• YouTube gripped by unsubscription mass delusion

    Thousands of YouTube users have recently reported being automatically unsubscribed from channels, forcing them to resubscribe. I am going to explain what seems to be going on, and why YouTube is, in fact, not doing any such thing.

    Creators of some of the biggest YouTube channels, like PewDiePie (>50 million subscribers), have complained about the problem.

    Recently, YouTube made a video featuring its own subscriptions manager explaining that this is not something its software is capable of doing. They investigated 100 claims of the glitch and found no software issues. At the moment, that video has a striking 36,000 dislikes.

    How can so many users be wrong? Are they lying? Is YouTube? This phenomenon has been seen many times, and the latest version we are witnessing now has an identical profile to those in the past. To understand what is happening, let’s first consider a previous case where the truth is objectively known so we can see how this works.

    Audi 5000 “sudden unintended acceleration” (SUA)

    In the mid-1980’s, the National Highway Traffic Safety Administration (NHTSA), a part of the Department of Transportation, received at least 700 complaints that the Audi 5000 suddenly accelerated against the wishes of the driver. Drivers reported that when they started the vehicle, depressed the brake, and put it into gear, it immediately lurched forward in spite of the brake. The phenomenon was allegedly related to 271 injuries and 5 deaths. Things really took a turn for Audi when 60 Minutes profiled an alleged victim, a child killed in such an incident, and also aired footage of an Audi accelerating by itself. Audi denied any mechanical problem was at fault, blaming the drivers. This all seems damning for Audi, but it’s fiction. Fiction conjured by confused drivers and a predatory media. Here’s some of the reasons we know this.

    • A comprehensive, independent study [pdf] of SUA commissioned by the NHTSA concluded  “no malfunctions were found which could cause high engine power without opening the throttle.” All cases were attributed to driver error.
    • The woman who ran over her child lied on 60 Minutes. In subsequent court proceedings, witnesses to the accident reported her admitting that her foot slipped off the brake.
    • 60 Minutes fabricated the footage of the Audi accelerating on its own. It staged the scene by drilling a hole in the transmission and pumping in air at high pressure.
    • In no cases of SUA were cars found to have any mechanical defect in the brakes or throttle.
    • No standard production car on the road then, or now, has torque sufficient to overcome its braking system.
    • The same car sold hundreds of thousands of units worldwide. Yet, only in the US, where 60 Minutes aired a deceitful smear piece, did an epidemic appear.
    • After an interlock system was added, preventing shifting into drive without the brake being depressed, reports of SUA plummeted. This includes those for retrofitted cars, in which all the other systems were unchanged.
    • In more contemporary SUA studies, the very young and very old are dramatically over-represented. This shouldn’t be true for any mechanical failure that occurs when shifting a car into gear (see graph below or source pdf from NHTSA). This strongly suggests driver error is a major cause, or the cause:

    Driver error

    In order for these claims about the Audi 5000 to be true, the Audi 5000’s mechanical (none of these were electric or computer controlled) throttle and brake would have to fail simultaneously. Then, when inspected afterward, both systems repaired themselves and showed no sign of any trouble. But only in the US, not other countries driving the same car. And not the same car model with a manual transmission, whose identical brake and throttle systems somehow fared much better. This is, of course, absurd.

    Were hundreds of people lying? No, they just made a mistake and panicked. All of them? hundreds or even a few thousand? Yes, and actually, those are comparatively small numbers. By the mid 80’s Audi had at least 250,000 5000’s on US roads. If we assume each car has about 1.3 drivers (some driven by 2 or 3 people), that means 700 is just 0.3%. That’s a highly realistic proportion of drivers committing one rare type of error. We know many other types of errors are far more common over several years of driving.

    But millions of people were taken in. Audi’s sales crashed immediately and took more than a decade to recover. Now let’s compare the Audi SUA with  YouTube’s SUU, sudden unintentional unsubscription.

    1| Both alleged problems are things a user can do by mistake, but blame on a machine.

    2| The alleged problem makes no sense at all. 

