Me: “But they can’t give you a better deal.”
Them: “That just proves we should leave!”
This person has a Cambridge degree, yet this topic allowed them to mix up hopes and preferences with objective economic benefits and end up with cult-like thinking where the conclusion was kept despite the negation of the argument just used to reach it.
From time to time, I ponder their mistake and ask if I have made a similar one of my own.
I have found that the smarter the person, the easier it is to fall into the trap of asserting something confidently without actually thinking through the details. Our brains are really good at abstracting out detail and so there can be a whole tree of things to consider that our brain simplifies into a ready made answer. When you are really bright, you are used to being right (or at least being able to argue the point so well that other people concede the argument). With that confidence, it's easy to overlook pretty massive gaps.
I once worked with a brilliant physicist. One day he went off on a holiday and when he came back I asked him how it went. "Terrible", he said. It turned out that his plan was to drive from Winnipeg (which is right smack in the middle of Canada) to Vancouver (on the west coast) in a day. Then he was going to drive down the coast to Los Angeles the next day, do some sight seeing and find a hotel. Then we was going to drive to Las Vegas the next day, and gamble for the rest of the day. Finally he was going to drive back to Winnipeg the next day. He made it from Winnipeg to Moose Jaw before he gave up.
"When I looked at the map of North America in my map book, it was the same size as the map of Germany, so I thought they were the same size in real life", was the excuse.
I also have scale problems whenever I visit California (I’m from the UK, which is about half the land area of that particular state). I’m glad Google Maps gives me a sense of journey time, or I might make similar mistakes.
Regarding the person themselves, when it came to subjective questions — philosophical sensation of self, where I rather than they wanted to live, if cyclists (they didn’t cycle) preferred cars to overtake quickly or slowly — this person tended to have the “used to being right” hubris that you described.
> The pro-evolutionary theory videos would typically be substantive scientific talks or lectures that require patience and critical thinking in order to understand.
There's a LOT of very low-brow, fast-food content on the "left", too. Videos that just makes fun of how stupid anti-vaxxers are, generate outrage about global warming while being very light on scientific data and hard of apocaliptic imaging, generate fear and anxiety about the rise of fascism and nazism, repeating the idea that Trump is literally Hitler and Theresa May is his Mussolini...
I'm not trying to make any point about whether those things are true or false, by the way. Just about the level of propaganda and emotional manipulation on both sides.
You're at least somewhat correct, at least judging by the headlines on realclearpolitics, for example, or by browsing politically opposed subreddits simultaneously. But I'm really astounded that you've said that anti-vaxxers are specifically right-wing. I thought it was a bit more like the flat-earth phenomenon where it's unclear who is being satirical, who supports it simply because of some disposition to conspiracy, and who actually believes that they've found evidence more compelling than the near entirety of the foundation of our scientific system - the same scientific system that has managed to create the device of incomprehensible complexity on which you're reading this comment.
The point being made is that, in order to promote the divisiveness that yields more ad revenue (through increased "watch time"), the YouTube algorithms disproportionately promote positions that are morally wrong, factually incorrect, or both.
Brexit may not be the best example, because (perhaps, arguably) there could be legitimate arguments for the pro-Brexit stance.
Vaccination might be a better illustration.
Promulgating the view that "vaccines are a hoax, and vaccinating your child will cause autism, or let the government achieve mind control" is both incorrect (scientifically, factually) and wrong (morally). Those convinced by this view put their child and other children around their child at grave risk.
A system that promotes this view over the established, normative, scientifically validated consensus view that we should vaccinate our children, to the point that it changes opinions and behavior, is harmful to public discourse (and harmful to children).
This is true whether automated/algorithmic, or intentionally devised and operated by people with malevolent intent.
At scale, these phonomena become harmful to democracy itself. Democracies are diminished when their subjects believe false information, especially objectively harmful false information whose belief causes disease, stuctural collapse of infrastructure, etc.
Since we are now observing this happening to various extents in various areas of society, we should use our power as democratic citizens to compel operators such as YouTube to provide real data on what their algorithms are selecting, so that we can understand the effect on our society.
