Edemilson Paraná is an assistant professor of sociology at the Federal University of Ceara, Brazil, and authors of the books Digitalized Finance: Financial Capitalism and Informational Revolution and Bitcoin: a Utopia Tecnocrática do Dinheiro Apolítico (English translation forthcoming). You can check this interview conducted by Evgeny Morozov on Bitcoin.
This Monday, he will give a talk at the University of Toronto about his new research involving artificial intelligence as financial infrastructure. In this interview with Rafael Grohmann, he explains the core question, how it relates to the broader agenda of critical AI studies, the notion of infrastructure, and other technology debates from a Marxist perspective.
DIGILABOUR: What does it mean to understand artificial intelligence as a financial infrastructure?
Edemilson Paraná: The idea is to understand that the penetration of artificial intelligence in financial markets – which has been occurring more rapidly in recent years – is not something that happens out of nowhere but has to do with a set of transformations that involve different scales and enable AI to be implemented in this way in contemporary markets. How is AI implemented? It is used for risk assessment, credit assessment, real-time trading of the most diverse financial instruments, market administration, and management. AI is spreading very quickly across finance. The dimensions in which it is most intensively used have to do with three aspects: 1) credit score and ranking of access to credit in the case of banking services; 2) accounting and risk management in financial companies; 3) management of investment portfolios and trading in the capital markets. There are large funds that use algorithmic trading and artificial intelligence and sell access to these products to their clients and investors. There are huge funds like BlackRock and Bridgewater that are using artificial intelligence on an increasing scale.
Thus, it is necessary, first, to understand at what scale this occurs. There has been a major change in the financial market in recent decades that practically makes the market merge with a technological infrastructure, with a socio-technical system that serves as the basis for other interactions. I have been discussing this subject in my research for quite some time regarding the idea that we live in a context of “digitalized finance,” as I’ve defined it. It is no longer possible to understand finance outside of informational technologies dynamics. These are electronic markets where the negotiations through information and communication technologies become ubiquitous. With the advancement of computational processing capacity, these trading models become more refined. As all these layers overlap each other, there is, on the surface, “intelligence,” the “intelligent” layer of artificial intelligence. For this, it is necessary then to assemble a huge dimension of overlapping infrastructures so that this artificial intelligence can gain dominance in these fields. This has an important implication, which is to understand that this ‘game of scales’ is fundamental for us to access what artificial intelligence is in the financial market today.
There is a “micro” application of artificial intelligence in specific services, and there is a macro systemic deployment of artificial intelligence – less understood – in the financial system. That is, artificial intelligence can enable gains, returns, profits, and efficiency in the strictest economic sense at the micro level for some agents, above all, agents that are better positioned in the economic and technological infrastructures of the financial market. But at the macro level, there is increased risk, unpredictability, and perhaps inefficiency. So, this contradiction needs to be better addressed. Increased “efficiency” at the micro level, with increased risk and complexity, in many cases means inefficiency at the macro level, with increased concentration of power and informational control in the markets. That’s what I try to explore by treating artificial intelligence as a financial infrastructure.
Nowadays, AI is becoming increasingly unavoidable. To enter the market, whether as a small or large investor, accessing these resources is becoming more and more necessary. However, understanding the difference in scale is fundamental to both the agents in the market and the socio-technical functioning itself. This relates to what is called the fallacy of composition, something widely explored by Keynes in economics. People who analyze technology in the market often do not see what is fundamentally happening because the analysis is always at the micro and descriptive level. What is the fallacy of composition? It is the idea that the whole is not merely the quantitative sum of the parts. The whole has emergent properties that are qualitatively distinct from the sum of the parts. To use another example, it is analogous to the basic functionalist postulate of sociology according to Durkheim. The social is something different from the mere sum of individual interactions as it has properties of its own. I want to demonstrate in this work that this also applies to the application of AI in the financial market and that it is producing some worrying effects. This needs to be evaluated in light of the contradiction between these two dimensions.
DIGILABOUR: What is your central argument?
