Welcome, everybody. So welcome to the first of our conversations on AI. It's been a somewhat long road getting here, at least in Web3 time, but here we are. And I'm so glad to be present with you at this dialogue between Reza Nagarastani and Nick Cernicek on the political economy of AI. In 1983, Elaine Rich defined AI as the study of how to make computers do things at which, at the moment, people are better. Yet, in much mainstream discussion of AI, the conception of it as an industry produced by human labour at many levels of skilledness tends to be obscured by a focus on the sublime of the feats of the machines, and it is a sector in which an enormous amount of labour is expended.
Many of the worst features of capitalism are on display. At the top of the AI industry, an overwhelming proportion of the means of production is in the possession of a handful of American and Chinese companies, and, to a lesser extent, universities. Working for them, we have an extremely small number of individuals with the requisite PhDs from elite institutions to make them attractive candidates to DeepMind, Baidu, or Carnegie Mellon. And as you'll probably see, the decision between the corporate world and academia is often not a terribly difficult choice for them to make. Skipping a couple of strata, at the bottom, we have poorly paid workers in the global south. who spend their days manually applying ever more complex labeling systems to images and other data sets the better to improve the abilities of the machines as well as those toiling in the
lithium mines of Bolivia the Congo Mongolia and indeed Nevada. Other conversations in our series will consider AI as interfaces with different types of intelligence with art making terraforming and, indeed, with the inhuman as much as with the human. But this conversation aims to look at AI through the lenses of labour and capital. We want to consider questions including the following. Does the capitalist development of AI narrow down the overall development of AI? Setting aside narrow AI, is artificial general intelligence possible under capitalism? And what does AI look like after capitalism? Our speakers Reza and Nick need a little introduction. Reza Nagarastani is a philosopher
and a former systems engineer. He originally became known for his pioneering work of theory fiction Cyclonopédia. Since the early 2000s he has contributed extensively to journals and anthologies and lectured at numerous international universities and institutes including Duke University, MIT, Ecole Normale Superiaire and Haus der Kulturen der Welt among others. his latest philosophical work is intelligence and spirit a book investigating the meaning of intelligence at the intersection of artificial intelligence philosophy of mind theory of computation and german idealism nick cernicek is a lecturer is a lecturer in digital economy at king's college london he's the author of platform capitalism inventing the future post-capitalism and a world without work and with helen hester he's currently finishing writing a
new book, After Work, The Fight for Free Time. The format for today is that we'll have a conversation purely between Reza and Nick for the first approximately hour and a half, then a break, then we'll reconvene to have half an hour or so of questions from the audience. The conversation will be transcribed and an edited version of it will form the basis of Nick and Reza's chapter in Conversations on AI. A couple of housekeeping notices. so first please mute your microphones during the first portion of the conversation I can see that most of you have done that already so thank you very much there will be time for questions at the end but please do keep them brief no more than 60 seconds and one question at a time to give our speakers a chance to answer questions from several people in some degree of depth if you want to ask a question
please put an emoji in the AI chat channel and I'll come to you also please could you mute the notification sounds from your other Discord servers again I'm sure most of you have done that already It's probably just me who had to spend about half an hour doing that yesterday with all my huge number of Discords. So, Reza and Nick, who's going to kick things off? I believe I'll kick things off. Yes, our Canadian friends should do it. I'll politely ask to go first. Yes. So, I mean, I'm quite looking forward to this conversation. I mean, obviously, I think Reza and I come at this issue of AI from quite different perspectives, but in many ways that I think are quite mutually compatible.
And I'm sort of interested to explore some of these areas of compatibility and potentially differences as well. My own research on AI has been focused a lot on the political economy of it. And I'm particularly interested in moving beyond what I see as a quite facile take on it by most economics research on AI. So if you look at what economics has said about the political economy of artificial intelligence, almost all of the work, I would say 80, 90 percent of it is focused on automation. So it's interesting how AI can automate different aspects of the economy, what jobs can potentially be automated. Is it cognitive labor? Is it physical labor? High-skill labor, low-skill labor, whatever.
And then on top of that, what is the impact of that potential automation? What has been less discussed by any research on economics and AI has been monopolization. So the ways in which this technology, in a sort of classic Marxist way, is all about the concentration and centralization of capital. So there is some research that looks at monopolization and the ways in which AI sort of facilitates that. But when you look at this research, what it's really interested in is businesses who are using AI. So think like, you know, it could be some random warehouse doing logistics.
And, you know, what is the impact of them adopting an AI system to help manage their logistics planning and things like that? And the idea here is that, you know, companies that adopt AI and use it in their business are going to become better than the others. They'll get more value. They'll capture more value. And eventually they'll grow larger and start to monopolize their markets. I'm much more interested in the big tech companies which is the providers of AI rather than the users of AI so these are the companies like Amazon, Google Facebook to a lesser degree, Microsoft Alibaba, Baidu a little bit these companies are obviously using AI but where the real power and wealth
comes from is in the provision of AI to other companies. Now, for a variety of reasons, I think a lot of this also tends towards a monopolization. You could, on your two hands, name the companies which are in the entire world actually participating meaningfully in this market. So it's only a handful of companies. And for economics, the research has largely been on the impact of data on this sort of concentration of AI power. So the argument from, say, the venture capitalist Kai-Fu Lee is that more data leads to better products, better services. That leads to more users.