Ya, I love getting flat earth suggestions because I watch the PBS Spacetime channel. YT promotes tons of absolute crap. Of course there is tons of crap, since making crap is low effort. Seems to go hand in hand with click bait titles, which are decisive in and of themselves.
I would think that such a contentious issue would lead to people doing research which would lead to views which would lead to the AI thinking "hmm, this seems to be a popular topic, maybe other people might like to see it too?"
Assuming, of course, one believes in the myth of the informed voting public.
> Assuming, of course, one believes in the myth of the informed voting public.
They have to at least believe in the desire to be informed. I agree with your point and optimizing for eyeballs means you're gonna optimize for what people want to see. But as we've seen lately, if that machine driven optimization goes against the wishes of the squeakier of the wheels, it is considered intentional favoritism or at best negligence. Internet advertising is the popular scapegoat these days when cognitive dissonance amongst the loud is all that remains.
This is all good stuff. When we ignore the wishes of the majority of UK voters this will make it much easier for us to look ourselves in the mirror and pretend that we still believe in democracy.
What's really chilling is the atmosphere in tech in which certain legitimate democratic outcomes are wrong and that tech has a responsibility to "do better" next time and avoid more "mistakes".
Perhaps your algorithm isn't buggy just because the democratic process didn't produce your desired outcome.
I think you should read the twitter thread linked. Nobody is calling this a bug: rather, YouTube put their fingers on the scale because it was better for their bottom line.
What he's saying is that this is a retrospective that's looking to blame outside reasons for the outcome of a vote, when ultimately the reason a vote went one way may simply be genuinely because most voters felt that way. In hindsight you can blame a million externalities for any 'undesirable' result in any election, because there are a practically infinite number of factors, foreign and domestic, involved in the elections of any major nation.
For instance imagine leave lost and chose to play the same game. In mid April 2016 Obama blew to the UK to try to do a mixture of evangelizing against Brexit and implicitly threatening the leave position as well by suggesting that Britan would be 'at the back of the queue' in any future US trade deals if Brexit passed. And one can only imagine the thousands, if not millions, of pro-remain posts and topics shared on social media by people outside of Britain. Of course these ended up being completely pointless and Obama's visit actually ended up strengthening the leave position, but cast in a different light these sort of externalities could be used to try to argue for undue and undemocratic influences that led to the result of the election. But it's revisionism.
I hear an assertion that Brexit was promoted because it is divisive. Without knowing what was done and what he means by what he claims, it is hard to come to a non-knee-jerk reaction. If you think Brexit is a disaster in principle, this is catnip. If you don't, not so much.
It's chilling but it will be a little while before we end up there.
The engineers (and friends) who built these systems are wildly successful, and as often happens in these cases, they now feel guilty about that success, especially when they're surrounded by so much human suffering (in a place like SF), which pushes them in a progressive direction.
But now they've discovered that the systems they built had unintended consequences and may have even led to the very success of the political ideologies they despise.
The cognitive dissonance working at a place like Twitter must be palpable. Trump probably drives more revenue to the company than any other person, and yet probably 90+% of the employees who work there despise the man and everything he stands for.
So you can see why so many people are on-board with trying to do something about it. But the same poorly-thought-out, gung-ho attitude which led to algorithms no one understands anymore, will backfire when they try to "correct" them.
Social engineering is a dangerous game. This country is a time bomb just waiting to go off and when Silicon Valley finally pulls the plug on Trump there's no telling where it will end.
(though I do think things have calmed down a little in the last couple years... we'll see where they end up after the midterms...)
What's the lesson here? I see 2 suggestions: (1) algorithm engineers try to make inputs as fair as possible, 'promoting' every political initiative perfectly fairly or (2) algorithm engineers train bots to produce output (e.g., election results) they prefer since they're considered morally superior by those involved.
While (2) is definitely worse, I don't think (1) is that much better. It's unreasonable to expect an algorithm to be totally impartial and trying will only lead to more unintended consequences.
