Ed Finn: Jennifer Golbeck is an Associate Professor in the College of Information Studies at the University of Maryland, College Park. She also directs UMD’s Human Computer Interaction Lab, and studies how people use social media, and thinks of ways to improve their interactions. Ian Bogost is Ivan Allen College Distinguished Chair in Media Studies, and Professor of Interactive Computing at the Georgia Institute of Technology. Founding partner at Persuasive Games, LLC, and a contributing editor at The Atlantic. So, welcome both of you.
Alright. So, our topic is “What should we know about algorithms?” What should we know about algorithms, Jen?
Jennifer Golbeck: You know, so I talk to people a lot about algorithms, and the ones that I work on as a computer scientist are building algorithms that can take the digital traces you leave behind, whether it’s from the Fitbit, especially social media. But any of these traces, and use them to find out secret things about you that you haven’t volunteered to share. Because all kinds of things about you come through in those patterns of behavior, especially when you take them in the context of hundreds of thousands, or millions of other people.
So when I go talk about this, the thing that I tell people is that I’m not worried about algorithms taking over humanity, because they kind of suck at a lot of things, right. And we’re really not that good at a lot of things we do. But there are things that we’re good at. And so the example that I like to give is Amazon recommender systems. You all run into this on Netflix or Amazon, where they recommend stuff to you. And those algorithms are actually very similar to a lot of the sophisticated artificial intelligence we see now. It’s the same underneath.
And if you think about it, most of the time the results are completely unsurprising, right? “You bought this Stephen King novel, here’s ten other Stephen King novels.” Sometimes they’re totally wrong, and you’re just like, why would you ever think to recommend that to me? And then sometimes we get this sort of serendipity that you mentioned, these great answers. And my favorite example is that I had bought The Zombie Survival Guide, which is exactly what the title suggests, like an outdoor survival guide but for zombies. And I read it very quickly, and the next day I go back and Amazon is like, “Oh, you know, since you bought The Zombie Survival Guide you might also like…” and it has other books by the same author, World War Z, which was made into a Brad Pitt movie which you maybe saw, some other zombie books, a couples zombie movies, and then this camping axe with a pull-out 13″ knife that’s in the handle? And I was like, “That’s exactly what I need.” The book was telling me this. And then I was like, okay probably not something that I need. But I bought it anyway. I thought it was just such a great example of like, I never would have gone looking for it, but it was such a cool thing to recommend.
And so, I think the thing to know about algorithms is that that’s generally what they do. They usually tell us stuff that’s not super surprising, or that we kinda could’ve figured out on our own, but sometimes they give us great insights, and sometimes they’re wrong. And just like you don’t watch, in order, everything that Netflix recommends, or buy, in order, everything that Amazon suggests that you should buy, the thing I think we really need to keep in mind with a lot of algorithms today is that they’re going to tell us stuff but we absolutely have to have intelligent humans taking that as one piece of input that they use to make decisions, and not just handing control over to the algorithms and let them make decisions on their own, because they’re going to be wrong a lot of the time, or they’re… They’re not going to do things as well as a human would do.
Finn: Ian, what do you think?
Ian Bogost: I’ve become really interested in the rhetorical register of this word, algorithm. How we use it And I did this piece for The Atlantic earlier this year called “The Cathedral of computation” in which I sort of said anytime you see the word “algorithm,” especially in print, in the media, if you try replacing it with “God” and ask if the sentence kind of works, it usually does. So there’s this anxiety we have, you know. “Google has tweaked its algorithm” or “What are the algorithms doing to us? How are they making decisions on our behalf?” and in what we are we sort of pledging fealty to these algorithms?
So there’s a sort of techno-theocratic register to the concept of the algorithm. And there’s this mystical notion about it, too. I think one of the reasons we love “algorithm” instead of “computation” or “software” is really we’re talking about software, is what we’re talking about. When we say algorithm, we invest this kind of Orientalist mysticism into fairly ordinary experiences and services and so forth.