More users leads to more data. And you start getting this virtuous cycle going on. So if you start to have a marginal advantage in AI provision, you'll get more data, you'll get more users. you'll get more data, more users, more data. And that leads you to sort of eventually monopolize this potential market. So the focus again here has just been on data as that driver of monopolization. I think what hasn't been acknowledged within the research so far, and it's starting to change a little bit, is the importance of hardware. The sheer fact that to train particularly these large language models that are emerging over the past two years, to train them is incredibly expensive. It requires a huge amount of computational power
and also computational knowledge. These things are becoming the real sort of ways in which Amazon, Google, Microsoft are pulling away from the rest. It's not just because they have lots of data. They do, but it's more and more important is that they have lots of computing power. So that hardware aspect is, I think, really crucial and has been missed by the research on the political economy of AI so far. So, I mean, a lot of my research then has been on that, the providers of AI and on the hardware, and the argument being that these things are driving this concentration of AI resources in the hands of a few companies. Now, why is that interesting?
Well, for a few reasons, I think. and I will flesh out I think a lot of these reasons as the conversation goes on but just very briefly I think there's three key reasons. One is first that having control over this key infrastructure of artificial intelligence gives these companies an immense capacity to capture value. And I say very specifically capture value rather than produce value. I think when we start talking about sort of traditional Marxist approaches to thinking about the way in which value is produced in a capitalist economy and circulates. I'm not entirely convinced what these companies are primarily doing is producing value so much as capturing value. And this partly has to do with the rise of a rentier economy. You own this infrastructure, you then rent it out,
and you capture value from other industries which are producing value. That's the first reason, this capture of value. The second reason why I think capitalist concentration of AI is interesting is to do with political power. So the control over this infrastructure and the political power that goes with it, whether it be, you know, most clearly, I think, in sort of the Chinese case, you know, various stories about social credit systems and whatnot, but also, you know, the very tight-knit ties between, you know, these Chinese capitalist platforms and the Chinese government, the ways in which these are having a sort of political impact. Similar things also exist in the U.S. as well.
You know, Amazon and Microsoft, for instance, had a massive battle. And as far as I know, actually, I think the battle is still going on because there's been an appeals process about it. But a massive battle to get a huge contract to provide cloud computing for effectively the military and the security state of the U.S. apparatus. so you know there's there's these increasingly you know fascinating and and and and sort of terrifying in many ways these ways in which the platform power provided by ai is being tied up with political power both directly and and indirectly so that's i think the second key reason why it's capitalist concentration of ai is is quite interesting to to think about and discuss
The third one, which I think may end up being a primary focus of the conversation today, is control over the direction of AI research. So I'm particularly interested in this because capitalism has a particular set of interests and capitalists as embodiments of capitalist accumulation process, again, have that sort of particular set of imperatives that they have to follow to survive as a capitalist company. And these imperatives, though, drive them to develop technology in particular ways, which aren't identical, let's say, even if they do at times parallel. But they aren't identical with the interest of, say, scientific research or, you know, with something like a sort of philosophical program set out by Reza in Intelligence and Spirit.
you know there's this project for instance of AGI I think potentially is impossible under capitalism and you know something we'll try and discuss today but I think the short term interests of capitalism the sort of settling for local optima over sort of greater goals these sort of things are well founded and well studied within the research about innovation under capitalism. And within AI we can see the same sort of thing. Deep learning takes off in 2012. This sort of unique conjuncture of ideas which had been around since the 1980s but suddenly finds a material embodiment
in big data and GPUs sort of brought about by video games. That unique conjuncture causes deep learning to suddenly become incredibly successful in a variety of narrow tasks. And what we've seen over the past 10 years now is a huge amount of investment in that particularly narrow field of AI. It's not to say there's no research or funding going elsewhere, but the vast majority of the research and funding is going towards this particular narrow idea of AI. And what's getting sidelined then is alternative approaches, which would probably be much more fruitful for any sort of development towards AGI. So my provocative sort of claim here, and this is maybe where I'll let Reza take it up,
but one of my provocative claims, I think, is that AGI is actually impossible under capitalism. Reza, I'll sort of pass it over to you, you know, if you want to lead off of that or if you want to talk about, you know, more from the sort of computational side of things, the various limits built into particular frameworks. Sure. Many thanks, Nick. I mean, I can top this off. I mean, you're way too articulate that I feel that I'm going to fail. Nevertheless, I will try to make an answer. I think best avenue to approach this is by way of
actually understanding what sort of research scenarios, what sort of biases, fundamentally theoretical and ideological biases, we are in in the AI. And that, I would say, that's not even considering the triad of labor, capital, and machine under capitalism, right? I want to start with this idea that what you might call to be research myopia of AI at this point is not particularly clear where its origin is
or where it is coming from. But we have some good ideas when we read the theses of people who are proponent of this narrow AI, big data, where they are coming from, actually. I would say that number one, at least in computer science And computation is not really AI, right? But nevertheless, it has affected the research on AI, precisely because of some of the bold claims that has been stated a long time ago, right?
I would say that the first one is what Anil Kavya, in his recent book Logisiel, called Turing Orthodoxy. What is Turing Orthodoxy? So Anil talks about the major point of this orthodoxy is that syntax manipulation is sufficient for basically doing jobs that could be otherwise not done. And we can do it blindly, right? By following certain procedures
in the computational sense. But then you always bring back some sort of semantic explanation of what you have done or what you are doing. So it is the case that the first... The horn of Turing orthodoxy concerns with bringing semantic from the back door, right? The semantic is not actually part of the Turing orthodoxy of computation. But you're always trying to bring it from the back door, saying what I am doing or what this thing is doing, right?