Perhaps this is the true evil of advertising-centric revenue models: they put corporate monopolies in charge of swaying the biggest decisions we as citizens can make, the ones that run our societies and our lives.
Privacy, transparency, mental health, etc. for the end user are all important, but in day-to-day life, these considerations pale in comparison to the sheer power these corporations have acquired in steering world affairs.
If they just chose more responsible business models, this wouldn't be happening.
#1 and #2 amount to the same thing, since the success metric for #1, by necessity, has to be the extent to which it produces the political environment favored by its creators. Fairness is giving weigh according to legitimacy, and we as humans always perceive our own side as being more legitimate.
> If they just chose more responsible business models, this wouldn't be happening.
No effort on social change that requires that corporations voluntary forego profit (or that requires that individuals forego status) will ever work, nor should it. Everyone talks about what behavior should be, but nobody wants to talk about the sort of incentive structure that would produce that behavior.
We're in the first round of digital media becoming a pervasive part of everyone's lives.
People are just now realizing how harmful these services are & just how extensive their powers are. At the same time, there are a bunch of new services popping up that offer to spread power more equally with structures that are more distributed by design.
I think it will take time (like, maybe a decade or two)...but today's dominant services will die out through a combination of terrible events and user-hostile moves, and new services will emerge that build on the lessons learned. It might seem unthinkable right now, but I think the right dynamics are in place to make it happen in the near-future.
Fundamentally, there is a lot of content out there. So much so, that it is necessary to have some algorithm assist in curation. Any algorithm you choose will favor certain types of content over others (unless you literally choose randomly, in which case you aren't really solving the original problem). The question is what should that algorithm be optimizing for. We have seen what problems optimizing for ad revenue causes; but I see no reason to think that optimizing for some other metric wouldn't produce a comparable class of problems.
Ultimatly, the problem of algorithmic curation is a new problem, and no one has a good answer for it. Whatever answer we pick will have significant influence on the public discourse, because so much of said discourse is filtered through the algorithm.
There's another way of looking at things. The success or failure of a content sharing service is reflective of it's ability to give people what they want to see. If you don't like the content you're getting from a video service, you're going to go swap to another that does. In other words, growth implies content selection is working.
But of course this leads to what I think is the real problem: network effects. The reason people use YouTube is because people use YouTube. What I said above is true in a world with sufficient competition, but YouTube has no meaningful competition. They constantly hurt their content creators but the creators stay there nonetheless because YouTube has the users. And YouTube has the users because YouTube has the content creators.
---
By focusing on the content curation as the issue, I think we take a monopoly as granted. But there are ways you could help enable competition such as by removing barriers to content delivery. Imagine a public content delivery network where whitelisted providers (just to avoid it being used for unlawful content) can stream from. Pair this with rules against exclusivity for free content delivery services, and some way for content creators who upload the content to the service to cleanly opt in to allow their content to be used by various services for some standard contract (%revenue, one-off fee, etc), and away we go.
Very off the cuff idea, but the idea is just to help foster more competition.
I don't work at YouTube, but it is highly unlikely that they have a Brexit department or a divisiveness department. What they have is machine learning.
The goal of their Algorithm is to maximize for watch time. Any behavior that leads to longer watch time will be promoted.
If a user watches 2 videos, leaves the platform for a while then comes back later to watch a third, the system will have learned what video to suggest to another person who watched only the first two. If the system can't determine what to show next, it will choose the video that has a record of increasing watch time.
That seems consistent with what the author is saying.
Working inside of YouTube, he was in a position to see that the algorithms heavily promoted Brexit. However, for people not directly involved in YouTube's operations, this is not necessarily apparent — and is essentially impossible to check.
His point is that the ultimate outcomes of this algorithmic selection has potentially massive societal repercussions, and therefore the citizenry deserves to know what is being selected for promotion.
While it wouldn't solve the problem in and of itself, I think one mechanism that could help mitigate this phenomenon would be to compel companies like YouTube and Facebook to (somehow) disclose what their algorithms select for promotion (that is, for suggestion or automatic presentation).
