Understanding machine learning – a philosophical inquiry of its technical lineage and speculative future

This dissertation presents a philosophical critique of machine learning based on an investigation into its technical lineage. It begins by explicating Martin Heidegger's remarks that cybernetics would take the place of philosophy and that "only a god can save us" from a technocratic society. But whereas Heidegger's critique assumes the universality of cybernetics, this assumption can be challenged by examining the transactions of the Cybernetics Conference. Such an examination exposes the inherent conflicts between disciplinary knowledge, which helps explain the failure of cybernetics in attaining scientific achievements. This dissertation further argues, even though cybernetics has shaped the historical development of computer science, artificial intelligence (AI), and machine learning, cybernetics and universal computing can be distinguished as two mutually imbricated intellectual traditions. AI research explores how human intelligence can be simulated on a universal computer, departing from the cybernetic objective of understanding the mechanisms of living organisms. In particular, Ray Solomonoff showed how his abstract machine-learning algorithm can recognize any subtle data patterns, anticipating the capability and limitation of deep learning. Deep learning has made possible generative AI applications such as DeepBach, raising questions about the possibility of computational creativity or emotivity. This dissertation deliberates such questions by turning to Gilbert Simondon, whose model of philosophy derives from subatomic quantum behavior and other modern scientific theories, as opposed to everyday intuition applied to large objects. While cybernetics has also been influential to his philosophy, Simondon rejects the cybernetic mechanization of the living and its blurring of the life-machine boundary. Rather than conflating the human and the machine, Simondon's theories of concretization and individuation of transindividual relations suggest how technology co-evolves with the human and the social. These theories were adopted by Andrew Feenberg in his critique of the Internet and by Bernard Stiegler in his critique of algorithmic governmentality. Even though Feenberg's critique emphasizes the openness of the Internet while Stiegler's reveals the closed character of the 24/7 computational infrastructure, their interpretations of Simondon are compatible, as both recognize the revolutionary potentiality in human-technology co-evolution, which can be differentiated from J. C. R. Licklider's human-computer symbiosis.