So there is a certain sort of what you might call to be semantic unconscious about computation in the classical sense. The second one is Turing's own claim that for any infinite jobs that we could do, We do not need infinite machines. But all infinite machines can be based on one simple machine, right? A simple machine of highly, you know, high sophistication,
but nevertheless in its procedures, when you look at it, it's highly simple, and that's Turing machine. so we see a certain sort of reduction already kind of in the computer science in the classical computer science from Church and Turing thesis onwards that's you know everything that we could do we could do it by way of a simple procedure a simple machine so these are the two horns of Turing orthodoxy, which I think have fundamentally narrowed down certain sorts of research that should have been done. And of course, there are research that try to
escape the ambit of this theory and orthodoxy in today's theoretical computer science, but they are highly niche. They are not funded precisely because they don't fit into the trend. So this is one, as I said, certain sorts of computational philosophy, flawed computational philosophy that is already there that has been responsible for that. But that is not all. Second one is the emergence of a new sort of empiricism. is what we might call to be,
we might call it the rise of black box socioeconomic systems. This empiricism purports that if we have sufficient data, and we are not really talking about the volume of data here, but rather how we read from it, right? We have a sufficient volume of data. We can access the meaning of the patterns. And these patterns can be universal, global. They can be applied to everything, right? It's the very foundation of big data.
So, this idea that's essentially the new empiricism tries to ultimately make scientific theorization redundant. that we could do everything with a sufficient volume of data in the same way that we could do with scientific theories. And scientific theories are becoming antiquated now, right? All we need at this point, and I highly recommend this article
by the proponent of big data Wolfgang Peach. It's P-I-T-S-C-H. Ten Theses About Big Data. So we could do the same. All we need essentially to do is to, instead of using a statistical, all the statistical models, using new inductive methods of explorations of the data. And that is sufficient for us to actually learn patterns. Science has always been trying to discover, right? And these patterns are bottom-up. You can apply to everything on higher scales.
so this second one would be the sort of as I said the constraining of research is sort of what you might call to be the new form of empiricism which purports certainty, predictability, inductive exploration, and most importantly, the blurring of theoretical exploration and simple, basically,
inductive manipulation of data. Right? The third one would be a certain sort of what you might call to be hard naturalism in the sense of sort of philosophical ideologies that go into AI research and constrain them. The third one is hard naturalism. A good example of it is Dennett, right? Dan Dennett. So Dan Dennett, if you read the last chapters of Darwin, The Dangerous Idea, he makes an example of why AI denialism is wrong.
his example is trying to explain what nature already does and we are just simply reflecting back upon what nature does so his example of nature is a blacksmith blacksmith who tries to make a sword the procedure of making a sword is a fundamentally algorithmic one In the sense that you get a chunk of metal, you put it in the furnace, it comes out hot, you start to hammer it, you know, constant hammering. Then iteration of hammering, and then you put it back in the cold water, heating again, hammering, heating, cooling, so on and so forth.
A certain sort of recursive procedure. Right. A recursive procedure through which a blade can be forged. Right. The molecules, the crystals of metal start to reshape through these tiny sort of movements. Right. and says that, look, if this is already what nature does, then there is absolutely no way for us to say that artificial intelligence is impossible precisely because it's what already nature does, right? Truly, simply through sheer brute recursive processes, right?
this was the third one the third sort of ideology that goes into this sort of valorization of this sort of research to the point of them becoming highly bloated claims the fourth one and I'm sure that there are far more factors but the fourth one I'm just talking about the main basically factors is what you might call to be the status of philosophy of computation today.
It is essentially a war between two highly flawed system of thinking philosophies about computation and the nature of it, and hence, by extension, the nature of intelligence. Not only intelligence in the parochial sense of processes, but also intelligence in the sense of general intelligence, right? A qualitative form of intelligence. This is the battlefield between computational functionalism and computational realism.
So, briefly speaking, computational functionalism claims that there is a one-to-one map between mental estates and computational estates. whereas computational realism is just what you might call to be a sort of computational panpsychism, right? Where computation is actually in the world itself. So in the sense of computational realism partially overlaps hard naturalism of the Danetian way.
So all of these combined, when we look at their presuppositions and the implications of the claim, then it wouldn't be too difficult for us to say that, well, this is why we, for example, think that every intelligent behavior can simply be turned into a real thing. recursive task, or for that matter, you know, all it takes to create AGI is by essentially what Dan Dennett has been talking about, is by enlarging the designer space of algorithmic
functions, algorithmic programs, because there is nothing else to general intelligence that simply a bundle of algorithms, right? Which is, these are, if we think about them coherently, we notice that there are, you know, there's nothing really stable here. It's not, nothing is set in stone theoretically, pragmatically, or epistemologically. So, yeah, so these are the sort of limitations and we should think about these limitations that are coming from different sources and they are essentially stifling the research on AI
by virtue of having extremely bold and unwarranted claims to begin with. Right? So I want to stop here and... I ask a different sort of question. And of course, I will get back to the AGI versus AI, whether AGI is possible under capitalism or not, probably later in the conversation with Nick. But first, I want to start with a different sort of scenario. I think that Nick is on a good track here
with regard to the political economy of AI and, you know, of course, its complicity with ideological, theoretical, epistemological biases. I want to ask, let's go back to, you know, as a thought experiment, go back to Marx. ask ourselves this question that the sort of machines that Marx witnessed were not the sort of machines you know, softwares, AI that we are witnessing today. Marx essentially witnessed mechanization and mechanical machines
you know, thermodynamic machines. so we can't assume that his thesis about machines can also be extended to a thesis about automation because the nature of these two processes mechanization and automation are fundamentally different the sort of machines under which they have been formulated are different So how can we, or if we can recuperate some of the Marxist or Marx thoughts about machines, about today's AI, given the fact that it witnessed something else and we are thinking about a different phenomenon.
