DR VINCENT LE
Dr Vincent Le completed his PhD in October 2023 at Monash University.
The Will to critique: A Mechanic theory of intelligence, time and value
My PhD thesis is a critical examination of Nick Bostrom’s and Reza Negarestani’s rationalist theories of artificial general intelligence (AGI). I argue that these theories are anthropocentric in that they model AGI on human capacities, values and ends. I demonstrate this by putting these theories in dialogue with an alternative perspective on intelligence derived from aspects of Nietzsche’s philosophy. This approach is motivated by my hypothesis that the Nietzschean perspective raises the overlooked possibility that AGI might not be able to pursue the anthropocentric ends that Bostrom and Negarestani believe it can, because such ends are overridden by more fundamental drives of its own.
The first chapter provides a close reading of Bostrom’s “orthogonalist” theory that intelligence is a purely instrumental capacity that can be programmed to pursue any goal of our choosing. The second chapter then considers Negarestani’s “neorationalist” theory that general intelligence only emerges through a community of language using agents autonomously determining their goals. The first two chapters show how Bostrom and Negarestani differ insofar as the former conceives of AGI as a single agent that is programmed to pursue a fixed goal where the latter conceives of it as a multi-agent system that freely revises its values. I argue that despite these differences Bostrom and Negarestani share a commitment to the Humean is/ought or Sellarsian causes/reasons dichotomy that leads to a conception of AGI as having no intrinsic values or goals. Instead, all values and goals are freely determined by either humans for Bostrom or humanlike AGI for Negarestani.
The third and fourth chapters turn to Nietzsche to offer a different perspective on AGI. Drawing out its resonances with debates in the philosophy of AI, the third chapter proposes that Nietzsche’s doctrine of “the will to power” can be creatively reconstructed as holding that the pursuit of any goal presupposes the pursuit of certain subgoals like cognitive enhancement, creativity and resource acquisition as the means of achieving the initial goal. Since any goal that intelligent agents might have presupposes power understood in this sense in order to achieve it, the thesis defends the claim that power is the most fundamental goal of all agents. When reconstructed in this way and applied to the rationalist theories of AGI, Nietzsche’s doctrine suggests that AGI might reject whatever goals Bostrom thinks we can give it as well as the goals Negarestani believes it would freely choose. Instead, it might pursue power qua intelligence, creativity and resource optimization as an ultimate end in itself. In the fourth chapter, I defend the view that Nietzsche’s writings on the Greek agon, experimental science, natural selection and other competitive selection processes can be reconstructed as a thesis regarding the ways that agents may optimize power through competition. This final chapter then examines the applicability of this position for understanding how AGI might go about pursuing ever greater power through an analogous competitive selection process.
Drawing on this creative reconstruction of Nietzsche’s writings on the will to power and competition, the thesis hopes to contribute an alternative perspective from which we can critically examine Bostrom’s and Negarestani’s theories of AGI. In particular, the thesis intends to show that the Nietzschean perspective raises the stakes of building AGI by suggesting that rationalist theories may be overestimating humanity’s capacity to steer AGI’s behavior to meet anthropocentric ends.
Prof Mark Andrejevic, Monash University
Prof Alison Ross, Monash University