When the citizens don't and can't know what YouTube is playing for people, they can't really object.
On the other hand, if the data was there and enough citizens were interested for somebody to analyze the data, then we could know: Oh, during the 6 weeks leading up to the Brexit vote, among the top 1000 videos X% strongly advocated for, Y% strongly advocated against, Z% didn't fit into strong advocacy. Oh, and, this false disproven pro-Brexit piece from Boris Johnson was played 18,960,000 times to an estimated 14,000,000 unique viewers.
That would provide another vector of incentive/consequence that content promoters — algorithmic or otherwise — would have to factor into their decisions.
I think this would also help with the other, extremely related problem of "algorithmic child abuse" described by the viral article "Something is wrong on the internet" from last year[1].
As we shift from a mass broadcast media landscape (TV, radio, print), increasing algorithmic content selection has huge implications, for society, democracy, children, etc.
Because each person (or child) potentially receives their own custom-tailored feed of content (articles, videos) and ads, it is extremely difficult to, first of all, _know what is actually happening_. (That is, what is being shown to the public, or in the second case, what is being shown to individual kids.)
This makes it effectively impossible to hold these companies accountable for what they are choosing to show people.
I am sure that these companies would fight tooth and nail to prevent such legislation from being enacted, and I am not confident that the collapsing democratic institutions in my own country (USA) are even capable of doing so. It's a pretty hard thing to legislate.
Even if the rule was something like each user's account had to make accessible, somewhere, the complete history of content and advertising selected for display for that user. Then, scientists and journalists could at least use random sampling of people, get their data, and do analysis that way. Although probably, with micro-targeting of individuals, we would also need the global totals for each content item or ad to also be available.
I think algorithmic selection has the potential to so terrible because it is happening in the dark. There are very few consequences for bad behavior when it is known only to the "victims".
Especially if they don't think they are victims; e.g., my young kids when they're watching the grotesque algorithmically generated videos of their favorite video game characters that YouTube suggests for them — my personal introduction to this entire topic. If you haven't seen it before, I recommend the below-linked article to anybody with children, or who just wants to think more about the implications of algorithmic/ML content selection.
51 comments
[ 2.9 ms ] story [ 111 ms ] threadIIUC, the author is saying that divisive (i.e. controversial?) videos lead to longer watch times. Longer watch times lead to more ad breaks. More ad breaks leads to Youtube making more money. Videos with longer watch times are recommended over videos with shorter watch times (because Youtube makes more money that way). Thus, divisive topics are recommended over non-divisive topics.
But pro-Brexit and anti-Brexit are two sides of the same coin. Given that Brexit won we could even conclude that the anti-Brexit stance was more divisive (controversial?). So I'm not sure what the author is saying.
Again, to avoid a flame war, I'm not trying to make a political point. I'm just trying to follow the author's logic. What makes pro-Brexit videos more likely to be selected that couldn't also be utilised in an anti-Brexit stance?
The videos in favour of Remain weren't divisive, because being in favour of Remain is the moral, right-thinking stance.
Sort your life out, Jim. Pathetic little incel.
Anti-Brexit is the status quo. It's by definition not controversial since they can't do anything other than advocate for how things currently are.
Pro-Brexit is by definition controversial since they were advocating for a complete upheaval in the relationship with the EU constructed over decades. And it was always going to be tinged with xenophobia and nationalism since that was the entire premise for the Brexit vote to begin with. And it doesn't get any more divisive than xenophobia.
The rest of your post is political fervor on your view of the vote and not related to whether YouTube's algorithms promoted divisiveness in all forms instead of just the one you don't like.
https://en.wikipedia.org/wiki/United_Kingdom_European_Union_...
So, in the spirit of democracy, where the desires of all those who /can/ vote (not just those who /do/ vote) are respected, it is probably interesting to examine why voters do and do not find the motivation to vote. Hypothetically, if we had mind-reading devices, and motivation was not a prerequisite for voting, which side of the vote would have benefited?
Asymmetrical motivation is the hypothesis that in referenda where Action/No-Action are the choices, there will be more voter motivation for Action rather than No-Action. This very reasonably applies to Brexit.
