Author: Dr Implausible

Tiktok has a noise problem, and it’s indicative of a larger issue ongoing within social media, that of “context collapse”. But even context collapse is expanding outside its original context and evidence of it can be seen in the rise of generative AI tools, music and media, and the rise of the “Everything App”. Starting with a baseline in information theory and anthropology, we’ll outline some of the implications of noise and context collapse in this episode of the Implausipod.

https://www.implausipod.com/1935232/13516713-implausipod-e0013-context-collapse

Transcript:

 TikTok has a noise problem, and it may be due to a context collapse, something that’s been plaguing music, social media, and it’s even showing up in our new AI tools. And if you don’t know what that is, you’ll find out soon enough. We’ll explain it here tonight on episode 13 of the Implausipod.

Welcome to the ImplausiPod, a podcast about the intersection of art, technology, and popular culture. I’m your host, Dr. Implausible. Now, when it comes to the issue of noise and context collapse, there’s a little bit more going on, of course. The problem for TikTok is that it started out with a pretty tasty signal, one that kind of really encouraged people to stick around.  But as that signal amps up and it gets more and more noise in the system, it gets a little chunkier and crustier and maybe not as finely tuned as you’d like. Now, for some people that noise isn’t a problem, but for a lot of people it can be. And the reason it’s a problem for TikTok is that the noise can be actively discouraging from using the app.  It can make it Unfun, and this is what I’ve been noticing lately. So let’s get into how context collapse is impacting life online.

When TikTok rose to prominence throughout the pandemic, it was a very tasty experience for a lot of people. I mean, if you had negative interactions there, there was probably reasons for it, but there was also ways to mitigate it.  You could block people, you had a lot of control, and generally the algorithm would be feeding you content that you wanted to see. Or even if you know, you didn’t know you wanted to see it, you know that the joke goes. To that end, it was pretty good at sussing out what people found engaging. So TikTok had a very high signal to noise ratio.  Yeah, there was some noise there, but that was because it was feeding stuff that it wasn’t quite sure that you liked. But once it kind of honed in on what your preferences were, it was really good system for delivering content to users.

Over time though, as more and more content goes out and more and more people start participating, the amount of tasty content, the amount of good content, the amount of interesting and novel content drops off.  So you see less and are aware of pieces of information that everybody is seeing less, and less stuff – even within your niche from people that you’re following – gets shown to you. So this is all noise in the system. It’s the amount of stuff that you don’t want to see increasing.

Now we’re talking about signal to noise, and as we’re talking about a very old theory here, we’re talking about Claude Shannon’s Mathematical Theory of Communication.  Now, it was “A mathematical theory of communication” when it was published in 1948 as a paper, and then it was reworked as a book with Warren Weaver in 1949, where it was The Mathematical Theory of Communication as they realized that the theory was more generalizable, and this theory undergirds the entirety of the internet and most of our modern telecommunication systems, and it’s just a way of dealing with the noise in a system and ensuring the signal gets sent as it was sent from the transmitter to the receiver. And you can talk about it in terms of human communication or machine to machine communication. Device to device. Point to point, and this is why it’s generalizable.  It can be pretty much black boxed, and you can see this in how it gets used in multiple contexts. The point of the theory is that there’s a certain throughput that you need where the amount of information is greater than the noise to ensure that the signal is “understood”. And then there can be systems that are used to error check or correct or whatever, what’s on the receiving end to ensure that you know what was transmitted comes through as an intended, and that’s the gist of it.

Now for something like TikTok as the signal, you know, the signal is the content that’s supposed to be delivered to the end user, and the noise is anything that isn’t part of that. It’s the stuff they’re not necessarily looking for or asking for. And as TikTok has branched out and provided more types of content, starting with the 15 second videos and then 60 seconds, three minutes, 10 minutes, live stream, stories, whatever, you get more types of content in there.  Not all of it’s gonna be relevant to all users. If somebody’s watching for some quick videos, even a 60-second or three minute video is definitely not gonna be what they want to see. So we have a variety of content in there and that increases the noise, the amount of stuff you don’t want to see in a given block of time.