And you know, this idea of the poetry of computation is interesting because I think it helps us kind of get under the skin of the rhetorical nature of the word algorithm, and not just the word but how we use it. When you think about that idea of the poetry of computation, it should kind of terrify you that okay, if we’re going to run our lives, our airplanes, and our automobiles, and our businesses on poetry, on these sort of poetic modes— It’s not because we distrust poetry, or because poetry isn’t good at what it does. It’s because what poetry does explicitly is to defamiliarize language. To take ordinary speech and to show us something about that speech. To reconfigure the words that we normally use in a different way.
And this aesthetics of the algorithm common in computer science of elegance, of simplicity, of tidiness, of order, of structure, rationalism, all of those sorts of features, are fantasies. To some extent, these are messy, disastrously complicated computational and non-computational systems. Like Amazon has a logistics system, and warehousing, and all these factory workers and warehouse workers they’re abusing and so forth. And all of that stuff, we’d like to kind of cover over it. But when we’re able to simplify it, to kind of point to this mystical God-like figure and say, “Oh, the algorithm is in charge,” then we feel better about that gesture.
So maybe one way of thinking about algorithm is as a kind of synecdoche, you know that rhetorical trope where you take a part and you use it to refer to the whole. So, we talk about Washington instead of the federal government. And when we do something like that, we kind of black box all this other stuff. And we pretend like we can point to Washington and that that sufficiently describes the way that the federal government does or does not function. Which of course it doesn’t do. It allows us to simplify the abstract.
So yeah, the technical aspects of algorithms, I think have become much less interesting, culturally speaking, than the rhetorical functions of algorithms. How we see this term and this concept weaving its way into our perceptions. Into the media, into ordinary people’s conceptions of the things that they do, and kind of— “Oh, Fitbit knows something about me, and so I’m going to use it.” I think those are somewhat underserved perspectives.
Finn: I think yeah, as we engage with these systems more, they become more and more important for everyday individuals, no longer sort of technical experts or somebody who’s designing an airplane or flying an airplane. And we’re all dependent on algorithms in many ways, now. And in many new ways that we weren’t even say, ten years ago.
I’m really interested in this notion of defamiliarization that you both brought up in different ways. In part this is about blackboxing things, and abstracting things. In part, it’s also about sort of the unintended consequences, you might say. You were talking about the digital traces that we leave online, which is a topic of great interest for me as well.
And one thing I think about is all the copies of ourselves, or the versions of ourselves, that are created. These profiles that are aggregated by different companies and then potentially sold. How little access we have to them. And so, is that something that you think about as well, Jen? Or do you think that there are— Is the conversation moving forward about that? Are people learning how to read these digital versions of ourselves more effectively? Or is this a morass we’re just sort of beginning to work through?
Golbeck: Yeah. I mean, I want to say that we’re getting more sophisticated about it. But then if you actually look at it, I’m not sure that we are. And there’s so many facets to this. But I guess a couple that I think are interesting.
One, I like to start with that Netflix/Amazon example because it’s a way that we’re all in interacting with this technology that, if we talk about it it sounds like these terrifying black boxes who maybe are so much smarter than us, and we don’t even know how to handle it. Except we totally do, because we use it on Amazon and Netflix all the time, right? And that’s exactly the same thing as the scary AI that Stephen Hawking says is going to ruin humanity, right? We actually know really well how to deal with it when it’s presented in that way.
On the other hand, if we look at the kind of virtual versions of ourselves, I think we can look at our own virtual versions and understand and process those. And when I talk about the kind of algorithms I make, I get a lot of pushback. Like, “Well you know, the version of myself that I have on Twitter, that’s a really professional version. That’s not how I actually am, and so maybe it’s not going to find the right things out about me.” And maybe that’s true and maybe not. Sometimes, depending.
Finn: Are you saying that the axe did not feature prominently in your Twitter persona?
Golbeck: Actually, you probably could totally find the axe, looking in my Twitter persona. I talk a lot about zombies online.