And if we can do that, if we can extend Mark's thesis about mechanization to automation, what would be the link, the bridge for us to, you know, recuperate Mark's thesis about machines back then, from back then to now? what does allow us to revisit Marx again, given the fact that we are talking about different machines, different sort of processes here, mechanization versus automation. So, my dear friend, you have the microphone. Yeah, thank you for that.
I'd like to return to your stuff in a moment. But, yeah, to respond to your question, So I think, you know, on one level, this is partly a question of how to interpret Marx as, you know, a sort of fixed set of claims that are simply repeatable over time or more as a sort of, you know, perspective on the world that's adopted. And as the times change, your claims change as well. And for me, it's, you know, it's definitely the latter. But at the same time, what I take Marx to be doing is describing this particular socioeconomic system that we live under, describing its emergence in his time, and looking at how it, in particular, creates new sets of imperatives on individuals.
and these individuals can be workers or they can be capitalists but the key sort of point is that there's this emergence of a new structure which then drives certain behaviours from people given their position within this social system and insofar as we still live under capitalism and against McKenzie Wark or anybody else sort of arguing against the opposite I do think we're still living under capitalism I think that that sort of broad analysis about the structure determining imperatives and incentives and constraints and opportunities on individuals, not determining everything, still, you know, individual agency, but that broad structure, structuring the possible behavior is still applicable.
So on one level, you can simply sort of take that quite abstract framework and think about, as I've been trying to do, if capitalists are driven to capture as much value as possible for themselves, what does that then mean for their behaviors today? and this partly goes back to my issue around the rentier economy because when you look to what mainstream economists are interested in sort of big economic problems right now one that keeps percolating underneath a lot of their thinking is capitalist stagnation the fact that from the 1970s onwards the growth rate has been in
decline in many advanced capitalist countries growth has basically stagnated since 2008. Here in the UK, for instance, the classic measure of growth, GDP per capita, has actually reversed since 2007. So there's this real period of lack of growth in the aggregate economy. And what I find so interesting about thinking about these major platform companies and the rise of this sort of, you know, AI as a service, you know, AI infrastructure as a service, is that they're simply building these things and then capturing value from others.
And there's this move from under conditions of capital stagnation, a move from the production of value, a real focus on the production of surplus value, to a move to simply trying to capture as much value as possible. And I think that's a quite interesting shift from the industrial economy that Marx was talking about. I think the conditions today of capitalist stagnation mean that increasingly companies aren't interested in producing value so much as simply trying to grab as much of it as possible. maybe let me concretize this in one way as well which is advertising as sort of the classic example i mean advertising doesn't produce value it helps realize value that's already been produced
so the company produces you know product and then it uses advertising to try and sell as many of those products as possible so those products have been built and then you get advertising to realize the value within them. But then, of course, you pay to get that advertising done. And the advertisers themselves are not producing value. They're helping realize value, but not produce new value. So what they're effectively doing is capturing existing sources of value. And I think in many ways, that's what the AI companies are doing today. And so, again, to return to your initial question, but what does Marx have to say about all this stuff? Well, those broad frameworks of capitalists searching for as much value as possible, the competition that they're driven into, I think it has changed with the technology, but it's also changed with that basic macroeconomic condition of stagnation.
And AI has become the most recent way and most profitable way to do that sort of thing. on the level of mechanization and stuff, I think, yeah, it's, it's, it's interesting because yeah, mechanization, you can think about it as almost a sort of deterministic system, um, which is linear, uh, which is, uh, you know, a linear set of causes and which in many cases, at least initially was, was primarily based upon sort of, uh, replicating human behavior rather than, you know, trying to sort of reorganize the production process in completely inhuman ways. You know, eventually things like the assembly line become this massive breakdown of the division of labor.
And they start producing, you know, products in ways which were never simply possible beforehand. So you start getting machines built for these sort of purpose-built divisions of labor. I think with AI, and you get this not just with AI, but earlier with sort of cybernetic systems, but the rise of various feedback loops breaks up that sort of traditional linear causality of mechanization. And you start to get feedback loops which are responsive to workers in the production process, but also responsive to other machines in the process. and trying to keep the system
within a particular range of optimal functioning. So you get some of that. And then with AI, yeah, I mean, it's... With what we have today, I'm not the narrow AI that we have today. And AI is something completely different. But with the narrow AI that we have today, it's in many cases largely about a sort of predictive aspect. So it's not, you know, it's not this cybernetic feedback. It's not this linear mechanical sort of system of functions. Instead, it is a predictive quality.