    Audi SUA: It’s impossible for the car’s engine to overcome the brake under any circumstances. Brakes don’t fail, then work perfectly, etc..,
    YouTube SUU: YouTube’s business model revolves around its users easily finding and consuming its content, as much as possible, for as long as possible. Unsubscribing users would be one of the most pointlessly self-destructive things it could do.

    3| The nature of the alleged problem is such that no hard evidence is ever left behind. User testimony is the only evidence.

    Audi SUA: Mechanics and engineers investigating accidents were unable to find any problem.
    YouTube SUU: Engineers investigating reports are unable to find any problem.

    4| Large media channels drum up the alleged problem, leading it to spread and become more common. 

    Audi SUA: 60 Minutes
    YouTube SUU: Boogie2988, h3h3, PewDiePie, Kotaku

    5| Conspiracy theories proliferate.

    Audi SUA: Audi is denying safety concerns to protect its reputation.
    YouTube SUU: Many. Minimally, it’s alleged that YouTube is lying. See Boogie2988 or h3h3.

    6| Both involve motivated reasoning. It’s in one’s own interest to scapegoat.

    Audi SUA: It’s very embarrassing, legally damaging, or even psychologically damaging to admit to momentarily forgetting how to drive a car, resulting in damage, injury, or death.
    YouTube SUU: Content creators don’t want to admit they’re losing subscribers because viewers don’t want to remain. Creators almost always mention declining views when talking about this issue. Also, criticizing YouTube is always a popular topic.

    7| Reports of the alleged problem is always present in small numbers, even across many years and totally different technological iterations of the system (until sudden explosion caused by media exposure).

    Click to embiggen.
    Click to embiggen.

    Audi SUA: The SUA reports exist for all common cars. All makes, all models, every year. The numbers are trivially small, until the alarm is sounded, at which point they leap massively —>
    This is pretty amazing, considering all these different cars have different control geometries, and cover the transition from mechanical to “drive by wire” systems, power brakes, etc.., Every system has had hundreds of iterations and changes across makers, models and years. The only common denominator? The driver is human.

    YouTube SUU: The first accounts I can find come from early 2011. But you can find them from almost every month since then. Here’s a video from 2012. 2013. 2014. 2015. This makes the “glitch” explanation far less likely because YouTube has regularly, substantially updated its technology over the course of 6 years. If there was a code bug in 2011, it’s not likely that code would still be around or not otherwise obviated by other code in 2017. The common denominator is the human user.

    8| Trivial or innocuous things taken to be evidence by believers.

    Audi SUA: the placement and geometry of the pedals (note that these are not a problem in manual transmission versions of the car); small malfunctions in the throttle stabilizer that cause idling to be somewhat higher than normal sometimes. And the fact that Audi claimed no mechanical faults existed.

    YouTube SUU:  In this video, the speaker comments on the loss of thousands of subscribers in a single day for many big and small YouTube channels. This is because YouTube was scrubbing inactive/dead accounts. These did not drive views or revenue. Getting rid of them was a very good thing for content creators in the long run because advertisers are quick to realize inflated subscriber numbers are meaningless.

    The video below having “jump cuts” or the unpolished delivery of technical people whose job isn’t PR, is construed as evidence of deceit.

    9| Believers conjure irrational denials of the facts.

    Audi SUA: Michael Barr, an interested party in the suspiciously similar Toyota case, refuses to accept the facts about Audi SUA to this day. Even after the 1989 NHTSA report exonerating Audi, suspicion among consumers lingered, as did belief in real SUA.

    YouTube SUU: Apparently in an effort to be open and honest, YouTube presented the facts straight from the source. Naturally, believers read all manner of insanity into this reply. Here’s the video.

    To summarize briefly, the subscription product manager says that nobody is being automatically unsubscribed. 100 individual cases have been investigated without finding any underlying glitch. YouTube encourages users to use the feedback mechanisms when they experience problems. She also clarified that even if a spammer/bot account does not have its subscriptions removed.