Now, couple that with the other types of content that get filtered in. It can include ad sponsored posts or posts that are just generally low value. This can include things like, oh, so-and-so changed their name, so-and-so signed on, or what we’ve seen recently is the retro posts like on this day in 2021 or 2022 or whatever, where people will revisit old posts, and a lot of times there’s nothing special about those unless you haven’t seen it before. It’s just whatever’s that person was talking about a year ago. So that feeds into the pipeline with all the current content that’s also trying to get out to the user base as the user base is increasing. So we have this additional content that’s coming through the pipeline, increasing the signal, but there’s also more stuff, more stuff that you don’t want to see.

It’s noisy,

and that noise, as we stated earlier, makes it unfun. It’s like it directly interferes with the stickiness of the app, the ability for it to engage the audience and have them participate in what the actions that are going online. And as that’s directly part of what Tiktok’s business model is: capture an audience and keep them around, then that can be a problem for them.

But it also brings us into that idea of the collapse of context. Now context collapse is something that was theorized about by a number of media scholars in the early 2000s, including danah boyd and Michael Wesch, and a few others. In its most simplest form, it’s what happens when media that’s designed for one audience or a single audience gets shared to multiple audiences, sometimes unintended. For early social media, and in this case, that means like MySpace and Facebook and Twitter, media that was shared for a particular group – often a friend group – could go far beyond the initial context. And while those websites or apps, along with blogs and web forums were co-constitutive of the public sphere, as we talked about a few episodes ago, along with the traditional media. Context really didn’t start smooshing together until Web 2.0 started shifting to video with the advent of YouTube and the other streaming sites, and that’s the technical term, smooshing. You can update your lexicons accordingly.

But the best way to describe context collapse was captured by cultural anthropologist Michael Wesch in a 2009 issue of Explorations in Media Ecology. He describes it and the problem as follows, quote:

“The problem is not lack of context. It’s context collapse, an infinite number of contexts collapsing upon one another into that single moment of recording.  The images, actions, and words captured by the lens at any moment can be transported to anywhere on the planet and preserved the performer must assume for all time. The little glass lens becomes the gateway to a black hole sucking all of time and space, virtually all possible contexts in on itself.” End quote.

So he is talking then about the relatively new phenomenon of YouTube, which had only been around for about four or five years at that point, and what we now call creators producing content for viewing on that platform. It was that shift to cam life that had started previously, obviously, I mean there’s a reason YouTube was called what it was, but it went along with that idea of democratization of the technology, of the ability for pretty much anybody with a small technological outlay to produce a video and have it available online for others to see.  Prior to the YouTube era, that would’ve been largely restricted to people with access to certain levels of broadcast technology, whether it was television or cable access, or a few other avenues. It wasn’t really as prevalent as we saw in, you know, the 21st century. And now with the growth of YouTube and the advent of Snapchat and TikTok, it really has completely taken over. But this is why it’s also still useful to look at some older articles because they give us an idea of what was novel at the time, what had changed, and this was really what was different with what was going on.

Michael Wesch is really drawing a lot from Goffman here and that idea of “the presentation of self in everyday life”, that we have different behaviors and there’s different aspects of ourselves that we will bring to the forefront in different contexts. So whether it’s at school or work or with our family or parents or friends or loved ones or what have you, we’re all slightly different in the way that we act around them. And this has been observed for a lot of different people in a lot of different contexts. But with the rise of what I’ll call here the mediated self and the complete flattening of all contexts due to, you know, Snapchat and Reels and TikTok, it has really taken a new turn.

Now, that idea of presentation of self for multiple audiences through vlogging, through YouTube, it isn’t exactly new because there was other versions of that before.  In a presentation by Dr. Aiden Buckland, he goes into some of the critiques of this, that a media archeologist or media historian could draw a pretty straight lineage from diarization and life writing as a practice that occurred on blogs through to the modern practices that we see with video logs or just TikTok and Snapchats.  This, in turn, is drawing heavily on the works of Dr. Michael Keren, who wrote a lot about blogs and their political action in the late nineties and early 2000s. But I digress. I’m starting to get a little bit further afield.

