The following is a loosely edited transcript of the Author Meets Critics event devoted to Hatim Rahman’s Inside the Invisible Cage (University of California Press, 2024). The event was held on April 8, 2025 and sponsored by Work In Progress. The actual hour-long video will be posted soon.
Inside the Invisible Cage provides an in-depth account of “TalentFinder,” the pseudonymous platform that has become the dominant provider of on-line freelancer services in the world. The book stands as the most important analysis of the mechanisms that crowdworking platforms use to control the behavior of the highly skilled contractors and consultants they attract.
Panelists:
Hatim Rahman, Associate Professor and holder of the Pepsico Chair of International Management, the Kellogg School at Northwestern University.
Lindsey Cameron, Assistant Professor of Management and the Dorinda and Mark Winkelman Distinguished Faculty Scholar at the Wharton School, University of Pennsylvania.
Steven P. Vallas, Professor Emeritus, Department of Sociology and Anthropology, Northeastern University.
Vallas: Hatim, can you tell us a little about the origins of the book and what you were trying to accomplish?
Rahman: So the origins of the book. This is coming out of my dissertation work, and one of the things that kind of struck me, or when I was encouraged to write a book, is, as we know, that
in particular, one thing that I wanted to emphasize was about the impact eBay has had on almost all gig platforms out there or online platforms, and that they like kind of copied every online platform at one point or another and pasted the rating models or the model that they had, like a 5 star, without thinking much about it. And so I do think that has like huge implications, for, like just understanding how the history of how the platforms have evolved.
One of the chief architects of the algorithms that I described in the book, which made it more opaque, was actually in eBay higher up at one point or another. Right? So I think it’s always fascinating to see these connections that are there. So yeah, I mean, but broadly speaking. The last thing I’ll mention about this was that you know again, people who are studying (and on this call, doing amazing work on gig work broadly online work) weren’t paying as much attention to what I call this higher skilled online labor market. And I’ll just mention this last point, I think, a key distinction that animates the book, was that, I was fortunate from a study perspective to stumble upon a platform that was trying to create differentiation in their ratings.
Many online gig platforms, implicitly, or maybe very explicitly, want conformity. They want everyone to act in a certain way and behavior. There’s a very strong ethos underlying the change that I described where there’s a belief that there is a normal distribution of skills and that should be reflected in the ratings. So I thought that was a really fascinating, perhaps entry point.
Cameron: The first question I have is just to talk a little bit about your process, because this is different from the articles you’ve published. So just for some of us who are aspiring to write books, can you walk us through the experience?
Rahman: I did try to very consciously try to bring together data or arguments that are as best as possible, like not as much in in the articles, because there again, there are various ways to write books, and they all have their strengths and weaknesses. But you can see, I guess, in the 1st couple of chapters especially, there is a flavor of again like, let’s start with the history like, why do people come on onto the platform. How did it start? Who were the founders like? What were the founders thinking? And I don’t know again, if you caught this. But the founders wanted to convince their investors about they could create a model that would scale and attract investment. So like doing a what they call a hamster wheel staffing agency wouldn’t move the needle right? Their thinking was, we wanted to try to create something that was a global platform. So you know, things like that. As I said, I didn’t find space for it to do in articles and then put the emphasis on the organization right? A lot of it, and maybe in some of my other work does talk about like oh, it’s algorithm, and it is. It is there. But I hope I try to convey that ultimately, like organizations are the ones who are responsible. There are people behind this. It’s not just coming out of thin air.
Cameron: There’s been a lot of research about sort of like lower paid gig work. And then you’re in this space where you’re looking at more higher paid gig work. Can you talk about that empirically? That’s a distinction. But also theoretically, how do you think about these two contexts differently, because that’s always the trick. I think when you’re studying a topic that’s hot is like, how do you move behind the empirical uniqueness.
Rahman: That I do think theoretically, when you’re designing a platform, how people react like if the platform is aiming for differentiation. There was an iteration of the platform for many years when they had the Ebay five star rating the inputs, the outputs and processing are transparent. Everyone could understand what went into it more or less. And so at that point, like many other kind of lower paying platforms, 90% of the workers had close to 5 star ratings. Without notification to the workers, they switched the algorithm such that then only 5% of them had what they consider a top score.
If you think about, you know, we have when we design our courses right when we think about like how we design our grading systems like we think about like, Oh, yeah, do we care about like, if everyone gets an A, it’s a very different consideration when designing the course and in the relationship it forms with you and your students right. But now this is a context where people, as you know, their livelihoods are at stake.