    A large number of YouTube users have responded by calling it a pack of lies outright. Many, such as the popular The Know, expressed incredulity that “only” 100 cases were investigated. “If you’re scratchin’ your head because that just doesn’t seem like a very big sample size, you’re not alone…. that’s only 0.00001% of 1 billion” users.

    This is fallacious reasoning for several reasons. One, when investigating reports of something, you investigate those reports. If someone calls animal control claiming there are lions running about in central Illinois, you don’t respond by surveying all of North America to check general lion content. Two, when you want to basically establish something exists, you only need one case. If Illini are reporting that lions are around, you should easily be able to verify that with a few, if not single, follow-up. If you chased 100 leads that all turned up no evidence at all and no lions, you’d have good reason to think there’s some other explanation. Three, even in a controlled scientific experiment, 100 is an excellent sample size for statistical power purposes in most cases. Statistical power doesn’t rely on being proportionate to a sample population. Stop talking about statistics if you don’t know anything about it. Four, nobody is considering that 1 billion term when thinking about the base rate of complaints. Let’s say that there have been 5,000 complaints (I don’t know the real number, I doubt it is more than this). That’s just 0.0005% of all users. Could that be the result of user error, wayward browser plugins, or trolls? Of course it could. Were it ten times as much, it could. Five, proper case investigation takes time, and in this case relies on communicating with users who may or may not reply promptly or coherently. Without knowing the methods, there’s no reason to think 100 is too small.

    What’s really going on here?

    User error. However, there is some reason to think a technological component exists. It doesn’t involve YouTube unsubscribing anyone, and maybe it doesn’t involve YouTube at all. In this video “Helvetica” claims to have proof of this bug. The only proof we can see, is a channel he “resubscribed” to double-listed in his subscription list. He misidentifies the problem he describes as “being unsubscribed” when it sounds more like seeing a false indicator of not being subscribed. This is very similar to the first account I found from 2011. Some sort of glitch in presentation could lead people to click on the button thinking it will sub them, but since it is a toggle, it unsubs them.

    This could be the result of a web browser problem, caching/scripting problems, meddlesome browser plugins, connectivity problems, to name just a few. This is because the button reading “Subscribe” is the default presentation. It looks that way even if you’re not logged in and can’t subscribe until you do. Subsequent (honest) investigation by YouTube would probably not find evidence of any of these things if they are transient and rare.

    This can’t explain the broad phenomenon because some accounts state the problem is noticing they are not seeing videos they should from channels that they frequent. In these cases, if they had had to suddenly re-subscribe out of the blue, they should remember having done so.

    The explanation is thus most likely a very small number of related technical glitches + user error (misreading the subscribe button) + people prevaricating in support of fans, favorites, or over general angst about YouTube. Much like the last time. Much like the next time.

    Advice to YouTube to fix this: Remove this non-functional default “Subscribe” button entirely; for users that have been subscribed to a channel more than a week, pop up a confirmation prompt asking if they are sure they want to unsubscribe before they can unsubscribe from a channel. This will remove all possible client-side glitches and bears a certain metaphorical similarity to Audi’s brake interlock.


    Read more about historical mass delusions at the Committee for Skeptical Inquiry. I especially like the story of the genital bandits.
    Mass Delusions and Hysterias: Highlights from the Past Millennium

    Category: Critical ThinkingSkeptic Ink News and Reportskepticism

  • Article by: Edward Clint

    Ed Clint is an evolutionary psychologist, co-founder of Skeptic Ink, and USAF veteran.
    • im-skeptical

      Much wailing and gnashing of teeth results from a flawed user interface.

      • The sheer size of YouTube probably magnifies this problem. The software could be designed very well, such that only a tiny fraction of users have problems- and those problems could be caused by other software YT doesn’t control. But there’s so many users that these instances can snowball.

    • LindaL

      Did you speak to several of the big YouTube content creators who make this claim? They could show you their own histories to see if that contradicts YouTube’s claims.

