Episode 3: Eternal Memory of the Facebook Mind

Social media and streaming platforms such as Facebook and Spotify analyze huge quantities of data from users before feeding selections back as personal “memories.” How do the algorithms select which content to turn into memories? And how does this feature affect the way we remember—and even what we think memory is? We spoke to David Beer, professor of sociology at the University of York, about how algorithms and classifications play an increasingly important role in producing and shaping what we remember about the past.


Jason Lloyd: Welcome to The Ongoing Transformation, a podcast from Issues in Science and Technology. Issues is a quarterly journal published by the National Academies of Sciences, Engineering, and Medicine, and Arizona State University. I’m Jason Lloyd, the managing editor of Issues. On this episode, I’m talking to David Beer about social media and how these platforms, algorithms, and classifications play a role in shaping our memories and reality. David is a professor of sociology at the University of York. His new book, Social Media and the Automatic Production of Memory, which he co-authored with Ben Jacobsen, explores these themes.

Dave, thanks for joining us today. I just finished your book and it got my brain spinning in really terrific ways. You also recently reviewed David Arditi’s book Streaming Culture for Issues, which dealt with similar topics, and may be a good place to start. Could you tell me a little bit about “Chamber Psych?”

David Beer: Thanks. The “Chamber Psych” thing I started the review with, that was from Spotify, and it was the End of Year Review thing that they produce. It’s like an automated narrative of your music tastes that Spotify creates on your behalf and provides you with. It tells a bit of a story. And one of the aspects of the story is about genre.

So I’ve had this interest in genre as part of a broader interest in archiving—how media, new types of media platforms, [operate] through an archival lens. And then you start think how are they organized, what are the classifications that are going on within these spaces, and that sort of thing. They’re really interesting, the grids that we put culture into are really interesting and very vibrant, I think, as a result of the new types of media structures.

So my top five genres last year—I can’t remember now which number it was at, but in the top five was “Chamber Psych.” I didn’t know what that referred to, or which artist, which songs that referred to. It was a genre label I was unfamiliar with.

I found myself, as I mentioned in the review, searching for the genre that described my own taste so that I could understand how my tastes were being classified, really, or categorized. I thought that was interesting. I think the thing that drove it—I don’t mention this in the review—I think the thing that drove it was listening to the band Super Furry Animals and then a couple of other things that I’d listened to must have been categorized as that as well, I think.

Lloyd: And that’s how the platform slots those artists, or does it categorize them by album, or song?

Beer: Yeah, I think it’s by song. But I suspect it filters through from the artist. So the songs, I imagine, can be tagged with more than one classificatory label, but that seems to be the one. But I’ve been listening to Super [Furry] Animals for about 25 years and I’ve never heard of Chamber Psych.

It’s interesting how these labels then start to take on a presence within these platforms, even if they’re not perhaps labels that we might use ourselves. The other genres are more familiar, but it just struck me that it’s indicative of the vibrancy of the kind of classificatory systems that are going on in media structures.

Lloyd: I was really struck by the fact, I think you mentioned in the review, that Spotify has more than 2,000 different types of classifications for the various types of music that they host—Chamber Psych, obviously, being one of them. Could you talk a little bit more about the role of classification—obviously, on a streaming platform like Spotify, but also in other social media platforms such as Facebook?

Beer: Well, I think consumer culture is full of this kind of classificatory system. The comparable thing on Spotify you get on Netflix, you get on Amazon: these labels, very specific, granular types of labels that are used to organize content. I think this is driven, in part, by people’s involvement in the platforms, but it’s also to do with the amount of content that there is out there, and that needs to be organized.

We’re familiar with ideas about algorithms presenting content back to us, but I think alongside that, we sometimes focus less upon the classificatory systems, the kind of archival systems that are active. So once you get all this massive, massive content—like you get all these films, all these TV shows, all these podcasts, all these songs available, and you move to an access-type cultural consumption—you need ways of organizing that, so that it’s manageable. It needs to be rendered fathomable to a consumer, and one of the ways this works is through classificatory systems taking on a greater level of significance within the media structures, so that people can then find their way around and find culture they might be interested in.

So you get the automated thing presenting you with suggestions, but there’s also the classificatory structures that allow us to organize this content in different ways. You get that in the consumer-culture platforms like Spotify, Netflix, Amazon, and so on, but you also then get it within social media with people using hashtags and stuff to try to organize content within those spaces. There’s classificatory systems that work there as well, and you had it with tagging in the past—tagging photos and so on.