So a simple example is, you know, you watch a video on YouTube and you're recommended a series of videos at the end. And that's, you know, a particular piece of narrow AI, which is predicting what they think you, as a particular individual viewer, are likely to want to watch next in order to keep you on the platform. But again, I think that indexes a shift within the basic technologies that capitalism is using. Marx himself in Capital talks about various distinctions between tools and large-scale machinery. And I think today we've seen the continued further development of that.
That's, I think, one way to extrapolate the Marxist analysis to the approach of AI today. today. Reza, I don't know if you have anything you want to sort of jump in on on that. Yes, I have many thoughts. I mean, we can obviously, this is for our dear audience, but we can obviously, you know, kind of reticulate and iron out some of these during the writing process of our chapter. Two things that came to my mind. One, I think that missing link that I was talking about is essentially the very spirit of capitalism itself, right? Which is, in fact, the very spirit of the machine itself, recursive procedures.
What's, you know, that's essentially capital survival is about optimization and efficiency, accelerating efficiency of applying its own produced values and concrete technologies back to itself, such that it can expand the socio-technical domain of its powers, but under the conditions of creating surplus value, right? So the process of valorization, which is essentially a recursion, right? Applying information back to itself, applying value back to itself
such that it can actually expand is ultimately what aligns capitalism with machines. But now, if that is the case, then, should we ask, are machines always, all the time, on the side of capitalism? Or rather, capitalism is always on the side of machines. Precisely because of its Asian nature. I mean, I think capitalism is always on the side of machines. I mean, this is the classic thesis of the rising organic composition of capital. Yes.
the basic idea that technology is constantly replacing humans within the production process, as you say, for optimization reasons. But interestingly, I want to also sort of problematize that a bit because I think any given... Capital as an abstract structure has... My apologies. Can I actually... One more question so you can actually cover this one. So essentially, it is on the side of optimization, which brings us back to that example that Dennett was making. So essentially, capitalism ultimately wants to justify its existence as a new nature.
Like capitalism is being made in the image of nature itself, and hence you cannot escape it. yeah yeah and and even uh i mean even more than that you know capitalism is is remaking nature and it's in its own perspective um i mean i think yeah just on this this sort of optimization issue i think there's there's different levels of abstraction that's sort of important to to keep in mind at the most general most abstract level of of the capitalist system um it is it is a recursive process, you know, C or M, C, M prime. You know, it's this accumulation of value with, you know,
different elements partaking as M or C or M prime. And that is, you know, that is all about optimization of that process of speeding it up, of maximizing the extraction of value, of accumulating endlessly and continuously without any end to that process. but in practice and this is at a different level of abstraction any human capitalist for instance has a lot of different different imperatives placed on them that aren't always that aren't always compatible with that overall capitalist accumulation process so I think one of the interesting examples from the empirical research on capitalist
adoption of technologies is that in many cases is they actually don't choose the most optimized technology. And there's classic work from David Noble on this, a book called Forces of Production, which looked at the machine tooling process in 1950s America and looked at the various technologies which were available at this particular point in time to optimize this process, to get large amounts of volume of the product, but also to be able to do so with the requisite specificity of the details needed. And there was basically two, three options. And capitalism didn't, or the capitalists I should say, didn't choose the most efficient option.
They chose the option instead that separated the knowledge of the production process and put it in the hands of managers and left instead this sort of de-skilled process with the workers on the factory floor. And so I think this is an interesting aspect because you get the optimization imperative, yes, but then you also get the control over workers imperative, which is at times at odds for individual capitalists. And so I just, yeah, I find it interesting because there is too easily a tendency to just say, well, capital wants something or capital demands something or organizes something in such a way, without sufficient sort of attention to the complexities of the lived personifications of capital, if we could.
Yes, yes, I completely agree. I mean, but that is precisely because capitalism ultimately wants to be a mimicry of nature in the sense of Roger Caillois, mimicry that a twig looks like a predator and a predator looks like a twig, right? In the sense that nature doesn't have an end, right? So capitalism should have an end either. But I completely agree with you that there is absolutely a certain sort of disjoint among different sort of capitalist or capitalist paradigms. They don't always go for the optimization paradigm, right?
But this is precisely the anarchy of capital itself, right? So capitalism ultimately, as part of its interest, should create a certain sort of rabid competition among these disjointed capitalists to create essentially that image, whether it is fake or not, of tendency toward optimization, right? Even if, you know, not all capitalist systems work towards that end. Yes, chaos, not anarchy. Anarchy by that, I mean it's anarchy in the old sense of, you know, the Greek sense of, not anarchy, but of chaos, yes.
Someone corrected me here. Yes, absolutely. So coming back to this, the project of valorization and something that I always, you know, I think that our audience would like to hear about this. So we know that capitalism, or at least capitalism in the sense of competition between different individual capitals, strive to shorten labor time.