      While the user interface could be an issue, and your suggestion is needed (a confirmation pop up), my own anecdotal experience is that I am being unsubscribed from accounts I normally see on my TV, so I don’t have easy access to rate or subscribe/unsubscribe.
      YouTube can tell I am seeing the vids from these accounts, but I may not get to rate them or make comments. Could I be wrong? Sure. But as I never get a notification I am no longer subscribed to a channel, I may never know until weeks later when I suddenly notice I don’t see them in my feed that much anymore.

      If your contention is they only remove dead accounts, that wouldn’t affect me.
      If it is only dead accounts, the content creators claiming they lose a far larger than normal amount of subscribers every time they upload new videos. These are people that have been content creators for years, over 5 in several cases that come to my mind. These are people that pay attention to when they gain and lose subscribers because this is actually their job. They alter their content to gain new viewers and curtail content if it seems to turn people away. They will notice patterns and larger than normal losses.

      I still think this could be a real issue.

      • I do not know how their history could contradict YouTube’s claims. If it happened that a person apparently ceased to be a subscriber to a particular channel, would it prove they were SUU’d? No.They could have made an error. Especially if they are subscribing to channels frequently.

        I don’t know what explains your experience. But one must proceed on the evidence available. When there is a lot on the line, we can’t simply take one’s word for it in lieu of harder evidence. If you think that’s unreasonable, imagine someone accused you of a crime that you could go to jail for. Now do you think their word alone is good enough to convict you?

        re: content creators and subscriber counts. Creators are the people who are least likely to objectively view their subscription gains and losses because they have so much self-interest invested. I mentioned the case of YouTube scrubbing dead/fake accounts. This was a big net benefit to content creators because it makes their numbers accurate and therefore, useful to advertisers who pay those creators. But they didn’t respond positively, did they? They did not view the situation objectively, or correctly, because they attached ego and emotion to their counts. It’s a powerful bias, even self-destructively sometimes.

    • Great explanation and comparison with the Audi thing. I actually didn’t know about that.

    • ncovington89

      This is interesting! Great for understanding mass delusion.

    • TruthDigger

      This is no example of a mass delusion?! Nothing more than mistaken YouTubers. But CLINT! Why DEFEND a BIGOT and a RACIST?! Why CUNT?! WHY?? Sorry – Clint!! That says alot about you and this website. Will you at some point be doing the Nazi salute like your favorite leader Trump? You think you so smart? HA!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! HATER!!!

      • This sort of language and aggression are not appropriate here, even if I had some idea what you’re talking about. Edit your remarks or I will remove them. If you can’t compose yourself and speak calmly, this is not the site for you.

    • TruthDigger

      You sure don’t give someone much time to respond or edit their comments. This was a test. My first comment you deleted without comment. I was polite and asked a simple reasonable question of you and you dismiss it. Here I berate you and you not only keep my comment but reply!!! Not very consistent of you. Since you seem to be in a more talkative mood I ask again. Why do you tolerate the prussion’s obvious bigotry? Do you agree with his positions or not? If so why? If not why not? It’s a very simple question. If this comment gets deleted without reply I think I will know where you stand and think it’s disgusting and your disregarding every principle you claim to hold dear, namely evidence and liberalism is shaemful. But I don’t jump to conclusions just yet. I will see what you do first.

      • You had hours. You weren’t polite. You called me “Cunt”. You called me many absurd, insulting things.
        Because of your poor behavior, you’ve no right to question me here about anything. Which I probably would not address anyway, because it’s not the topic of this post. It’s very easy to contact me. You chose not to do that, and to misuse comment sections instead. You are not welcome here. Please do not comment again.

    • David Giarratana

      A few points (note: I have not, to my knowledge, experienced this problem):
      1) Youtube has done many, pointlessly self-destructive things since Google took over. The response to the smear pieces by the WSJ and others stands out most, as demonetizing independent creators in favor of competing with Netflix, Amazon and Hulu is just about the most idiotic plan I have ever heard. They have the market on a specific type of content monopolized, so they decided to make even less money and discourage current and future independent content creators from providing the very content that attracts people to the site in the first place. There are other examples, such as the banning of content many people want to see through poorly-made algorithms meant to target other types of content, but you get the point.