So that’s one distinction. And the other thing that that is interesting is that these contractors are forming meaningful relationships with the clients. I do think that people’s relationships with workers were very similar, and that they viewed them as expendable or transactional. ‘Just get it done. Don’t ask as many questions.’ So even in this context, there. I guess what I’m saying is like the nature of the work was, it was one differentiating factor. And then the from the organizational standpoint, the purpose that they sought to create with their algorithms and evaluation…
Cameron: So where I thought you were going with, and one distinction I see theoretically between these 2 samples is lower paid work tends to have more of a pseudo relationship. If you’re familiar with Gutek’s use of the term, it’s repeated interactions with the same service organization, but a different provider, and I felt like with TalentFinder. It’s more of an actual or personal relationship. But then it seems that TalentFinder actually treated them more like commodities.
You talked about how the rating system changed and how the platform tried to create more differentiation. Can you talk about that relationship, and how it informed your analysis. What did it allow you to think about that you couldn’t have thought about if you didn’t have access.
Rahman: Yeah, for the platform. So you know, this is early on, like there has been great work charting how to do online ethnographies. It was very clear, early on that. Everyone wanted to know what was my relationship with the platform.
And fortunately, early on, I was doing exploratory analysis. It became very obvious that workers would change their relationship and responses to me, had I had a more formal relationship with the platform, given what many of us have talked about this triadic, tenuous, actually non-existent, real relationship, like at least, legally speaking, you know, just being a service provider. And I kind of talked about in the book. How like ‘I was cut off from the platform, even though even though it didn’t do anything wrong’, because I had inadvertently violated the Terms of Service. So as a worker, I was completely cut off, with no recourse, no ability to get back on whatsoever. So that was one thing. It turned out many of these platforms are providing very curated data sets and experiences to researchers. There have been articles in Science and Nature, where the editors had to release kind of editorial notes, talking about how like, later on, they found out the platform provided a very selected data set. So again, honestly, I didn’t think about it that way. Going in like that [without company authorization] wasn’t the primary reason why I did it. But as I went along, I realized that it was important to take that approach.
Cameron: I think particularly like related to the ratings and the changes you could see in the app was that from TalentFinder you were able to see that or more from the workers’ perspective.
Rahman: It was primarily through the workers’ perspective and kind of what I talked about in the methodological appendix, like I learn ways to try to track this more acutely. One is logging on, taking screenshots, scouring the forms, and then triangulating between workers themselves, just because, again, there were differences, right, like some changes were made to certain profiles, and others which was likely revealing of an experiment as workers tried to cope with shifts in the app.
Cameron: So I mean, I see your work sort of connecting to Edwards and Braverman, and sort of these big theories about control. But there are times where you don’t specifically like invoke them, you know. Think of as control, technical, bureaucratic or simple. So I’m just sort of, you know, this is the question I think about all the time like, this is control that you’re talking about and it’s algorithmically managed? So is it just a form of technical control? Is it a new form of control? Is it an amalgamation of all 3 like? How are you thinking about your work? Vis-a-vis sort of these big theories of control.
Rahman: Yeah, I mean, my orientation as a scholar in general is like nothing is completely new right, like there are elements of things that are new, but it definitely builds upon past iterations of control. I mean to me where I came back for what I think is new is one at the speed and scale at which these things are operating. And I kind of talk about that, and the nature of the employment relationship being very different than other forms of control that you might have in bureaucracies and other areas.
I don’t think it’s only a form of technical control. Like, what about the frameworks that they like the legal and institutional environment that they’re operating in. And the fact that there aren’t necessarily alternatives for these people because of the monopoly position of TalentFinder.
Cameron: I don’t know if I fully grasp or agreed with, because I thought about it more as like reputation lock-in, like I have a high platform or high score on like 5 or Upwork, and there’s no way I’m going to go to another pro another site because of the “cold start” problem. So how are you seeing the reputational “lock” on a specific platform? Now that you have Google, that’s supposedly, you know, searching and creating this composite reputation profile of you that creates interdependence.
Rahman: Some of the workers were specifically talking about how they didn’t realize that the ratings and reputations that they’re forming were feeding into these other aggregating platforms, right like on Google and Linkedin. People who wanted to leave, were like, ‘Hey, like, I don’t really care if they delete my profile type of attitude. But why is it still showing up?’