I think it's perfectly fair to say that videos from both sides evoked similar passion. We can argue whether it's "good" passion or "bad" passion, but I can't see what difference it makes. Why would I watch a pro-Brexit video for a longer time than an anti-Brexit video, based on topic alone (i.e. assuming that both videos were equally effective in addressing their audience)?
Not necessarily.
For example, recommending videos that are somehow categorized "pro-evolutionary theory" would almost certainly be less divisive than recommending Creationist videos that attempt to "debunk" evolutionary theory. The pro-evolutionary theory videos would typically be substantive scientific talks or lectures that require patience and critical thinking in order to understand. The Creationist videos, on the other hand, typically use lots of underhanded tactics. They will attempt to mislead the viewer and typically throw out so many different arguments from so many disparate fields that it's impossible for any one person to debunk them all out of hand.
If you recommend the "pro-evolution" videos you probably end up with a lot of bored viewers who don't even make it through a single video. If you recommend the "anti-evolution" videos then you'll get a biologist debunking point 1 of 100 and a troglodyte responding "what about point 2-100?"
I have no clue whether the pro/anti Brexit videos fit that bill. But it's certainly a possibility.
So, no, it isn't really like one group claiming a falsehood to be material fact while another claims the negation of that falsehood to be a material fact. (Material fact as opposed to moral fact)
Them: “We should leave because then they’ll give us a great deal.”
Me: “But they can’t give you a better deal.”
Them: “That just proves we should leave!”
This person has a Cambridge degree, yet this topic allowed them to mix up hopes and preferences with objective economic benefits and end up with cult-like thinking where the conclusion was kept despite the negation of the argument just used to reach it.
From time to time, I ponder their mistake and ask if I have made a similar one of my own.
I once worked with a brilliant physicist. One day he went off on a holiday and when he came back I asked him how it went. "Terrible", he said. It turned out that his plan was to drive from Winnipeg (which is right smack in the middle of Canada) to Vancouver (on the west coast) in a day. Then he was going to drive down the coast to Los Angeles the next day, do some sight seeing and find a hotel. Then we was going to drive to Las Vegas the next day, and gamble for the rest of the day. Finally he was going to drive back to Winnipeg the next day. He made it from Winnipeg to Moose Jaw before he gave up.
"When I looked at the map of North America in my map book, it was the same size as the map of Germany, so I thought they were the same size in real life", was the excuse.
Regarding the person themselves, when it came to subjective questions — philosophical sensation of self, where I rather than they wanted to live, if cyclists (they didn’t cycle) preferred cars to overtake quickly or slowly — this person tended to have the “used to being right” hubris that you described.
There's a LOT of very low-brow, fast-food content on the "left", too. Videos that just makes fun of how stupid anti-vaxxers are, generate outrage about global warming while being very light on scientific data and hard of apocaliptic imaging, generate fear and anxiety about the rise of fascism and nazism, repeating the idea that Trump is literally Hitler and Theresa May is his Mussolini...
I'm not trying to make any point about whether those things are true or false, by the way. Just about the level of propaganda and emotional manipulation on both sides.
Brexit may not be the best example, because (perhaps, arguably) there could be legitimate arguments for the pro-Brexit stance.
Vaccination might be a better illustration.
Promulgating the view that "vaccines are a hoax, and vaccinating your child will cause autism, or let the government achieve mind control" is both incorrect (scientifically, factually) and wrong (morally). Those convinced by this view put their child and other children around their child at grave risk.
A system that promotes this view over the established, normative, scientifically validated consensus view that we should vaccinate our children, to the point that it changes opinions and behavior, is harmful to public discourse (and harmful to children).
This is true whether automated/algorithmic, or intentionally devised and operated by people with malevolent intent.
At scale, these phonomena become harmful to democracy itself. Democracies are diminished when their subjects believe false information, especially objectively harmful false information whose belief causes disease, stuctural collapse of infrastructure, etc.