      2) “Engineers investigating reports are unable to find any problem.” This does not overlap well with your Audi example, as Youtube’s code is proprietary; it’s a black box.

      3)”Content creators don’t want to admit they’re losing subscribers because viewers don’t want to remain.” It is impossible for you to come to any such conclusion without looking at the numbers. I have seen data from several Youtubers showing precipitous decline in less than 24 hours (we’re talking sheer cliffs). If you don’t find that suspicious, I have to ask why thousands of people would unsubscribe in less than a day. Is there a Twitter movement I don’t know about? Perhaps this is all explicable by the scrubbing of inactive accounts, but do you actually know that to be the case?

      4)”YouTube has regularly, substantially updated its technology over the course of 6 years.” Do you have access to the source over the last 6 years? How do you know how “substantial” their updates have or have not been?

      5) Part of the problem is that Youtube is inaccessible. It’s not possible for the vast majority of people to interact with anyone who knows any details about the goings-on behind the scenes, and most complaints are shunted to a forum led by Youtube sycophants who really don’t know anything useful and never provide much in the way of help. Not only does this provide bad optics, it makes it impossible to fact check their claims. Yes, they could be lying. You don’t know, and neither do I.

      I’m not saying I buy into the idea that Youtube is unsubscribing people, just that you have not demonstrated that this is not the case.

      • 1) I don’t believe this is so. Self-interested mistakes aren’t the same thing as pointlessly self-destructive actions.

        2) It’s a fair point. However, in this case Google, like Audi, was facing lots of angry people. It would have been in its interest to say “this was a bug, we’re fixing it, sorry”. This would have validated people’s concerns and made them feel like at least there was a chance they were being heard and addressed. The SUA issue with Toyota years after the Audi hysteria left Toyota lying about having found mechanical or electrical “causes”, just to get some of the heat off. So the interest here isn’t covering up bugs. A bug would be great.

        3) Yes and lots of creators had legions of “dead” subscribers. For any one channel, of course I do not know the exact reasons. But yes I do know for a fact that YouTube sometimes scrubs inactive/dead accounts and has told creators they may see drops in subs.

        4) Your claim is that there has been no substantial change in YouTube software in 6 years? I think that would be astonishing considering all the obvious changes to its functionality, capability, and performance.

        5) I have made my case. If you do not find it persuasive, then that is your prerogative.

        • David Giarratana

          1) So your contention must be that this putative unsubscribing problem falls decidedly into the latter category while the other examples fall decidedly into the former. I don’t agree, but I suspect that it would take undue time to expound on the difference. I’ll count this as a case of “agree to disagree”.

          2) An interesting counterargument, though I have to say that, at this point, “this was a bug, we’re fixing it, sorry” has been their stock response, and I don’t think the majority of people are buying that, anymore. Regardless, there are no engineers outside of Google who can comment on the issue, unlike the Audi case.

          3) I am not asking whether you know that this does happen, I am asking whether you know that these cases can all be explained as such. I am saying that you cannot know, that you are arguing probabilities without trials, statistics without data.

          4) No. My claim is that the changes have been mostly “under the hood”, and their substance is as much a mystery to you as it is to me.

          5) Fair enough. I’ll stand by my rebuttal by your leave.

          • 2) I don’t agree. And the truth would have the virtue of being.. true. Meaning one could produce evidence for it, both before addressing it and after. And this doesn’t just affect creators, but regular users too. The people YT needs to stay on their site as much as possible.

            3) Correct. I do not know what lead you to believe this blog was a scientific or forensic laboratory. It is not. It is a blog.

            4) That I am not familiar with the nuts and bolts does not alter the fact that there have been great changes. I do not know what the “hood”, what is under or over it means in your metaphor. One does not need to understand the details of a turbojet engine to understand a jet aircraft is significantly different from a prop-driven plane.

            I am glad you are skeptical. But you do not seem to be as skeptical of views that are critical of YT. You seem mighty sure they’re icky bad, and tend to suggest explanations that cast it in a negative light, even without any evidence to support those views. If you can dismiss my lack of direct empirical evidence, I can discount yours, too.