One of the ways to theorize Context collapse is that it’s like if every moment that you have that is recorded was available for instant replay at any time.  And with the advent of video services moving to the cloud and having everything accessible (and looking at YouTube’s archives, now you can go back to basically when they began), we have that idea of instant replay. So it isn’t just a context collapse in terms of anything might be available to multiple audiences, but it’s also a Time collapse in that everything is always available to all potential audiences, and this extension of the context collapse to encompass multiple times or at least all times that are recorded and stored in the cloud has been discussed by authors Petter Bae Brandtzaeg of Oslo and Marika Lüders. Now there’s a very obvious link to this, to the rise of what’s called cancel culture, and I’d be remiss if I went without mentioning it, but that’s kind of beyond the scope of what we’re discussing here. That’s a different thread, a different track that we will have to pursue at some time in the future. The other implication of this time collapse is something that we’ve discussing here on the podcast more recently, namely media, especially music,  and AI.

In terms of media, this context collapse, this time collapse is happening because obviously everything is available everywhere, all at once, at least for the most part. Things are currently in a state of flux, especially when it comes to television and film. The advent of the streaming services where each carved off a particular portion of the IP catalog that they happen to own has really changed how things have been interacting, but when it comes to music where streaming can basically all be done through one particular service, Spotify, with a few additional ones with minor catalogs, the impacts of that time collapse and context collapse are much more noticeable.

In an article published on The Atlantic in January of 2022, author Ted Gioia asked “Is old music killing new music?”. The author found that over 70% of the US market was going to songs that were 18 months or older, and often significantly so. Current rock and pop tracks now have to compete with the best of the last 60 years of recorded music. And while it is possible to draw some direct comparisons between the quality of the music as YouTuber Rick Beato did in a live stream on August 26th, 2023, where he asked: “Is today’s music bad?”, and looked at the top chart toppers from 50 years ago in August of 1973. You can argue that the overall production of music may be significantly better in 2023, but the overall composition, songwriting, and other elements may lack that magic that we saw, you know, 50 years ago. The most popular trend in music right now seems to just be a remix, a sample, a cover, or an interpolation of an older song.  Even a chart topper like Dua Lipa draws heavily on the recreation of a seventies dance club aesthetic and sound. So context collapse, even if it isn’t necessarily killing new music, is definitely changing the environment in which it may be able to, you know, survive and thrive. The environment’s almost getting a little polluted.

It’s very noisy there.

However, one of the other places we’re seeing the impacts of this noise, this context collapse, is in the generative AI tools, or at least this is one of the places that the noise is being put to use. On a post on his blog on July 17th, 2023, author Stephen Wolfram talked about the development of these generative art tools and the processes that it goes through to actually create a picture.  We work through the field of adjacent possibles that could be seen in something like a cat with a party hat on, and a lot of those images that are just a step or two removed for being a image that we as humans recognize shows up as noise. It turns out that what we think of as an image isn’t necessarily that random, and a lot of the pixels are highly correlated with one another, at least on a pixel-per-pixel basis. So if you feed a billion images into one of these models, in order to train it, you’re gonna get a lot of images that look highly similar, that are correlated with each other. And this is what Wolfram is talking about when he is talking about the idea of an “inter concept space”, that these images generally represent something or close to something. It’s not an arbitrary one either, but it’s one that’s aligned with our vision, something that we recognize, so a “human-aligned inter-concept space” that’s tied to our conception of things like cats and party hats.

But this “inter-concept” space is not only like ‘representative of’, but ‘fueled by’ the context collapse.  It requires the digitization of everything, like a billion images that go into it in order for it to be trained. But it also, you know, squishes everything together. Again, our technical term, smoosh. And this smooshing brings us back to TikTok because everything is there. That’s part of what’s contributing to the noise, but it also is why there’s such a volume of a signal that’s there. You can likely find something and it’ll get algorithmically delivered to you if you like it enough or you interact with. But this is also how it’s captured so much of the public sphere in a way that the owner of Twitter wishes it could, and that idea of the context collapse seems to be made manifest in these apps that are trying to capture the public sphere, that they have to capture everything, everything all at once.

