The Philosophy of Artificial Intelligence
The field of Computer Science has long studied Artificial Intelligence (AI). However, since the advent of early large language models like GPT-1 in 2018, the pace of progress has accelerated dramatically. Some have argued that the Turing test [1], which was designed by Alan Turing in 1950 to assess a machine’s ability to exhibit human-like intelligence through conversation, has been largely solved. Natural questions arise following these rapid advances in AI.
A founding objective of Artificial Intelligence (AI) has been to replicate Human Intelligence (HI). With behavioral benchmarks like the Turing Test accomplished by modern deep learning architectures [2], the field of AI is in need of a new grand vision. In particular, what is the ultimate limit of AI, and how does our understanding of HI inform this limit?
Historically, the relationship between human and artificial intelligence centered on the Computational Theory of Mind [3], [4], which sought to understand human cognition through the lens of early computational architectures. Philosophical critiques of AI focused on the inherent limitations of disembodied, algorithmic approaches to replicating human understanding [5], [6].
Contemporary AI alignment literature focuses on managing future capabilities, but it tends to ignore or abstract away the human element. Bostrom [7] frames artificial superintelligence as an optimization process that is independent of human origins. Similarly, Russell [8] centers the “control problem” on human values, but views human intelligence primarily as a reward function to be reverse-engineered and optimized.
Rather than treating humans merely as a biological baseline we intend to overcome or as an optimization objective to be satisfied, I propose a new framework: the “Philosophy of Artificial Intelligence.” This human-first approach lies at the intersection of the philosophy of mind, cognitive science, and contemporary AI alignment. It is concerned with the study of human and artificial intelligence, their influence on one another, the relationship between their components, and the capabilities of both.
The Philosophy of Artificial Intelligence aims to understand the origins, mechanisms, and limits of human intelligence, which will inform a deeper understanding of artificial intelligence systems. Intelligence is an inherently human construct, which we discuss in terms such as “understanding,” “reason,” and “alignment.” The most accurate understanding of these terms arises through our personal experience and interpersonal relationships as we perceive and interact with one another. Therefore, I propose that the “ground truth” information which informs the grand objectives of AI systems must be our interpersonal perceptions of intelligence.
Introspection of intellect is a difficult task. Therefore, to gain a functional view of intelligence, we must rely on the great philosophers, theologians, and scientists of history. Critical to our exploration are the divisions of human decision-making that have been proposed. Many, from Plato and St. Augustine to Descartes and Freud, have proposed forms of dualism, arguing that intellect is divided among physical and non-physical components. Others, such as Nietzsche, present a purely physicalist or materialistic explanation for the origins of the intellect. Still others view the mind through the lens of functionalism, arguing that mental states are defined by what they do, regardless of the physical substrate on which they run. If intellect is functional and substrate-independent, is it possible to reproduce it artificially in silicon? Would this reproduction be in mechanism or only in outcome? If it is true there is a non-physical component to intellect and that component cannot be reproduced, then what remaining parts of the intellect are physical and can these be fully reproduced artificially? Finally, of such questions, which are experimentally verifiable and which, by their nature, are not? Such are the questions that the Philosophy of Artificial Intelligence seeks to answer.