But you know, we can say that for ourselves, right? But then, if you look at how we treat other people online, these digital versions, and especially when people get themselves in trouble, the one bad thing that somebody does online becomes the entirety of that person, as we view them. And algorithms can see beyond that. But we as humans often can’t, where this person put out a tweet that seems racist. And then that person starts getting death threats, and gets fired from her job, and all of these bad things happen because the one bad thing that you did that gets shared widely and that there’s a record of becomes the representation of you as a person to the Internet.
And so we have all these digital traces, but it’s really hard for us as humans to process those. And as just one one more example, we’re doing a project now looking at people on Twitter who have admitted that they got a DUI. And we’re looking at what sorts of things they say, and can you check if they’re kind of changing their ways or whatever. As I had my students presenting this week, “Here’s the people we found who said they had DUIs. And here’s this guy who got a DUI.” And then the student was like, “Actually, he seems like a really good guy, you know. Here’s this stuff with the baseball team he volunteers for. And here’s these things with his kids.”
And I was like oh, we have to be morally ambiguous? Like, we can’t just hate him because he got a DUI and admitted it? Like, there’s all this other good stuff? And we’re so used to kind of seeing these digital traces and making our own inferences like oh, because this is there, that’s a bad person, or that’s a good person. And actually, we’re all very complicated people, and we all do bad things and good things. But we’re not great at judging it when we have a full record of people. And I think that that’s a problem that comes with all this, is that we don’t forget, and things don’t fade. Everything is there, and we have a hard time dealing with that. Algorithms can kind of deal with it a little bit better, or we can program them too. But as humans we have a hard time handling that.
Bogost: We also take computers to have access to truth in a way that we don’t take poetry to, for example. So to kind of come back to this poetry business, if the purpose of poetry is to defamiliarize words, then the purpose of algorithms is to defamiliarize computers. They show us how computers work, and they don’t work, kind of. Or they work badly, or they work in this very wonky, strange way. And you see it when you go to Amazon. You see that you ordered some button cell batteries because you needed two of them. And then it’s like, “Oh, perhaps you’d like these other button cell batteries.” And no, no, but I see what you’re doing. I see the caricature that you’ve built of me, and ha ha that’s inter—
But then we flip that on its head and we’re like oh, actually this is truth. Amazon knows something about me. Google knows something about me that’s true, and therefore I can know something about you by seeing the way that Twitter or Facebook or whatever is re-presenting you to me. Whereas we tend not to do that with poetry if you wrote— You know, here’s your book of high school poems. It’s like oh yeah that’s a sort of caricature of you at a particular mo—ha ha ha, we’ll look at that and then put it aside and understand that you as an individual are more than that set of words, right.
Golbeck: If I can give you a quick example on that, my dissertaon work was on computing trust between people online. So, if we didn’t know each other, could I guess how much I trust you? And I was presenting this—this is like 2004, 2005, so early in the social media space. And I was giving this talk like yeah, you know, we tell if our algorithms are good because you’ll say how much you trust me, and then I’ll compute it, and I’ll compare what the algorithm said to what you did.
And I would get these answers from these older computer scientists who were like, “Well, if the algorithm says on a scale of one to ten you should trust me to three, but you said a seven, maybe you’re wrong.” Like, the the algorithms says a three, so that’s probably right, as opposed to all of our personal history of interactions letting you make this very human judgment. Like oh, but the algorithm says three, so maybe you, human, are wrong.
Bogost: It’s super interesting to think, “Well, what does the computer think about me?” But not so interesting to think, “I absolutely trust the computer to make decisions about me.”
Finn: Yeah, I think that battle of trust is really interesting, and the ways in which we now— The space of human agency and the space of shared agency, where we’re sort of collaborating with computational systems. And then the space where we just sort of trust a computer to do something. Those are all moving around in really interesting ways.
For example, now I find myself questioning my frequently blindly obeying the instructions of Google directions about which way I should drive home. And then sometimes questioning my pathetic slavishness to this system that obviously doesn’t get it right all the time. And then pausing because of who I am, wondering to what extent I’m just a guinea pig for them to continue testing that this isn’t actually the fastest route, this is just that I’m in Test Group B, to see whether that road is a good road.