So these are classificatory systems. Some of them are user-generated classifications, others feed off that but are led by the platforms, and there’s this interesting mix there of everyday classifications that people use, combined with the classificatory schemas that are applied, or imposed, onto culture by the platforms themselves, or actors within them. It’s a quite interesting mixture, I think, of agendas going on within classification of culture and content. But behind it all is there’s so much there. You need ways, then, of managing it, and these are vast archives, as I see them as, of content that classificatory systems allow us to access and allow us to retrieve the things that we might be interested in.

Lloyd: I was really struck. I had not given this any thought, how a platform like Facebook takes a post and assesses it for how it’s going to classify it, then how it ranks it, and then how it determines whether or not it’s going to feed it back to you as a memory. I assume it has some sort of AI that looks at the image itself or classifies the image in some way, maybe fairly generally as “two people on a beach” or something like that, and then it also looks at how it’s tagged, the comments on it, I guess they do some linguistic analysis on whether those comments tend to be good or bad, and so the platform purports to have a sense of whether or not this post or this photograph will be positive or negative for the user. I had not thought about that.

Beer: And that’s where the interesting thing of these things is archival, and thinking about classifications. To understand the kind of politics of those archival structures, what they allow to be said, what comes back to us, what we see, what we encounter, then, are all a product of those archival structures and that kind of politics.

Memory, then, is a part of that. So that was the collaborative work I’d done with Ben Jacobsen, who has written widely about algorithmic memory as well. And, yes, we were trying to understand, in particular, the classification and ranking processes, and then how people responded to them. That’s the three movements in the book.

One of the starting points, really, was understanding that Facebook had a taxonomy of memories—types—and the content was then slotted into, literally, a grid. And, in the way that you’ve described, it’s assessed, and the images, the comments, and so on are used to place memories within a grid. … As we live in social media, they become memory devices. And that past content then becomes slots into these pigeon holes, these categories of types. And in that moment, you’re deciding, “Well, what, of that content we create, constitutes a memory?” And also then “What type of memory is it?” So at that moment, memory becomes part of the logic of social media. Because it’s the types of things that social media want us to engage with as memories—[those] are going to be the types things driven by the logic of stickiness, and sharing, and commenting, and engagement.

We thought that was quite interesting to see the grid. We use Facebook in this book, but what we were pointing towards is a broader set of trends in social media through memories: a memory culture. And we’ve also got mobile devices doing something similar, but we used Facebook as a way into that because we’d got this taxonomy. So we did that, yes. And then, once classified, they’re ranked for their worth or value, which is partly to do with the prediction of what the person will want to remember and when, and then we looked at how people respond to those recirculated and sorted versions of their own past.

Lloyd: You touched on this just now, but when I think about the goals of the platform, what this kind of classification scheme, the targeting and regurgitating of your memories, what the objective is for that, I’m thinking of other related concepts like James Scott’s ideas about legibility in the state: in order to get visibility and control over the population, they need a census, a sense of who lives there and where they are and what they do. In this case, it wouldn’t be for something like taxation or conscription—why is this useful for Facebook? What do they do with these systems?

Beer: Well, because people have been living on social media platforms and using social media platforms for a significant amount of time, we’ve built up a series of biographical traces about our lives within those platforms.Now, instantly then, if you run a platform and you want to have maximized personalization, then having people’s biographical traces is a significant, valuable resource for knowing them. It’s a way of knowing people through those pasts, and that becomes a resource for maintaining engagement in the present.

Obviously things like nostalgia, memories are powerful things for people and they’re understanding themselves, but also their understandings of their friendships and relationships, and connections with other people, collective understanding of what’s going on. So this can generate significant activity, because what you are looking to do is maintain people in that platform, as long as possible, each day.

That’s the objective, because, really, some of these social media platforms, they can’t grow much bigger in terms of the number of users. What they’re looking to grow is the level of engagement that people have with the platform, particularly as they’re competing more with each other, I suppose, a little bit on this now as well with generational shifts and so on.

It doesn’t have to be a long time in the past, it can just be a previous year or whatever, you don’t have to be recalling long periods of time for it to work, but it gets people engaging with their own past and with shared moments that then recirculate and trigger activity in the present.