And so we’re seeing the rise of the Everything app, the everything website, much like we talked about a few weeks ago in episode 10 with the rise of a o l and how it as a portal was for a lot of users. The internet, it was the entirety of it. And we’ve seen subsequently with Facebook, we’re seeing a number of competitors, sometimes in different places around the world, catering to a particular locality, but all of them trying to capture that “One thing to all people, to all customers”. In China, we see it with the rise of WeChat, which allows for calls and texts and payments as well. In Moscow, we can see it with the various apps that are run by Yandex, where you could use it for everything from getting a taxi to communications to your apartment, and there’s a lot of tools built-in and it actually has its own currency system built-in as well. A user by the name of Inex Code posted a list of everything that you can do with Yandex in Moscow. In North America, we can see it with not just Facebook, but also with Apple and Google and Amazon too. The breadth of services that they have available, and the continual expansion of services that they’re adding to their apps and platforms. And when Elon Musk bought Twitter, it was theorized that one of the things you wanted to do was turn it into a WeChat like app. His recent comments about LinkedIn and the option of adding that kind of functionality to the app now known as X indicate that he may well be headed in that direction.  And finally, the continual expansion of TikTok now include texts as well as a marketplace and music sales indicate there’s still more growth in that area too. As each of these walled garden “everything” apps try and gather up more functionality, we can see that this is one response to the context collapse: to provide a specific context within their enclosure.

It’s an effort to reduce the noise, or at least to turn it into something that happens outside their walls.

But setting up a wall may not be the only solution. It’s one way, obviously, that element of enclosure that’s taking place, but there’s other ways to deal with it as well. One way is a way we looked at with the Fediverse, where an everything app can be developed as long as it’s open. and there’s a lot of opportunity and possibility there, but that openness requires a fair amount of work by the user. It requires curation. It lacks the algorithmic elements that drive the enclosure of the other apps. Now, that doesn’t mean an algorithmic element couldn’t work for the Fediverse, it’s just that currently it’s not set up for it and may require a lot of effort to bootstrap something like that and get it going.

And absent an algorithm, it kind of points the way to the last two solutions that we have. The first one is just to lean into it to accept that there’s this change that’s happened to our society with the advent of digital media and everything being available. If the context collapsed, that’s fine. That’s just the way things are now, and we just have to learn to deal with it. And that leads into the second option. The one David Brin called The Transparent Society. And just that everything is available, and we’ll have to change our patterns of use. If we recognize that aspects of our culture are socially constructed, then we learn to live with that and we can change and adjust as necessary.  Things haven’t always been the way they are currently, and they don’t have to continue that way either. Because the last way forward to deal with context collapse is to look at some areas of our culture that have already experienced it and seen how they’ve dealt with it. Because context collapse is intimately tied with that idea of availability of everything as well as in video terms, what Wesch is talking about was the instant replay.

And the two areas that have managed that and have continued to succeed in an era of streaming media and context collapse are pro sports and pro wrestling. The way they’ve succeeded is recognizing that they have their particular audience, that their audience will find them, that they don’t have to be everything for all audiences.  And they’ve also succeeded by privileging the live, the now, the current event, something that revels in the instant replay, the highlight reel, the high spot, but also is allowed to continually produce new content because there might be a new highlight reel or a high spot in the very next game or match or show or finals or pay-per-view.  There’s always something new coming down the pipeline and you best not look away. It turns out that the best way to deal with the noise is to create something that cuts right through it.

Once again, I’m your host, Dr. Implausible. It’s been a pleasure having you with us today. I hope you join us next time for episode 14 when we investigate the phenomenon of the dumpshock. In the meantime, you can find this episode and all back episodes at our new online home at www.implausipod.com, and email me at Dr. implausible at implausipod com. Until the next time, while you’re out there in the busyness and the noise, have fun.

References and Links:

Brandtzaeg, P. B., & Lüders, M. (2018). Time Collapse in Social Media: Extending the Context Collapse. Social Media + Society, 4(1), 2056305118763349. https://doi.org/10.1177/2056305118763349

Gioia, T. (2022, January 23). Is Old Music Killing New Music? The Atlantic. https://www.theatlantic.com/ideas/archive/2022/01/old-music-killing-new-music/621339/

Shannon, C. E. (1948). A Mathematical Theory of Communication.

Wesch, M. (2009). Youtube and You: Experiences of self-awareness in the context collapse of the recording webcam. Explorations in Media Ecology, 8(2), 19–34.

https://www.quantamagazine.org/how-claude-shannons-information-theory-invented-the-future-20201222/

https://journals.sagepub.com/doi/10.1177/2056305118763349

Generative AI Space and the Mental Imagery of Alien Minds