Since we are now observing this happening to various extents in various areas of society, we should use our power as democratic citizens to compel operators such as YouTube to provide real data on what their algorithms are selecting, so that we can understand the effect on our society.
At least, that's what I got out of his tweet.
Assuming, of course, one believes in the myth of the informed voting public.
They have to at least believe in the desire to be informed. I agree with your point and optimizing for eyeballs means you're gonna optimize for what people want to see. But as we've seen lately, if that machine driven optimization goes against the wishes of the squeakier of the wheels, it is considered intentional favoritism or at best negligence. Internet advertising is the popular scapegoat these days when cognitive dissonance amongst the loud is all that remains.
However I will keep it here as a prediction of future events.
Perhaps your algorithm isn't buggy just because the democratic process didn't produce your desired outcome.
For instance imagine leave lost and chose to play the same game. In mid April 2016 Obama blew to the UK to try to do a mixture of evangelizing against Brexit and implicitly threatening the leave position as well by suggesting that Britan would be 'at the back of the queue' in any future US trade deals if Brexit passed. And one can only imagine the thousands, if not millions, of pro-remain posts and topics shared on social media by people outside of Britain. Of course these ended up being completely pointless and Obama's visit actually ended up strengthening the leave position, but cast in a different light these sort of externalities could be used to try to argue for undue and undemocratic influences that led to the result of the election. But it's revisionism.
The engineers (and friends) who built these systems are wildly successful, and as often happens in these cases, they now feel guilty about that success, especially when they're surrounded by so much human suffering (in a place like SF), which pushes them in a progressive direction.
But now they've discovered that the systems they built had unintended consequences and may have even led to the very success of the political ideologies they despise.
The cognitive dissonance working at a place like Twitter must be palpable. Trump probably drives more revenue to the company than any other person, and yet probably 90+% of the employees who work there despise the man and everything he stands for.
So you can see why so many people are on-board with trying to do something about it. But the same poorly-thought-out, gung-ho attitude which led to algorithms no one understands anymore, will backfire when they try to "correct" them.
Social engineering is a dangerous game. This country is a time bomb just waiting to go off and when Silicon Valley finally pulls the plug on Trump there's no telling where it will end.
(though I do think things have calmed down a little in the last couple years... we'll see where they end up after the midterms...)
While (2) is definitely worse, I don't think (1) is that much better. It's unreasonable to expect an algorithm to be totally impartial and trying will only lead to more unintended consequences.
Perhaps this is the true evil of advertising-centric revenue models: they put corporate monopolies in charge of swaying the biggest decisions we as citizens can make, the ones that run our societies and our lives.
Privacy, transparency, mental health, etc. for the end user are all important, but in day-to-day life, these considerations pale in comparison to the sheer power these corporations have acquired in steering world affairs.
If they just chose more responsible business models, this wouldn't be happening.
> If they just chose more responsible business models, this wouldn't be happening.
No effort on social change that requires that corporations voluntary forego profit (or that requires that individuals forego status) will ever work, nor should it. Everyone talks about what behavior should be, but nobody wants to talk about the sort of incentive structure that would produce that behavior.
People are just now realizing how harmful these services are & just how extensive their powers are. At the same time, there are a bunch of new services popping up that offer to spread power more equally with structures that are more distributed by design.
I think it will take time (like, maybe a decade or two)...but today's dominant services will die out through a combination of terrible events and user-hostile moves, and new services will emerge that build on the lessons learned. It might seem unthinkable right now, but I think the right dynamics are in place to make it happen in the near-future.
Fundamentally, there is a lot of content out there. So much so, that it is necessary to have some algorithm assist in curation. Any algorithm you choose will favor certain types of content over others (unless you literally choose randomly, in which case you aren't really solving the original problem). The question is what should that algorithm be optimizing for. We have seen what problems optimizing for ad revenue causes; but I see no reason to think that optimizing for some other metric wouldn't produce a comparable class of problems.
Ultimatly, the problem of algorithmic curation is a new problem, and no one has a good answer for it. Whatever answer we pick will have significant influence on the public discourse, because so much of said discourse is filtered through the algorithm.