Therefore it fills the gaps. It fills the voids within social media for us to say something, or respond, or act within the platform—to be active. So it gives us an option, little bit like memes, I think, they become these anchor points for activity that allow people to fill the spaces of social media and satisfy the obligation for activity, I suppose.

Lloyd: Yeah, and it seems really effective. Part of the book is about, as you mentioned, the user response to the memory feature. Could you talk a little bit about what you found? You did focus groups and some structured interviews, right?

Beer: So this was off Ben Jacobsen’s project. He’d performed his interviews, and has been writing about those, and this became a kind of side project to that, really, around classification and ranking. It was a kind of unexpected insight that came off the back of that. We started to think about ranking and classification and then how people were responding to their past content being classified and ranked. And we found some, we use Imogen Tyler’s term “classificatory struggles” within the space. So it’s not like these things fit seamlessly in. They do generate activity and they do generate content and stickiness, but they also create other outcomes. We detail this a bit in the book, but these are things to do with misunderstandings of memory. So, presenting back things that weren’t that significant to the individual.

That was one of the examples, one of the things we looked at. We also look at the way that, sometimes, it can feel invasive. It is part of the surveillance, I suppose it’s that creepy over-surveillance you sometimes get from these platforms that unsettles you.

And in other instances, it was almost like this reaction against the polished, sleek version of their past that didn’t feel quite right. It didn’t sit well with people to have their past packaged in quite such a neat way. That was one of the other things that we found. So people were engaging and showed that they were entertained and amused by these memories, or they saw it as useful in some instances, but there were also these struggles, uncertainties, unsettling properties to it as well, sometimes, that people found.

Lloyd: It strikes me as paradoxical that people would find these too polished, because the conventional wisdom about what you put on social media is that it’s your vacation photos and the studio pics of your baby and things like that. And so the idea that in a year’s time it would come back to you as a memory and you’d find it too sleek.

Beer: People use those types of terms in their response to it, but maybe part of this is that an automated story of your life can create a kind of unsettling presence or it can clash with your own version of your own past, and therefore feels wrong. Or it might just be that people have an uneasy sense of automation within the space.

We use Walter Benjamin in the book, there’s an illustration fragment from Walter Benjamin about how memories gain authenticity through the digging, through the actual digging them up, unearthing the memory actively, is how they gain legitimacy. And here, maybe it’s the fact that because people aren’t digging it for themselves when they’re presented back, there’s a sense that they lack authenticity, or lack a kind of legitimacy.

We problematize notions of authenticity in the book, but you can see how it might be communicated as a sense of not liking the kind of polished nature of what’s presented to them. That’s part of their response, perhaps, to automation.

Lloyd: That’s really interesting.So, you point to this in the book a bit as a potential path for future research, but I was wondering, if you were to speculate on what effect this automated production of memory, what these algorithms are doing, both to the individual and maybe to social relations more broadly—what do you think will be the effect?

Beer: It will change what we remember and how and when, because the things we’re encountering from our own pasts are coming up through the devices. So what we remember and how and when we remember it is going to be filtered through this archival structure. I think that’s already happening, that’s in place.

I think that there’s the potential there for a reworking of what the notion of a memory is. What we understand to be a memory could change as a result of this, that it is something that’s in the platform, as well, and that’s automated, and that is provided to us. And the selection of what a memory is by this system could then lead us to see memory through that lens, potentially. So there’s a possibility for that there. And then, I think, the third thing is that this will have consequences for individual notions of self, potentially, and identity, but also collective remembering.

I’m not sure we understand fully what the implications are for collective memory, and therefore for solidarity, social connections, and social divisions that could come from a transformation in the collective memory, when memory is something that’s personalized to algorithms and is fragmented at the level of the individual, potentially.

So I think there’s three things there: when we remember, what we remember, what we understand the memory to be, but also how individual and collective memory might operate in the future, particularly as these things become more and more embedded, more active, and, potentially, more predictive.

Lloyd: This type of feature seems to be everywhere. I assume it’s in part because they’ve just found it so effective, a really effective way of increasing and engagement. But it’s on your phone, Spotify now sort of famously has this year-end feature where they feed you back a memory of the year in music. So it doesn’t seem like it’s going away anytime soon.

Beer: I don’t think so. The first thing I did on the memory thing was maybe about four or five years ago, and you can just see it escalating. Most of these platforms and devices have got their version of presenting your past to you, or the automatic production of our past.