Why are clients or other people offline asking or telling me they first discovered me through Linkedins or Googles, or Facebooks, or whatever from that? I agree that that ultimately that the effect it creates a lock in right because it keeps them tethered to TalentFinder. But I was trying to highlight what I thought it was primarily some of the worker expressing that they’re like, ‘Oh, let me just try it out. Let me see what it’s like.’ And then when they get into it, it becomes like, ‘Oh, wow! Like, I had no idea that it’s not that easy to step away because it has implications on these other platforms.’ So that’s what I was trying to highlight with that.
Vallas: To review: I think one of the strengths of the book is that Hatim outlines the threefold mechanisms that this platform talent finder uses to control the freelancers that it attracts. One is the metrics, that kind of create unpredictability for the workers. They help the platform to achieve predictability, but they impose unpredictability on workers, especially when there are these volatile and opaque metrics that are used. A second mechanism is the terms of service that are remarkably expansive and actually open ended, so that the platform claims the right to change whatever it wants to change, and workers are obliged to conform to the terms of service. And the third mechanism is the one that Lindsay just mentioned, which is reputational interdependence. In their recent book, The Ordinal Society, Fourcade and Healy talk about the effect of reputational scores as a broadly important phenomenon that affects not only freelancers but all of us, affecting our access to services or insurance, the prices we pay for things and even the wait time on customer service phone lines. So Hatim is on to something important here.
But I do want to raise three questions about this book. I think many of us have studied the most vulnerable of gig workers, and not looked at equally important platforms that are reconfiguring the employment, the work situations of the middle class, right?
The book’s theme on the invisible cage is really in a sense updating Weber’s iron cage. So the question is: What type of domination is it that you’ve identified, Hatim? Weber’s bureaucratic rationality requires transparency, and a universal set of rules that are knowable to everyone, and in a sense, the rules or terms of service and metrics used by TalentFinder are opaque and uncertain and constantly changing. They’re different from a bureaucratic rationality. And I’m kind of wondering about that. It’s also the case that people don’t have recourse. There’s no right to appeal which would be fundamental to bureaucracy. And that reminds me of other competing typologies which perhaps begin to get at what you’re talking about, as when people use the term algorithmic despotism. Other people, like Piketty, have used the concept of patrimonial capitalism to capture the kind of arbitrary nature of the system.
So let me just raise two other questions too. I think the book is speaking about labor market uncertainty. Shifts in the metrics create uncertainty about whether or not freelancers will be promoted or visible to prospective clients. So in your analysis there’s uncertainty that’s baked into the experience that your freelancers have. So my question is whether you haven’t taken the opportunity to speak to the implications of your study for the literatures on precarity, like precarious employment. I think that the kinds of precarity that you’re talking about might have purchase on some of the experiences that self-employed or freelancing or entrepreneurs have, but they’ve always had. So if we extend this point about precarity, what are the experiences that you found on TalentFinder? Are they accelerating precarity that freelancers have previously experienced in the past or are they just reproducing them? What’s going on there?
My third point has to do with the workers responses to the conditions that they’ve experienced. So when freelancers are struggling or they lose a contract in the off line world, it’s not entirely clear who they might have to blame. But on TalentFinder, I’m wondering… If workers are attracted to the site and have a kind of a meritocratic individualism that they bring into the site, what happens when they experience this kind of arbitrary, unilateral, asymmetrical treatment? Do they respond with feelings of unfairness? Do they develop an injustice frame? What is the subjectivity that exposure to this kind of arbitrary rule provokes? Is it the case that whereas previously freelancers might have to blame thect, e market, which is kind of an abstraction, it’s faceless, but now, they blame the company? Does the existence of the app have a consequence for worker subjectivity? For people who haven’t read the book yet, there’s a very important chapter that has to do with how people respond, how they try and cope , how they desperately struggle to try innovation, innovation to improve their metrics. But your focus there, Hatim, is mainly behavioral, and I’m wondering about the kind of attitudinal and experiential effect, or the consciousness of workers when they encounter this kind of treatment.
Rahman: So yeah, I mean, what type of domination that’s interesting again. I do think that I just I haven’t thought about it as systematically as maybe I should. As many people are asking and push me to do so. You know, the elements that you listed were really interesting, right? Like, obviously, no recourse. There’s uncertainty and unpredictability right? I think that both of them are really key in my argumentation, because they’re similar. But I think the combination of those is really what kind of makes it even more related to the second comment that I’ll get to about the precarious nature of it. The reason why I kind of struggle with the question of what type of domination is I don’t know the right way to like, analyze it empirically.