But of course this leads to what I think is the real problem: network effects. The reason people use YouTube is because people use YouTube. What I said above is true in a world with sufficient competition, but YouTube has no meaningful competition. They constantly hurt their content creators but the creators stay there nonetheless because YouTube has the users. And YouTube has the users because YouTube has the content creators.
---
By focusing on the content curation as the issue, I think we take a monopoly as granted. But there are ways you could help enable competition such as by removing barriers to content delivery. Imagine a public content delivery network where whitelisted providers (just to avoid it being used for unlawful content) can stream from. Pair this with rules against exclusivity for free content delivery services, and some way for content creators who upload the content to the service to cleanly opt in to allow their content to be used by various services for some standard contract (%revenue, one-off fee, etc), and away we go.
Very off the cuff idea, but the idea is just to help foster more competition.
#gamergate
#comicsgate
#elsagate
#maga
Algorithms that dumbly promote stuff that gets people "engaged" have a huge back door called "pissing people off".
The goal of their Algorithm is to maximize for watch time. Any behavior that leads to longer watch time will be promoted.
If a user watches 2 videos, leaves the platform for a while then comes back later to watch a third, the system will have learned what video to suggest to another person who watched only the first two. If the system can't determine what to show next, it will choose the video that has a record of increasing watch time.
Working inside of YouTube, he was in a position to see that the algorithms heavily promoted Brexit. However, for people not directly involved in YouTube's operations, this is not necessarily apparent — and is essentially impossible to check.
His point is that the ultimate outcomes of this algorithmic selection has potentially massive societal repercussions, and therefore the citizenry deserves to know what is being selected for promotion.
I'd bet something along the lines of "absolute_value(upvotes divided by total votes - 50%)" is a feature somewhere. That's how I interpreted OP.
When the citizens don't and can't know what YouTube is playing for people, they can't really object.
On the other hand, if the data was there and enough citizens were interested for somebody to analyze the data, then we could know: Oh, during the 6 weeks leading up to the Brexit vote, among the top 1000 videos X% strongly advocated for, Y% strongly advocated against, Z% didn't fit into strong advocacy. Oh, and, this false disproven pro-Brexit piece from Boris Johnson was played 18,960,000 times to an estimated 14,000,000 unique viewers.
That would provide another vector of incentive/consequence that content promoters — algorithmic or otherwise — would have to factor into their decisions.
I think this would also help with the other, extremely related problem of "algorithmic child abuse" described by the viral article "Something is wrong on the internet" from last year[1].
As we shift from a mass broadcast media landscape (TV, radio, print), increasing algorithmic content selection has huge implications, for society, democracy, children, etc.
Because each person (or child) potentially receives their own custom-tailored feed of content (articles, videos) and ads, it is extremely difficult to, first of all, _know what is actually happening_. (That is, what is being shown to the public, or in the second case, what is being shown to individual kids.)
This makes it effectively impossible to hold these companies accountable for what they are choosing to show people.
I am sure that these companies would fight tooth and nail to prevent such legislation from being enacted, and I am not confident that the collapsing democratic institutions in my own country (USA) are even capable of doing so. It's a pretty hard thing to legislate.
Even if the rule was something like each user's account had to make accessible, somewhere, the complete history of content and advertising selected for display for that user. Then, scientists and journalists could at least use random sampling of people, get their data, and do analysis that way. Although probably, with micro-targeting of individuals, we would also need the global totals for each content item or ad to also be available.
I think algorithmic selection has the potential to so terrible because it is happening in the dark. There are very few consequences for bad behavior when it is known only to the "victims".
Especially if they don't think they are victims; e.g., my young kids when they're watching the grotesque algorithmically generated videos of their favorite video game characters that YouTube suggests for them — my personal introduction to this entire topic. If you haven't seen it before, I recommend the below-linked article to anybody with children, or who just wants to think more about the implications of algorithmic/ML content selection.
[1] https://medium.com/@jamesbridle/something-is-wrong-on-the-in...
Sad, but true.