EDEMILSON PARANA: My argument focuses on the idea that greater efficiency at the micro level does not necessarily result in greater “efficiency” at the macro level; rather, it is the opposite. Fierce competition in the market forces the adoption of information and communication technologies. The sociotechnical and infrastructural bases of market functioning – and financial markets themselves – are historically very sensitive to information. Perhaps the financial market is one of the most information-intensive economic sectors. As a result, these are technologies, dynamics, and sectors that have a strategic aspect. That is why the financial market tends to anticipate other sectors of the economy in adopting cutting-edge technologies. This is something I have been developing for some time. Nowadays, we talk about algorithms, Silicon Valley, media companies, and social interaction, but algorithms have been applied in the financial market since the 1980s. We talk about neural networks, machine learning, and deep learning for contemporary informational and educational products, but they have already been applied in the financial market way before they became present in our everyday social interactions. Information and communication technologies are, therefore, the infrastructural basis on which markets have operated for some time now. I discussed this in my first book, “Digitalized Finance.” These technologies anticipate and compress space-time flows, make it possible for the market to expand its base and speed of financial negotiations, and this produces an increase in complexity and concentration, with augmented risks and inequalities. In my first book, I called this the spiral of complexity of financial markets. AI now enters as an emergent financial infrastructure, composing this sociotechnical complex, which also, of course, has its political and institutional aspects. The major players in the financial market aim to adopt it as a general-purpose technology, to be increasingly used as the basis of all other financial services.
DIGILABOUR: What are the AI imaginaries involved in this?
EDEMILSON PARANA: Agents anticipate greater control, transparency, predictability, productivity, and profitability. The reports from regulators, large companies, and consulting firms are generally praising this transformation of AI in infrastructure and the benefits it can bring to the market. However, I argue that the question of scale complicates this sociotechnical imaginary as it fails to address the aspects of power, control, and political governance of these economic infrastructures and techniques. At this point, we encounter problems such as the lack of knowledge about the logic of causality within the models, the fallacy of composition, complexity, volatility, uncertainty, and other difficulties that artificial intelligence, in its current stage, cannot contain and may even exacerbate.
The argument, then, is that artificial intelligence often does the opposite of what these agents claim it will do. Tensions between the micro and the macro, between the material and the ideational, and between the technical and the political are not new but are fundamental to understanding the spread of artificial intelligence as a financial infrastructure. Augmented artificial intelligence, in its eventual use as a general-purpose technology in financial markets, tends to enhance rather than control risk and opacity and can bring new types of unforeseen risks.
DIGILABOUR: How do you conceptualize infrastructure?
EDEMILSON PARANA: I address infrastructure in a broader sociological sense. It’s not just about physical objects. They are not simply piles of things that make up the technical operability of certain informational processes, but rather a complex, scalar composition that involves natural resources, labor, and, of course, also the materiality of the objects that are mobilized in the sociotechnical and institutional functioning of these structures. From regulatory apparatuses and institutional arrangements to submarine cables, all of these elements make up the functional infrastructures of the financial market. Artificial intelligence is increasingly becoming a part of this infrastructural complex of market functioning.
Thus, the everyday base on which and from which AI operates often enters the dynamics in an invisible, subtle way, and we may not fully understand how the interrelationship of these different layers takes place to produce the markets that we have today, with all of their tensions. It is as if we are trying to access this “big global system of machinery,” a term coined by Marx and updated by my colleague Esther Majerowicz. This is a “machine system” that involves labor relations, data collection, use and storage, conflicts, tensions, and ideas and narratives. The way in which we create ways of visualizing, explaining, and presenting the market to society also shapes and conditions the way markets work. As the fields of Science and Technology Studies (STS) and the sociology of financial markets have been debating for some decades, the way in which these markets are permeated by performativity, discourses, and imaginaries has a material existence that shapes and conditions the way they work.
When I’m talking about infrastructure and thinking about AI as an infrastructure, I’m trying to connect artificial intelligence to a more integrated and systemic way of thinking about markets.