but also by positing duration of labor as an ultimate value. So I know you and Alex have talked much about this, about shortening of time of labor. my question is that to what extent shortening of labor time and this is of course coming back to the operismo and the whole even has been already talked about in Smith and Ricardo's
views about machines, labor and time to what extent shortening of labor time is an emancipatory? At what scale is it emancipatory? Considering the fact that capitalism does strive to shorten labor time, yet the ultimate trick of capitalism, as we have seen with automation and AI, it does actually create surplus value by exchanging one type of time of labor
with another time of labor, right? So it is not all about shortening of time, but exchanging durations of different sorts of labors with one another, right? One time with another time. And that's how it actually creates surplus value. So in that case, if this is the ultimate trick of capitalism, exchange of one time with another to extract surplus value, then under such conditions is it warranted
is it guaranteed that shortening of labor time can be fundamentally emancipatory project or can it be just in the service of a new capitalism a capitalism that now works with full automation yeah yeah so i mean there's a couple of things i want to pick up on this because i think you know alex and i very much um i would say follow in what was until maybe 20 years ago the quite orthodox marxist line of uh you know the development of the productive forces is part of
the necessary conditions for communism, post-capitalism, whatever you want to call it. That was a necessary condition for this beyond capitalism to emerge. And that was, you know, the orthodox line. It was why, you know, Soviet industrialization was such a key thing for that project. And then it sort of got rejected out of hand over the past 20 years, you know, for a variety of reasons. that sort of claim of the necessity of developing their productive forces. In part because I think the left in general has turned very critical of technology in a way that is sort of at odds with its long history.
And it's still today, I find it quite difficult to find many leftist takes that are sort of optimistic about potentials of technology. Why do you think that NIEF has really left this sort of fidelity to techniques, techne or technology? What do you think are the sources of this sort of paranoia? And it is a sort of paranoia. I mean, for me, the most obvious one is the collapse of the Soviet Union as any sort of counter hegemonic project. for all the many, many flaws of the Soviet Union. It did at least continue to keep this space open, this ideological space open.
But the dominance of capitalism after the collapse of the USSR and the sort of long 90s moment up until about 2007 and 2008, it meant that the technology that was being developed was just capitalist technology through and through. I think on top of that, you have the sort of progressive moment, briefly, of the internet and the worldwide web in the 90s that just becomes rapidly commercialized in the late 90s, early 2000s, and becomes entirely, thoroughly capitalist. So, you know, the sort of maybe one hope for escaping capitalist technologies is rapidly recuperated. and then from that point you have things like big data which from a critical perspective and
i think it's entirely right is is simply a a massive surveillance economy um you know it's it's i think under those conditions it is hard to be uh optimistic about the the potential aspects of technology um particularly when the levers of power all seem massively dominated by you know capitalists and their cronies. So there's not a lot of optimism. I think for me, 2007 and 2008 changed that. The capitalist end of history wasn't final. Other alternatives are and should be possible. So yes,
I think there's been more optimistic takes from the left on technology recently, but again, it's still in my experience a really minor position yes position I'm trying to think unfortunately I'm trying to think of what your original question was which sparked a lot of this it was quite a good one my mind has been flying trying to shoot you with more questions I forgot Ross do you know what was the original question trying to remember myself but was it to do with the emancipatory yes the emancipatory balance yeah so development of the productive forces
and one of the key reasons is simply the production and escape from want which Marx in the German ideology talks very explicitly about this you need to develop technology in the productive forces in order to be able to builds a sufficient amount that, you know, we don't have to fight over, you know, food, housing, education, all these basics of life. The other aspect, of course, is then that reduction in labor time. And, you know, this is where Marx himself is somewhat, not contradictory, I think he's working towards, working at attention. On the one hand, and this is the fragment of machines, which has been mentioned in the chat, in the fragment of machines,
it marks it as most optimistic about technology and basically argues that, yes, capitalist development of the productive forces leads to the reduction of abstract labor time. It leads to a growing wealth of free time. And he says that at some point, the social relations that are just becoming incompatible with a society that measures value through labor time it eventually just breaks free of that and we start to measure wealth not by concrete or abstract labor time that's been expended but instead by the amount of free time that we all have so that's Marx at his most optimistic and then in Capital Volume 1 he's at his most pessimistic
which is the development of productive forces leads to an increasingly gigantic mega machine that is increasingly automated, running on its own, and workers are just... What was that? The mechanical monster, as he said. Yes, exactly. And workers are just appendages to this massive machine. And I think Nick Land takes off from this point and basically just says, well, it doesn't need appendages, we'll just replace those. So humans are completely chucked out of the system entirely. Yes, yes. but you see you can see where all these sorts of paths come from and i don't think there's any necessarily easy answer to it ultimately i think no ultimately i think it's a matter of political
struggle um you know it isn't it isn't answerable within uh an abstract economic function like uh you know mcm prime it is something which it has to be struggled over the other thing i quickly want to mention that i'll pass it back to you reza is um i think it's an important point to say that capitalism doesn't reduce labor time in general. The best way, I think, to think about it is Marx... What is that? It strives to, but it doesn't. Well, it strives to, but there's an even more nuanced point here, which is that the division of the working day into a period of necessary labor, which is necessary
for reproducing the worker, and a period of surplus labor, which is the value and the profit that the capitalist captures. The point of automation is not to reduce that aggregate of necessary and surplus labor, but instead just to reduce the necessary part, which means that actually the surplus part grows for the capitalist and the worker experiences it as a continued eight-hour workday or 10-hour workday or whatever. So the worker never actually sees the benefits of that. this reduction of labour time yes but the reduction of necessary labour time in particular so that then becomes the point of political struggle it's not a natural or inevitable outcome of the process but it's the political struggle
over which aspect is being reduced I see I see I mean I'm not accusing you of saying this in the vein of Smith and Ricardo but Adam Smith actually and Ricardo also on his fragments on machines has a similar sort of vein I can actually read something by Adam Smith written in 1776, the workmen desire to get as much, the masters to give as little as possible. The
former are disposed to combine in order to raise, the latter in order to lower the wages of labor. It is not, however, difficult to foresee which of the two parties must, upon all ordinary occasions, have the advantage in the dispute and force the other into a compliance with their terms. And this is actually part of, you know, his paragraph on the rise of the machines. And Ricardo takes it to a different level. And by the way, that passage was from the Wealth of Nations. Something that struck me here.