So, this poses a question I think also comes out of “The Cathedral of Computation,” Ian, that we need to learn how to— So, seeing is one metaphor. I also tend to think of it in terms of literacy and learning how to read these systems. So, how do we begin to read the cultural face of computation?
Bogost: Yeah. It’s a great question. It’s an important problem. So, the common answer, let’s start there, is this sort of “everyone learns to code” nonsense that’s been making the rounds? Which, it’s not—I mean, I call it nonsense just to set the stage, right. But, it’s not a bad idea. You know, why not? It seems like it’s reasonable to be exposed to how computers work, and to some extent you learn some music, you learn some computing. Great.
But really the reason to do that is not so that you can become a programmer, but so you can see how broken computers really are. And you put your hands on these monstrosities, and just like anything they don’t work the way you expect. There’s this library that’s out of date and some random person was updating it but now they’re not anymore. And it was interfacing with this system whose API…who knows how it works anymore?
And once you kind of see the messiness, the catastrophic messiness of actual working computer systems, then it’s not that you trust them less or that now we can unseat their revolt against humanity. Nothing like that, but rather it brings them down to earth again, you know. But in addition to that, the way that we talk about these systems, and the fact that we talk about them, that we talk about them more is also important. That moment with Amazon is a moment of literacy. It’s a moment of you as an ordinary person recognizing, “Okay, I see the way that Amazon is thinking that it has knowledge,” and then working with that, and thinking about it, and talking about it. That kind of literacy is just as, maybe even more important, because it’s right there on the surface, and we can read it.
And then I think there’s a third kind of literacy that’s important to culture, which is the way that we discuss these subjects in the media. It really does matter. And the more that we present the algorithm as this kind of god when we write about it, especially for a general audience, then the more we don’t do our jobs of explaining what’s really going on and how a particular subsystem of a computational element of a very very large organization that has all sorts of things happening, we do a disservice to the public in that respect.
Golbeck: I agree with everything you said. And I think this literacy of just being able to understand what we know and what we don’t is so critical. Because when I talk about this artificial intelligence that I do, it’s completely unsatisfying, whether I’m writing about it or if I’m talking to people to say you know, what we do is we took all this data, and we put it in this black box, and we basically have no idea what goes on in there. And it spits out the right answer, and we kinda know it will do that in predictable ways. But we can’t tell you what it’s doing on on the inside. We spent a couple decades researching that, and we can’t. That’s a completely not-exciting article.
So what we do is we say, “We put your stuff in this…box, and it may be a black box. And it spits out this answer, and look, here’s some stuff that we kind of computed intermediately that sounds like it’s some insights that make you feel like you’re getting a story.”
So, the example that I use most is we take your Facebook likes, and they put them in this black box, and it can predict how smart you are. And that’s not too satisfying. And so we say, “Yeah, and if you look at it, here’s the things that you like that are indicative of high intelligence. Liking science and thunderstorms and curly fries.”
And everyone goes, “Curly fries?”
And then when I talk about it—especially like, market researchers—people get really angry. “How can you know that’s going to be true? And it’s going to change.” And it’s like, I’m just telling you that for a story. We don’t use that. We don’t care about that. It’s not part of the computational picture, but it allows us to tell a story that makes it feel like there’s something human going on in there. And that is a struggle for me, because you want to tell this story, “Here’s what these algorithms do, and it’s unpredictable and crazy.” But you can’t tell a story with just like, “black box spits out answer.”
Bogost: Yeah, but we can reframe that story. I don’t know if this is the best example, but it’s a kind of information derivatives trading that you’re doing, right?
Bogost: Which, I mean, I don’t know that that’s the the way to talk to the everyperson about the example that you— But it doesn’t have to be reframed as computation, right. There are other touchpoints we have in the world, where like, you know how there’s infrastructure? There are all these highways, and you didn’t build them but they were here before you. There are certain computational systems that were there before us, and we come to them and we actually have no idea how they work. We literally have no idea. So, the work of explaining how computational systems work that doesn’t rely on this appeal to mysticism, I think is super important.