DIGILABOUR: In recent years, there has been a proliferation of critical AI studies, but the financial market is still a blind spot in this discussion. Why?
EDEMILSON PARANA: I think a gap exists because finance still appears as something solely in the economic domain of social processes. Despite the efforts of STS and the overall sociology of science and technology, particularly since the 2000s, to demonstrate the social, constructed, performative, and even narrative character of markets, critical AI studies still rarely research this dynamic. Financialization studies, platform studies, and critical technology studies are not communicating effectively with each other. One of the goals of my work is to bring these worlds together and, together with other colleagues, fill this gap.
On the one hand, there is an idea that this is still a matter of purely economics, of economists, and not something of the social sciences. That is, we criticize power, but to do that, we take specific sociotechnical systems, unravel and show how the relations of exploitation and domination occur there, in that context. The problem with finance is that, even though it is shaped by important particular local arrangements, it is a complex system articulated globally, which creates methodological difficulties in accessing some of the dynamics of power, hierarchy, and inequality that occur in the financial system. So, on the one hand, there is a disciplinary challenge, and, on the other, a methodological one.
In my view, to critically access the financial market, it is essential that we look at the interrelated dimension of scales, taking seriously the systemic and structural level. It is necessary to think about the systemic causalities that occur in this complex game played in the markets based on the tension between the micro and the macro. This is key to understanding artificial intelligence as a financial infrastructure. For example, we can see this in the logic of systemic risk. How do we analyze the risk that AI and the increased use of AI can bring to financial markets? We need to think about this in a combined and coordinated way with the use of AI in several markets at the same time, by agents with different AI strategies interacting with each other. This is the modus operandi of the financial market nowadays, taking place in real-time in a global and interconnected way. If we only consider how it occurs in a specific financial market or in a specific product, it will be difficult to understand the contradictions that I am trying to address. Of course, there are problems of bias and black boxes that the literature has also been addressing for a long time. But it is necessary, in my view, to understand problems such as the ability of artificial intelligence to be pro-cyclical, meaning that the behavior of an AI tends to be reinforced by the behavior of another AI, producing market movements that, in the aggregate, can increase systemic risk. I think this is a fundamental issue for us to analyze, even though it has been a little outside the scope of this field of study.
Another important example is the problem of explainability and causality within AI models. In AI in financial markets, this is very serious, because you have portfolio management, buying and selling assets to have a certain financial performance and give a certain return. You throw in deep learning, which is also the dominant model in finance, and it gives you an accuracy rate, an excellent accuracy from the point of view of the financial return that you can have on that strategy. It’s just that you simply don’t concretely know exactly what produced that result. This is not a detail. This makes all the difference, for example, for market coordination, market regulation, monitoring risk logic, and even for the investor himself. Maybe there is a hidden cause working for it to have that profitability that can be extremely obscure from the point of view of the maturation of its portfolio, which goes in a totally different direction if minimal conditions change. This lack of explanation of the models, these black box dynamics, is fundamental to understanding how things work at the operational level, but they have extremely relevant systemic implications that often cannot be learned if we cannot understand the relative autonomy of these dimensions. Of course, this is a co-determination between these realms, but there is a relative autonomy between these micro and macro dimensions. Also, an intersection occurs between economics and politics in the power dynamics, particularly when the scales start to pile up. It relates to the issues of inequality and market concentration that are also very important to platform studies. So you have to think about these scales. Scale is always about power. It is not possible to think about scalability in a sociotechnical system without thinking about power dynamics. I think this is a very interesting point of contact to start addressing this gap.
DIGILABOUR: In addition to your research on AI as financial infrastructure, how have you positioned yourself in the debate on technofeudalism?
Edemilson Paraná: I am about to intervene in this very interesting debate, in which several qualified colleagues are taking part. I think this discussion is really important. Like [Evgeny] Morozov, I am a critic of the technofeudalism framework. I believe that we are not experiencing anything different from capitalism. Capitalism is a very adaptable system, with an extraordinary ability to rebuild and reinvent itself. Therefore, I do not think that, given the transformations that have taken place in recent decades, we are facing a new mode of production and heading in this direction.