Let me think for a second, put my thoughts together. So, as I mentioned, to me, it seems as if that's mere reduction in time of labor is not a fully collective imaginary project precisely because as we know it, so long as the conditions
of producing surplus value which pertains to exchanging one time of labor with another are intact left unexamined and unchallenged, it doesn't guarantee any more, any real concrete significant change. It does, however, change the life of individuals, and that's also important, right? But that would be a certain sort of reformation, so to speak. It wouldn't be revolutionary in any sort of sense. soft or hard. Another thing that I noticed that
you mentioned that you have a political economy stance on this. I'm going to be a little bit Marxian here, which I'm not really, but only to bait you, to differentiate your position. to hone it out. So a classical political economic position asks us how capitalism works, whereas a Marxist position asks us why is it or why it is that capitalism works
the way it does and not otherwise. right and there is a gap between these two sort of questions, between the classical political economy and that way of Marx asking why is it that capitalism works the way it does rather than just how capitalism works at this point after writing you know the manifesto platform capitalism, inventing future, so on and so forth. How can you distinguish your position at this point? Are you more comfortable to be a political economist
or a hybrid of Marx and political economists where it comes to the question of capitalism? if i can if i can just quickly uh jump in there as well um i wanted to to to call us back to the four constraints that you mentioned at the the beginning of your segment uh reza near the beginning of the the conversation um i guess what i wanted to do ask both of you is to what extent you feel those constraints emerge specifically from capitalism working the way it does? Or not, I guess. At the risk of asking a slightly naive question.
Are these things emerging from capitalism or aren't they? It's a naive question. I'm going to make it very, very rudimentary, highly problematic and controversial you know, allegory for this. You know, it only takes time after so many steps that things get entrenched, get rooted, right? Capitalism is the organon of the entrenchment. And that's why it tries to copy nature itself, right? just like think about Nuremberg trial
when Ludwig Keitel the head of the post after Hitler committed suicide is basically gets the death sentence and he says that look, we were just following the commands, right? It wasn't anything significant But now I understand that there's so many tiny steps that I have taken, and this nation has taken, has created a system that was fundamentally pathological, right? You can say the same thing about capitalism.
Of course, capitalism is not Nazism, right? But it is really about this. It's about tiniest steps, incremental steps, taken, added to one another, constantly updating one another, right? Applied to one another. and it creates a certain sort of entrenchment from which you cannot escape. It creates a sort of naturalistic ideology. Right? Just, well, if evolution takes this amount of time, then probably we can't actually change the rules
of how we have evolved. But I would say that communist thesis is a completely different sort of stuff. It tries to show the unconscious of capitalism, unconscious in the precise Freudian sense, not as the lack of information, but as a consciousness that appears to be working against consciousness itself, right? As a preventing measure. And in that sort of environment, as I said, you always carry a certain sort of pathogen,
certain sort of disease where you can't have a difference between nature and capital. right you can't understand can't imagine whether other worlds could be possible and Marx I think in the fragments of machine tries to actually do that to show that a different world can be possible with a triad of labor, time, and machine right, and capital it is actually quite interesting that this text was not published in Marx's lifetime. And there is absolutely no conclusive,
basically, commentary that whether Marx really meant it as something that will happen eventually in the age of capitalism, or whether it was a thought experiment or a hypothesis. Well, Christian Fuchs has written about this quite carefully and he says that actually it was more like a counterfactual thought experiment. That we are going to think about how we can create or suspend some rules in this world of ours, capitalist world of ours,
such that we can think differently from the possibility, we can think about the possibility of another world, right? In the sense that it is obvious, I genuinely agree with Fuchs here, that fragments and machines is not actually about capitalism. It is actually a thought experiment in the sense that it tries to say that the machines will betray capitalism ultimately, but not under the conditions of production of surplus value, namely the conditions of capitalism itself.