That seems to me to be spreading out into these platforms and devices, and they’ve just got more and more biographical traces on which to use, but also now the accumulation of data about people’s engagement with those recirculated memories, which then feeds into the system itself. So they can then use that to try to be predictive about which types of memories work and what to rank as being the memory to send back to you.

The consequence of that are difficult to predict really, because it might be that that narrows down memory, or it might be that they find that they want to try to create ways of being unpredictable, because that’s what people are. So you don’t know, but it’s going to get coded into the algorithms.

Lloyd: So you and your co-author on this book, Ben Jacobsen, did a really deep dive into this [memory] feature that is a fairly significant part of, but in some ways tangential to, the overall structure of the platform and what they try to do, which is increase engagement. So I’m wondering, what do you think about what social media is doing overall, after having looked at this particular feature that seems to have all these complexities and tensions in it, potentially a manipulative approach, although not necessarily—but it seems like this particular feature is such a rich source of research and tension. How does it make you think about the larger platform or social media itself?

Beer: Yeah, you’re right. This project is part of trying to build up a bigger picture around the way that these systems work, and what their objectives are, and what the politics of platforms is, and data, and algorithms, and that kind of thing. And I think you can understand this in terms of the broader transformation that we’ve seen through social media as it builds up.

I did another book called The Data Gaze, which is about how this gaze is exercised on us, how we’re watched through platforms and by data. So I think you can see the memory thing in terms of the broader political economy of platforms, and particularly social media, which is about the data.

The data of the archived users is really where the value is in social media. That’s where the predictions are, because the idea is you can use the data to be more predictive about individuals, and, therefore, target content towards them in ways where value can be extracted. So I think you can see the memory thing in that broad term.

So what you want to do is keep people engaging with the platforms as much as possible, because that generates the maximum amount of data about those individuals, which then lends itself value. Now I’m saying they can use the data to achieve the things they say they can achieve, but it’s the notion that that data is of value. The ideas around value attached to data are the really important things in terms of understanding the activities of a number of these platforms, I think. So you can see the memories thing, I think, through the broader ideas around data capitalism, probably.

Lloyd: And engagement.

Beer: Engagement really equals data production and stickiness. These are things that create, that increase the amount of data gathered, and therefore maximize the opportunities for value to be generated—or for notions of value to be generated, at least.

Lloyd: What’s been your experience with this feature on social media?

Beer: Apart from Spotify, which I think—I did the calculation—I think I spent 3.64% of 2020 on Spotify(you can work it out from the hours it gives you), I don’t actually have any social media profiles. And a student asked me about this recently actually, and I said, “Well, the reason I don’t have any social media platforms is because I do research them. I’m some sort of social media enthusiast.”

So I did my first project on social media in about 2006, 2007, I was working on a program called the e-Society program funded by the ESRC with a colleague called Roger Burrows. It was called Web 2.0 then, and I created a Facebook profile, and I found it very unsettling. So I deleted that after we had done a little bit of research on it.

And then I did have a Twitter account for about six, seven years, I deleted that. And the only thing I’ve really stuck with is blogging, really. Yeah. Blogs and working on blogs, used Medium for a bit, been using Substack, just experimenting with those types of mediums as a way of writing and communicating and being part of an online community. But it’s not quite the same thing, is it?

So personally, I never get presented with any memories about my past. There’s a problem. It’s “How do you understand social media from the outside?” is something that I’m always working with because I teach this as well. I actually find it’s quite useful, because you can look across platforms, internationally, to try to understand it, rather than being led by a targeted, personalized experience of the social media space, I think. That’s the way I justify it to myself, anyway. I have a kind of discomfort with social media, but I can see the value in social media and understand people’s engagement with it, absolutely do.

I see my job is to try to think skeptically about what’s going on and to try to think in sociological terms about broader transformations.

Lloyd: Thinking skeptically about the world and ongoing transformations is also our goal here at Issues. I’m grateful you joined us for this episode, Dave.

If you’d like to read more about the new process of digital memory making, check out his book, called Social Media and the Automatic Production of Memory, and visit us at issues.org for more conversations and articles. And of course, you can read Dave’s review there. I’m Jason Lloyd, managing editor at Issues. Thank you for joining us for this episode of The Ongoing Transformation.