Finn: I think this question of storytelling is really important. Not only because this is all an elaborate ploy for me to do research on my book project about algorithms, but also because humans are storytelling animals. And storytelling is essentially of a process of exclusion, right. It’s selecting the telling example that may or may not represent the broader history, but you have to find the examples in order to tell a story because humans aren’t going to sit down and read the phone book, right? We’re not going to sit own and read the database.
And so my question is, how do we grapple with storytelling as…is storytelling a fundamentally different way of knowing than what we might think of as computational knowledge? You know, when you’re talking about…the computational approach is the process of inclusion, right. We want to include as much data as possible to make the data set as rich as possible so that the solution will be more complete. Is that a totally alien way of knowing? Are there ways to bridge that divide?
Golbeck: I mean, it’s so hard, right. For the computers, you absolutely want to give it everything. And then when you’re talking about what the computers do, generally when you’re working with this huge amount of data, which is the exciting thing now, you’re ending up with not logical insights but statistical insights. And any human can look at the connections that are formed and go, “That doesn’t make any sense to me except that it tends to work most of the time.” And so we want to tell a story that says here’s some statistical insights, and and let me tell you a few.
But that doesn’t really give a picture, and it’s hard to give a picture, of “here’s how statistics work,” and little patterns emerge as important from this big mass of data. It’s a story that I try to tell all the time. But people, I have found, latch onto the specific examples and have a hard time grasping the bigger thing. And I think in terms of computer literacy that that is so much more important than being able to program. Programming is great, and you will see what a mess it is. But being able to grasp that this is a statistical insight and the individual example doesn’t matter, that’s the thing that I would like to be able to do better.
Bogost: Yeah. I mean, computers are more like marionettes, or like table saws or something than they are like stories. They’re these machines that produce things. And you design this machine such that you can then design things for the machine. So you have your table saw, and you make a bunch of jigs so you can get the right cut. And you build this puppet, then you have to kind of manipulative it in this perverse way that you can’t really even explain, in order that it produces an effect that appears to give life to the creature.
It’s a different way of thinking in the sense that whether it’s a story, whether its an outcome, or a business result. Whatever it is that the particular computational system is doing, it’s not doing deliberately, and it’s not doing it in a singular way. It’s a system that’s been designed to produce many similar kinds of outcomes. And this is a kind of weird way of thinking about behaving in the world, especially since we ordinarily think in and talk in specifics. In stories, in examples, in individuals. And that’s also still how we write about everything, including computation.
And you see this when you see computational arts, and you see the aesthetics of computation if you look at Twitter bots or generative text, or any kind of generative art. You know, the results are terrible when compared with hand-crafted storytelling, or humor on Twitter, what have you. What’s remarkable about them is not their individual utterances or individual effects, but that there is some system producing many of them, and when you look at it holistically you can appreciate it in a different way. And kind of getting that aesthetic ability, right?
I mean, we talk about ethics a lot when it comes to industry and to computing. But we don’t talk about aesthetics enough. Like, one other way into this literacy problem is through aesthetics. Understanding how computers produce results on the artistic register, right. Even if we kind of hate those results, or we can’t recognize them as art, and saying, “Actually, something just like that is happening inside of Facebook or inside of Google.”
Finn: Yeah, I think that notion of aesthetics is really important because I think it’s one of the ways that we can confront very inhuman or very alien ideas and systems, methodologies, without necessarily having the language to articulate what it is, right. Aesthetics can be a non-verbal way of engaging with these questions.
So, I think there’s a connection between aesthetics and what you referred to as illusion before, as well. And so my question for you both now is, are the illusions necessary? Or we could talk about it as that kind of faith, and you know maybe it’s a bankrupt faith or a misplaced faith. But is that something we have to have? Is that the only way that humans are going to interact with these systems?
