Aleš Bunta (ZRC SAZU)




Nietzsche, Embodied Errors and the Backpropagation Algorithm


In this paper, we once again revisit the often-overlooked yet crucial Nietzschean concept of ‘embodied errors.’ The comparative significance of this notion for psychoanalysis is evident in Nietzsche’s own use of embodied error to conceptualize instinct or drive. However, our focus here lies in a different comparison—namely, the parallel between the theory of embodied errors and the fundamental idea underlying the so-called backpropagation algorithm. As is well known, this algorithm serves as the core principle of ‘learning’ in artificial deep neural networks and, as such, underpins the remarkable capacities of contemporary artificial intelligence systems. At the heart of this comparison lies a critical observation: the backpropagation algorithm fundamentally functions as a system that produces the effects of ‘correct (re)cognition’ through the intricate manipulation of errors. In other words, within artificial neural networks, ‘correct (re)cognition’ emerges solely as a surface effect of a process in which errors serve as the only true ontological building blocks. In this sense, (re)cognition takes place—yet at no point in the process does it actually occur in any substantive sense or manifest as a discrete epistemic event. This paradox closely mirrors Nietzsche’s explanation of human, biological recognition and understanding.