AI has broad societal and economic consequences – impact entire societies
Verdegem 22 , P. Dismantling AI capitalism: the commons as an alternative to the power concentration of Big Tech. AI & Soc (2022). https://doi.org/10.1007/s00146-022-01437-8 Dr. Pieter Verdegem is a Senior Lecturer in Media Theory in the Westminster School of Media and Communication and a member of CAMRI (the Communication and Media Research Institute). – Maren Lien
An important aspect of understanding AI capitalism is to consider AI as a General Purpose Technology (GPT) (Trajtenberg ). GPT are enabling technologies, meaning that they open up new opportunities, in addition to offering complete, final solutions. Other examples of GPT are the steam engine, electrification and the Internet. They have three main characteristics: (1) they are widely used; (2) they are capable of ongoing technical improvement; and (3) they enable innovation in different application sectors (Bresnahan, 210: 764). AI qualifies this definition. Given their pervasiveness and the complementary waves of innovation they produce, GPT cause economic disruption. They affect entire economies, potentially drastically altering societies through their impact on pre-existing economic and social structures (Trajtenberg ). Economists study the impact of GPT in terms of the emergence of winners and losers. The winners are those associated with the emerging GPT, whereas the losers are those who cannot benefit from the unfolding GPT. However, looking at other GPT invites us to look beyond winners and losers and to consider GPT—and thus also AI—as a public utility. The importance of electricity and the Internet, for example, has opened debates about the need of regulation, and the decision to not merely leaving these technological developments over to the market alone, or at least to have some intervention from society in it. This is particularly important given the times we live in: during the pandemic, we all have witnessed the crucial role of digital platforms in everyday life. As such, we need to be aware that AI can facilitate a further polarisation of already unequal societies (Crawford ; Dyer-Witheford et al. ; Lee ). In any case, considering AI as a GPT and not just a digital technology that is owned and used by private entities but one that has broad impact on society, opens up new questions about how to conceptualise AI capitalism. This is where the work of Kate Crawford comes into place. In her book Atlas of AI, Crawford () offers a comprehensive and nuanced understanding of AI. According to her, AI simultaneously refers to technical approaches, social practices and industrial infrastructures (Crawford : 8–9). First, AI refers to technical approaches. Advancements in machine learning (ML) have been the most powerful contributor to the development of AI in the past two decades (Asaro ). ML is a paradigm that allows programs to automatically improve their performance on a particular task by learning from vast amounts of data (Russell and Norvig ; Lee ). It is based on statistical patterns and correlation in large data sets, starting to be used in the late 1980s–early 1990s. Earlier versions of machine intelligence—e.g., expert systems—were primarily rules-based, making use of symbolic logic and involving human experts generating instructions codified as algorithms (Agrawal et al. ). The problem was that they could not cope with the complexity of most applications. Unlike expert systems, powerful ML algorithms learn from the ground up, not from humans but from data (Alpaydin ). The rise of ML can be explained by more powerful and reliable computing infrastructure, which has made possible the development of systems driven by real-world data (Lee ). The availability of significant amounts of data further enables the development of learning algorithms that derive solutions using statistical methods. Deep learning (DL) and neural networks (NN) are the driving forces behind more recent developments in ML. In the early 2000s, ML pioneer Geoffrey Hinton (LeCun et al. ) demonstrated the power of DL neural networks: this allows automatically processing of unlabelled data, which has led to more effective applications of AI that we are now using every day (e.g., online services). Second, the social practices of AI refer to the classification systems, developed by humans, which are behind algorithms, ML/DL models and AI systems. Crucial questions that we need to ask here are: Who is involved in developing these classification systems? Who decides what classifications are used? and; What do they look like? Ultimately, these are political (power) questions about inclusion and representation (Crawford ). Important challenges exist around AI, bias, fairness and discrimination (Costanza-Chock ). Questions about how representative these classification systems are, are crucial in this. How to avoid bias and support inclusion in AI systems are important political issues that urgently need to be addressed (Brevini and Pasquale ). Last, the industrial infrastructures of AI refer to the computing power, algorithms and data sets that are the source of knowledge and production. This infrastructure not only entails the possibilities of collecting vast amounts of data—which are needed to train algorithms—but also the computational power necessary to develop and perform ML and DL models. Few companies have access to the required data sets, possess the necessary computational power to run ML/DL and are able to attract the brightest AI scientists, which means we are witnessing a concentrated industrial AI infrastructure, leading to AI oligopolies/monopolies (Dyer-Witheford et al. ; Riedl ). This gives a lot of power in the hands of a small number of corporations (Montes and Goertzel ) and is why we need to scrutinise economic power within AI capitalism. Offering an encompassing view of AI capitalism, is important: we need to be aware of how material AI is and that its production is based on natural resources, human labour and industrial infrastructures. Looking at the broader picture of change within technologies, beliefs and infrastructures simultaneously, however, also risks overlooking the issues of a concentration of power. To deal with this, we need to go back to political economy as this is the framework that puts power at the centre of its analysis. Political economy is particularly interested in the relationship between techno-economic systems and their impact of the broader societal structure (McChesney ). The industrial infrastructures of AI also contribute to a concentration of power, which has not only an impact on the social practices of AI but also how its technological development will happen in the future, which explains the importance of this perspective.