Rather, once capitalism, the condition of its production and social relations reformatting, are being suspended, yes, the machines will betray any traces of capitalism. And that becomes a task of communism itself. Nick, there's a lot for you to respond to there Yeah Also I read as a simpler question Are you still a political economist? I think that's what he said Let me put it this way Because I think actually the answer to that question
What was that, sorry? I said that it was no insult I just genuinely was interested how much you are actually wants to go back and forth between classical political economy and Marxian economy. Yeah, so for me, the study of the capitalist economy sort of gets to the point you just mentioned about, you know, Fuchs' sort of analysis about, you know, suspending certain aspects of the capitalist system and thinking about what might emerge, machines turning against its capitalist masters, which might make for a nice movie actually now that I think about it but yeah so I mean for me the analysis of the capitalist system as it exists today and this is
where inventing the future sort of jumps off from and it's also where platform capitalism comes from as well is this the analysis of the contemporary conditions of capitalism are crucial for understanding the strategic possibilities for building anything beyond capitalism, for understanding weak points of the system, for understanding crisis points, for understanding where class formations might be emerging, for understanding just what is possible under our current circumstances. So I think it's Marxist political economy in the sense that he wants to know how the system works,
not in order to make money off the stock market or anything like that, but instead to understand where its weak points are. And crucially, I think this is a really important thing which gets forgotten by most of 20th century communist thought and practice, is that his analysis is also about what communism would have to be as a suspension of key elements of the capitalist system, and particularly value production. I know this goes to the heart of the problems of Soviet Union is that it never suspended value production. It was value production guided not by capitalists,
but by state bureaucrats, eventually in not the aims necessarily of capitalist accumulation, but in the aims of geopolitical competition. and value production was still crucial there. It wasn't a suspension of this core axiom of the capitalist system. And I think that is a lot of what Marx is writing about when he tries to write about capitalism is that he wants to understand the system so he can understand what it really means to get out of it. And things like critique of the Gotham program and stuff like that are also aspects of that. He's trying to say that, no, you think this is going to work and get inside of capitalism. it's not going to do that. So there's all that to sort of play into. And then, yeah, the contextual sort of aspects
of the current conjuncture. Because capitalism is understandable at the most abstract levels in quite simple ways, and we can sort of derive macro tendencies from these basic claims about its most abstract functioning. But then in practice, there's a whole lot of other complexities which mean that things play out in unique and historically contingent ways. And part of the task of me as a critical political economist or whatever you want to call it, I don't really care, is to map out these aleatory encounters and think about how they're transforming control over workers, potentials for worker autonomy, potentials for instantiating certain communist values,
All these sorts of things are, you know, they can't be, you couldn't predict deep learning from Marxist period, for instance. You know, you needed that sort of aleatory encounter in the 2000s and 2010s that brought these things together. You know, GPUs, big data, and things like backpropagation. all of this stuff had to come together in a sort of contingent circumstance to then become the form of AI as we know it today and this goes back you know it go ahead no no please please go on my apologies okay very very quick point just that this this goes back to I think the one of the big themes of the conversation which is that you know these aleatory encounters that produce the particular form of AI as we know it today deep learning capital then just simply jumps on it
and drives all research and investment and attention towards this very narrow area. And there's no real sense of possibly expanding and diversifying into different forms of AI. It's just all deep learning and how do you apply it and how do you make it run and optimize it better. So yeah, I'll leave it at that. That seems like a good moment to segue to the, so we're in our kind of final 15, 20 minutes of this segment. And it seems like maybe that's a good moment to segue on to... Ross, can I say one thing? Of course, of course. Go for it, yeah. For our dear audience to reply and think about. I mean, this is one of the problems that I have.
I think that, yes, during the Industrial Revolution, it was a certain sort of history of contingency, of complicities, so to speak, in the Benedict Singleton sort of way of gyropolitics, right? But I think with self-reflection of capitalism unto itself, and capitalism has, in fact, created its own learning biases, its own learning, teaching, application biases, such that the aleatory encounters become ever more or less, right?
You can actually think about this at the four reasons, the four main reasons of narrowing of AI research that I mentioned at the beginning. are not actually aleatory encounters. They are essentially, if we think about them, we see that they are fundamentally made in a certain sort of technoscientific form of thinking within capitalism itself that feeds capitalism again and again and again. so the aleatory encounters I'm not sure about it
whether it is really aleatory encounters or just capitalism tries to pretend as if they were aleatory encounters but they are not to me essentially I was going to ask actually to ask you both to respond to the question what would a non-capitalist AI look like we've got about another 10 minutes, 15 minutes of you two speaking just you two then we'll have a break and then we'll have some questions after that let me just finish this and then we'll have a break come fresh we take for another 15 minutes and then open it up for 30 minutes, 40 minutes, whatever
it takes another thing is that you mentioned that there is this sort of kind of contrast between the state bureaucrats and the sort of capitalism in a Marxian sense. But I think that the contrast is being dissolved, is being blurred at this point, precisely due to automation. So that you can't actually, and it's an increasing progressive sort of movement, at some point we can't actually tell the difference
between estate bureaucrats and capitalism on the one hand, and capitalism and the chaos of individual capitalistic entities on the other. I think automation is the miracle for them, right? Precisely because it will eventuate the fundamental dream of capitalism itself. But of course, I might be way too pessimistic at this point. So, Ross, can we have a bathroom break? I thought I'd give Nick a chance to
to what you said just then one quick line which is I think a basic distinction between state bureaucrats and the USSR is slightly different so I won't refer to that but the state bureaucrats and capitalists I mean capitalists have their individual interests in mind and states have collective interest of capital in mind which doesn't always align with the interest of individual capitalists. But, you know, they are all participating in that same sort of process to greater and lesser degrees. And I'm not sure that that difference, that distinction is disappearing or not. But that's, you know, maybe a conversation for another time. Yes, good answer.
Yeah. No, I'm not sure. I'm not sure whether it's... But I would say that there is a tendency. There is a tendency. a very sort of rudimentary sort of tendency toward that blaring but yes, I absolutely agree with you that we shouldn't actually jump the gun and say that it would result in the fundamental disappearance of the distinctions between these two categories can there be a non-capitalist AI? should we take a break now? After bathroom break, we are going to talk about this. Okay. Well, stay on the edge of your seats then for the continuation of that one. So, yeah, we will have a bathroom break of what?
10 minutes? 15? 10 minutes is good. Yeah, 10 minutes. And obviously do keep on putting away in the chat. and put an emoji of your choice if you want to ask a question later. So yeah, see you all in about 10 minutes.