To me it’s algorithmic. Despotism seems a little bit strong to me that that verbiage. It does depend very strongly on dependence. So that’s another way to say you can have, like whatever three or whatever freelancers on the platform who have very different experiences with it. Right? So someone who is like, Oh, this is this is weird. I’m just gonna log off right? Like I don’t really care. I don’t understand it. Which does, that does and can happen. You have some people who just by chance do very well, and are oblivious to how it works. But I mean still in terms of what type I guess it. There is this, like also, like algorithmic dependence to it, too, right into the domination. So I would say that the mechanism of domination is definitely a combination of the algorithms and the terms of service that you mentioned. I do think they lean very heavily on the teams of service to give them the institutional legal cover to be able to kind of implement these chains without giving people sufficient recourse or to file class actions.
Well, I guess I guess I’m trying to be more descriptive and not theoretical in in what I’m talking about right like so yes, like the terms of service, the opacity, the unpredictability, the lack of recourse.
I do think that we mentioned, that the type of employment relationship is distributed. When you log on as a worker you don’t see any other workers. You just see like a kind of a for you page for the types of jobs that you can apply to. There is definitely similarity in the precarity people feel to what he has written about in the literature… So, like in contract work, the self-employed, especially the type that Barley and others have written about. There is this notion that you’re highly skilled in your changing the form of employment. But the community, the network you’re operating in is very similar and familiar to what you’re aware of. Actually, there were people in my sample where, like told me, “I’ve been contracting for 20 years, and I joined the platform. It’s like bananas here right? Because it’s like the person who hired me is like a undergrad from Berkeley you know, who wants his assignment done, and he doesn’t know he actually doesn’t have any expertise in the actual programming.” So you you again, the degree of of precarity you face with them prospective clients, and they don’t necessarily know or care.
And then I talk about this like the experimentation on the platform in another paper, like, we have a fantastic quote from someone who’s like he’s like, ‘Oh, I’ve been working in organizations that do changes all the time, but nothing like where they’re experimenting with my wage or my availability right without my consent.’ And then I do think like one thing, that that in one of the chapters I talk about as well like one of the reasons I think people go towards contracting is they’re like no more performance evaluations right? Either your contract gets renewed or it doesn’t. And then, if it doesn’t, you move on right? It’s one of the reasons why some people in high skilled work go towards contracting. ‘Oh, I don’t have to deal with the office politics, I don’t have to deal with, you know, like pleasing the manager in terms of like getting the next promotion.’ Other things like that, whereas in this setting, you know, the platform is kind of this central repository, where they’re publicizing information, they’re experimenting without your knowledge. So again, I think there’s a difference in degree there as well. But you are correct to highlight that it does. It does relate to other types of precarity that workers face.
Yeah, worker responses. I don’t have enough data to to analyze this, but I do think when you first experience something weird in the platform. There’s like confusion. It’s like, wait like, what’s happening. Is it me? Could I improve? Because, you know, a lot of people do are like very self, motivated and interested. And like they’re like, what can I do to improve like? Maybe it was my fault. So there is a lot of self blame initially and they’re like, oh, maybe the client wasn’t comfortable expressing the problem to me. And you do see this moment where people are like, ‘Should I reach back out to the client’ and the client’s like, ‘I gave you good ratings like, I have no idea what’s going on.’ And then you sign on people are like, wait, there’s something more that’s going on with just me or the dyadic relationship that I have with my client and that kind of triggers more self-reflection and this discovery, especially again… when you experience suddent changes and you have to ask, ‘What are other people experiencing?’ and then going in the forums and seeing and conversing and experiencing. So I think there’s this progression over time for worker responses. Now, as far as I saw, and just to be clear, the progression is moving from self blame to like, ‘The platform is messed up, and they don’t care about us, and they’re not responsive to anything that we’re saying or experiencing.’ So that is the progression that that if I were to see from the data that that I see amongst individuals
I have not seen collective activism yet from when I cut off the data or collection 2019 2020. Maybe that’s changed. And I would love to see follow up work or studies on it if it’s possible. But in part, for some of the reasons that I mentioned, like people are afraid of getting cut off. There aren’t a lot of competitors on this scale. I haven’t seen any of it rise to collective action for some of the other reasons that I mentioned.