Mobilizing typical aspects, processes and strucutres of other modes of production is a resource that capitalism has historically used to continue reproducing and remaking itself in light of the contraditicions and limits imposed on it. However, this does not change the fact that labor exploitation is a generalized social reality, that the pursuit of profit is an end in itself, and that the valorization of value is a central and structuring element of economic and social dynamics.
But although it is true that we are still continuing in the capitalist mode of production, it does not seem appropriate to me to think that things are the same, that they are exactly what they have always been. I think there are extremely significant and important changes taking place that need to be carefully addressed, as they can mean a phase change, a general change in the organization and disposition of relations within capitalism. This seems like a sensible hypothesis to me.
I believe that we are experiencing a phase change in capitalism in recent decades. Capitalism is morphing into something very different from before. Just as welfarism, Keynesian-Fordist capitalism was different from liberal capitalism, which, in turn, is different from contemporary neoliberal and financialized capitalism, I think we are crossing another line now, into another type of capitalism. I believe that this change has digital transformation at its core, with the digitization of processes, dynamics, and social interactions. This means, in my view, another qualitative form of functioning of economic and social relations within capitalism.
DIGILABOUR: What does it mean to be a Marxist researching technology today?
EDEMILSON PARANÁ: It’s a very dangerous, but also a very interesting time to be a Marxist – if we can ever come to terms with what “Marxist” actually means. We are, in my view, experiencing a fundamental qualitative change within capitalism. Perhaps an unprecedented change.
Why do I say it’s dangerous? In the face of these changes, there are two temptations, which are strong for all analysts, but perhaps especially for Marxists. One is to say that these changes are just a phenomenal expression of something that we already know. And that, for that reason, it only remains for us to make a good critique of these changes in the light of the basic propositions, of the fundamental concepts that we already know. Therefore, going into too much detail in the description, understanding, and careful investigation of these changes would be something not only unproductive but, at the limit, fetishistic and ideological. That is, this position argues that nothing has changed and that things continue to be exactly as they are.
A second position, that we have to be careful of, is the idea that everything is changing in an irresistible, irreversible, unavoidable way, and that these changes represent a complete reconfiguration of things, with possibilities for the end of capitalism itself. This is another idea we need to be cautious about. Let us remember that some Marxists – of course not only them, but also them – have already predicted the end of capitalism a few times, but capitalism insists on continuing to remake and reinvent itself. At every major crisis, some critical thinkers appear to say “the end of capitalism is near”. But these crises are instrumentalized precisely so that capitalism, through creative destruction – recalling Schumpeter’s definition – can reinvent itself, albeit producing a set of tragedies along the way within this “reinvention”.
So, I think this requires both openness and enthusiasm to understand what’s new, but a certain amount of caution and a good dose of scientific humility so as not to fall into the panacea that absolutely everything is new. Faced with this very delicate moment that we are living, being a Marxist means dedicating yourself with theoretical rigor, analytical depth, and a lot of empirical care to these novelties, but without naivety to think that they are a complete reversal of everything that exists. The new reproduces itself in the old, and the old reproduces itself in the new. Understanding the nuances of this dialectic is an arduous task, which requires the best of our efforts and our intelligence, especially in this area.
My diffuse, far-reaching theoretical bet is that the fundamental dilemma of the reconfiguration of capitalism in our time can be better addressed from the interrelationship between financialization and digitization. The reconfiguration of capital logics through the rearrangement between finance and production, on the one hand, and, on the other, the transformation of productive processes and social life with the broad and extensive digitalization, are – together with the environmental catastrophe and the need for social reconfigurations that this will demand – two of the most relevant contemporary processes.
As the relationship between technology and society becomes, for positive or negative reasons, increasingly central in social life, I think that it is necessary to have a critical, interdisciplinary, systemic, and rigorously attentive to complexity thinking – as a good approach should look like in this thought stream. In my view, this has clear advantages over other currently dominant approaches.