Simon Lindgren
Critical Theory of AI
ISBN: 978-1509-555772
209 pages
Polity Pres
If you ask what artificial intelligence or language models like ChatGPT are, you will typically get either an
answer about the technology behind them or an answer about what they can be used for. The first answer will
be about programming, machine learning, algorithms, and that sort of thing. You may also get the history of
the development of artificial intelligence as a story about brilliant scientists and engineers who have
constructed thinking machines. The second answer will be about the models' superhuman powers and
abilities, increased efficiency at work, clever solutions to complex problems, or how many jobs will be
eliminated and people will therefore end up unemployed.
Both types of answers are important and can raise good questions, but they may still miss the point when
we talk about artificial intelligence or any other form of technology for that matter.
At least that's what Simon Lindgren argues in his new book Critical Theory of AI.
The problem with technical and narrow questions about AI is that they come to see technology as something
isolated. This is actually a fundamental problem with many discussions about technology, namely that it is
seen as something that is in the world; something that can then be chosen to use or not. A very common view
of technologies is that once they are produced, they are simply present and can be used. Technology is
perceived as something specific with a given use.
Lindgren draws on critical theory and post-structuralist theory of power in his definition of artificial
intelligence. With Michel Foucault, he looks at how technology is always part of knowledge-power relations,
where particular privileged understandings are established as true and correct, which has effects on how the
individuals can live their life. In the contexts in which technologies are involved, the individual is positioned
or addressed in a certain way, which is part of the way the individual becomes a subject. It is Althusser's
conception of Ideologically determined and determining interpellation that is at stake here. Technology
becomes part of a discourse where one can fight and negotiate the 'truth'. “Discourse is power and knowledge
embedded in language and social practice” (p.130).
To understand artificial intelligence as something other than specific tools, Lindgren suggests understanding
it as an assemblage. Technologies will always be part of a wide array of different relationships with other
things and contexts. There are people who use these technologies in specific ways, and these technologies
carry significance as economic instruments. They also acquire ideological meanings that become part of the
relationship between humans and machines situated in a context involving a web of power and meaning.
According to this understanding, a technology (such as a coffee machine, a fountain pen, or a language
model) is never simply an addition to humans. As Latour (another inspiration for Lindgren) expressed, a
gunman is more than just a gun + a man. When considering a gunman in a context where other forms of
power and knowledge also exist, it becomes even more complex. However, it's precisely this complexity that
is crucial for understanding AI in a critical theoretical light.
Lindgren employs various terms for AI as an assemblage. He draws, among other things, on Foucault's
concept of the social as an apparatus or dispositif, indicating how all things and relations are interconnected
through knowledge, power, and organization. The assemblage or apparatus is a heterogeneous totality where
laws, power relations, institutions, buildings, organizations, morality, knowledge, science, etc., are
interconnected. "[We] can conceive of AI and humans as entities that are interconnected in a more broadly
encompassing 'machinery,' that is 'the social'" (p. 86).
Artificial intelligence and all other technologies must be understood in relation to the way they are
integrated into established economic and political structures. Therefore, technology is not something
separate from, for example, economic conditions, where it can be included or not. Technologies themselves
have agency (a capacity for action) that influences the contexts they are part of. However, technologies do
not determine societal conditions, which is another ideological understanding of technology that argues that
as a society, we must incorporate technology now that it's here. Technology is understood here as an urgent
driving force for societal change. But critical theory continually points out that it could be different; there is
room for political action and thus changing the techno-economic conditions and circumstances.
When it is claimed that technological development has consequences, it's a specific ideological use of
technology that is part of a power play. Technological development, when referenced, empowers some at
the expense of others, Lindgren asserts with the weight of critical theory. "[Tech] must be understood in
relation to political economy, and that it is not an autonomous force. [...] This means that AI must be seen
from the theoretical perspective of the social shaping of technology" (p.9 and 14).
Currently, AI and language models are celebrated as disruptive technologies that will change the future for
many. This is probably true. Not least, it will contribute to the ongoing climate catastrophe because the
energy consumption involved in training, maintaining, and using digital technologies emits unimaginable
amounts of CO2. Another consequence is the underpaid workers in the global South who provide human
feedback to image recognition programs. And the entire platform economy, which is also linked to the
development of AI, means a setback for the rights and conditions that labor movements worldwide have
fought for because the platform economy turns work into detached "tasks" that private actors can perform:
food couriers, taxi drivers, programmers, etc. AI also means precarity - casual laborers and day laborers
with minimal rights. It's "algorithmically driven digital capitalism," as Lindgren states (p.23).
But this kind of disruption is not highlighted in the ideologically determined perception of technology as
progress. Here, technology is part of a discourse that draws on an enlightenment philosophical understanding
of how history moves forward and upward through people's development of new technologies and
organizational forms. But in reality, technologies like AI are incredibly conservative when you look at how
they actually function, claims Lindgren. "For all its celebration of disruption and innovation, the tech
industry has always tended to serve existing power relations" (p. 65). When looking at the ideologies
programmed into AI, they are completely traditionalist - or perhaps even worse than that. Lindgren goes
so far as to include Yarden Katz's Artificial Whiteness, which argues that AI must be understood as a
'technology of whiteness' because it mimics and serves a logic of white supremacy (p. 138).
The point is that demanding ethical AI isn't enough, as the problem isn't how we as humans can relate to the
technology. AI is systematically connected to a specific form of socialization, contributing to algorithmic
oppression.
Interestingly, Lindgren points to Habermas' discourse ethics as a possible way to see and understand
technology differently. The ideal becomes deliberative democracy, with a focus on centering development,
decisions, and determinations based on the people affected by these processes. It's a fundamental
understanding of direct democracy translated into what Habermas calls discourse ethics: "Only those norms
can claim to be valid that meet (or could meet) with the approval of all affected in their capacity as
participants in a practical discourse. [...] Translated into the AI domain, this means that the moral
architecture that underlies algorithmic systems must be grounded in the practical participation of those that
are affected by the systems" (p.159).
When I call this reference to discourse ethics interesting, it's because it fundamentally breaks with the
analytical framework Lindgren has set up using original critical theory and post-structuralism. It's not
coincidental that Habermas is very critical of post-structuralist thinking, as it breaks with the enlightenment
philosophical and Kantian optimistic idea of reason as the saving factor. Habermas adheres to Kant's idea
that reason will ultimately find the truth - and his moral philosophy also pointed to the possibility of
morality. Habermas' concept of communicative action relies on the notion that reasonable conversation
leads to a shared truth: the common good.
However, the post-structuralist thinkers Lindgren draws on reject the existence of such reason. Therefore,
discourse means something entirely different for them. This contradiction between the analytical framework
and the solution Lindgren points out could have been highlighted by the author himself. There's an
unresolved duality in Lindgren, which is also evident in his use of the concept of ideology. On the one hand,
ideology means the values circulating in conjunction with knowledge and power (Foucault). But on the other
hand, ideology for Lindgren means false consciousness, drawing on a Marxist notion that citizens are deceived
through bourgeois ideology. These two concepts of ideology, as far as I can judge, stand side by side in the book
without being brought forward as conflicting.
In my opinion, these are blemishes in a very good book that offers a different and critical view of
technological development. Lindgren looks at AI as an assemblage based on various dimensions, providing
a critical theoretical perspective on the revered technology. There are insightful analyses and discussions
throughout the book. Lindgren writes very well, making the book accessible to those interested in a
comprehensive view of AI - a perspective that doesn't limit itself to viewing technology as either good or
bad but attempts to delve into the various contexts in which technology operates and influences people's
social relations and the individual's relationship with oneself.