Bogost: No, it’s starting point. It’s the thing you do when you don’t have better options. And then you realize oh, this is insufficient. And this is a good starting point. And then you recognize also the intrinsic flaws of the illusion. And you seek more knowledge and deeper understanding. And then you realize this [has] sort of know been demysticized now.
And you can do this historically. Maybe that’s one concrete example of something we can do. Go back and and unpack any historical computing system, and see the bizarre reasons why it was constructed in the ways that it was. What it did. How it had an influence on later systems. Then you’re just, “Oh, okay. This is just like anything else.”
Finn: The Atari, for example.
Bogost: The Atari, for example. Yeah, I’ve written a book on the Atari that tries to do exactly this. So, computing history has a role to play here. And as a kind of very quick aside on that matter, computer science as a discipline is one of the most ahistoric that I know of. Just completely uninterested in history. It’s just barreling forward, right, making that last algorithm slightly more efficient so they can do something slightly different.
Golbeck: Yeah. I think you’re marionette example that you gave. I’ve never heard that example before, but I think it’s so spot on, and gets to all of these issues that we’re talking about. Because if you’re watching this marionette perform, that’s one thing that you can see, right. And then if we try to explain it, “Oh, if I pull this string, this thing happens,” we can have all of these debates about why does that thing happen? And why isn’t this thing? And can’t you do it this other way?
But that’s different than the thing that is being produced for you to look at. And which of those conversations do we want to have? Maybe both. But they’re two really different conversations. And I think that’s part of the struggle, that as a computer scientist I always want to talk about both. Look at this amazing thing that you can see that it’s doing. And then also here’s all these crazy things that make that work.
But it’s really two different stories, and I find it’s hard to say, “Here, you pull the string and this happens.” And people say, “But how do you get this big complex thing at the end?” And it’s just too complicated [crosstalk] to tell it all the way through.
Bogost: Because it’s a lot of strings
Golbeck: Yeah, there’s a lot of strings.
Finn: Yeah, I think that the sort of unanswerable question about whether it’s really a marionette unless you’re seeing that complexity at the end, right? And that’s the thing that you focus on. Which I think is about aesthetics and kind of notions of performance, or when an algorithm or a system becomes a cultural thing.
We just have a couple of minutes left. So, what would be some, just to sum up, a couple of practical things that you would suggest if somebody wants to actually understand algorithmic systems better?
Golbeck: Oh gosh, that’s so hard. So, coming back to a point that you raised before about algorithms as poetry or algorithms as beautiful things. I’ve absolutely had that thought, that I’ve looked at algorithms and I’ve gone, whoever wrote this had this new insight to the problem that I didn’t have. You can learn about algorithms without having to learn about computer science. And so I guess if someone wanted to do, that someone like, “I don’t really know anything about computer science. I just want to start getting in to see what that is,” that you might start with some kind of basic tutorials on the Turing Machines.
You mentioned Alan Turing at the beginning, and he kind of put forward this fundamental notion of all computer science that says you can have a piece of paper and basically a little pencil that can write a one or erase a one, and that can represent all computers everywhere. And you spend a lot of time as an undergraduate doing that. It can get very complicated, but it is an accessible concept. And I think if you spend a couple hours playing around with that and seeing how you can do actually sophisticated math and all kinds of interesting things, with this really simple machine, it starts to give you an insight into the process that we use to develop these much more sophisticated algorithms.
It won’t help you figure out all of the strings and the training that you need to manipulate those strings in the right way to get the picture, but it starts to help you see like okay, these algorithms, it’s not this mythical thing, it’s like a bunch of people who were beating on this really hard problem, who kind of manipulated into doing the thing. So I think as a starting place for learning how the algorithms work, it won’t get you into all the complex algorithms, but it gets you in the space of thinking about them in the right way.
Bogost: Yeah, I mean, computing history is what I think we’re both pointing at. If we’re living in this deeply computationally age where computers are inside of and running so much of our lives, maybe we should know where they came from.
Finn: Thank you both so much. That was great.
Golbeck: Thank you.
The Tyranny of Algorithms event page.