            My evidence is circumstantial, but it is based in sociological phenomena that repeat themselves over and over again. An explanation is supported when it can account for an otherwise unlikely pattern of evidence that is not well accounted for by others (if any). If it can make predictions and offer insight. The pattern with Audi repeated again with Toyota with every feature intact. And there’s reason to think it is happening again with Tesla and self-driving cars (but this is being stymied by the fact that Tesla’s record every driver control action, thus puncturing the erroneous accounts of not having stepped on the accelerator). But those accounts momentarily do exist…

            • David Giarratana

              I get the impression that I should emphasize that I am not trying to put you on the back foot. I believe that I have made it crystal clear that I am not assuming anything about Youtube. Rather, I am raising the point that there are alternative explanations for these results, and that there are things Youtube have done in the past that might make their user base less inclined to trust them. I am not dismissing your lack of direct empirical evidence, I am calling attention to it, and to the fact that this lack is shared by everyone not working at Youtube HQ.

              2) I don’t get this. Are you saying that there are engineers outside of Google who can comment on this with comparable knowledge? Are you saying that “it was a bug” hasn’t been their stock response for years? Did you type the wrong number? I don’t follow, I’m sorry.

              3) I do not believe this is a laboratory of any kind. You know that, of course, but please try to assume good faith.

              4) Using your analogy, we’re really talking about the difference between a 2010 Boeing turbofan engine and a 2016 Boeing turbofan engine: the developments are incremental and difficult for the layman to understand. As such, the word “substantial” seems to be the word requiring further explanation, not the word “hood”.

              I am not discounting your expertise. I am merely calling into question the validity of your conclusions. Dismiss me as a selective skeptic if you wish.

            • I am not assuming anything about Youtube

              You appear to have a bias against it. You seem especially willing to offer “alternatives” that cast it in a negative light rather than alternatives that favor or neither favor or disfavor it. Having a bias does not make your opinion incorrect, but at least own up to it.

              I am not dismissing your lack of direct empirical evidence, I am calling attention to it

              Not sure why this was necessary as I’ve not claimed it (beyond what is cited above), and my argument does not rest on having it or not. Bringing it up suggests otherwise.

              2) No, I am not saying that.

              Are you saying that “it was a bug” hasn’t been their stock response for years?

              Even if that is true, it does not support your argument because in this case they are not saying it is a bug. If their “stock” reply is to disclaim concerns as bugs, they should do so here. They emphatically deny that it is and cite investigation that turned up nothing.

              3) You brought up ” that you are arguing probabilities without trials, statistics without data”. This suggests that is somehow expected or necessary, neither is true. I do not know why you would say such a thing, so I had to guess you were unsure what sort of page this is.

              4) That would only be true if the pace of change was identical between the items being compared. It is not. The point here is that the same bugs are unlikely to persist across many major revisions of a software; especially as precisely those features have been given to substantial change- rewritten and heavily modified. There didn’t used to be bells and notifications options. There didn’t used to a like/dislike rating system. For that matter, there didn’t used to be: ads, rentals, livestream, HD, 60fps, filtering, and the whole site was overhauled in 2011. I’ve talked about 6 years, but that’s just a relatively brief bit of investigation. I’d be surprised if the same sort of reports didn’t go earlier still.

              And yes, we can be sure there were major revisions even without being Google engineers. The world around YT has changed dramatically over its lifetime. It has had to scale up, from thousands to billions of views a day. You can’t do that without substantially changing the technology. Then there are things like the APIs that have all been born after YT. It had to be modified to interoperate with them. Then there are all the business changes. Changing demands of advertisers, changing demands of data analysis and reporting.. demands for more detail and new sorts of insights. There’s an evolving social and business landscape of content rights and copy protection, of censorship and ratings. When you serve 5 billion videos a day, you can’t band-aid each of these new changes and needs and not have it all fall apart.

              I don’t know of any bugs that have remained, essentially intact and identical, across 6+ years of major complex software like Windows, MacOS, or Facebook.