Why AI Cannot Formalize Human Intuition

Philosopher Hubert Dreyfus famously challenged the foundational assumptions of artificial intelligence by arguing that human intelligence and intuition cannot be reduced to formal symbolic rules. Drawing heavily from the existential phenomenology of Martin Heidegger, Dreyfus asserted that early symbolic AI—often called "Good Old-Fashioned AI" (GOFAI)—was fundamentally flawed because it treated the human mind like a disembodied computer. This article explores how Dreyfus used Heideggerian concepts of embodiment, practical engagement, and uncodifiable background knowledge to prove that human intuition operates entirely outside the realm of explicit algorithms and symbolic logic.

At the core of traditional symbolic AI is the assumption that the mind processes information by manipulating internal symbols according to strict, logical rules. AI pioneers believed that if they could just map out enough facts and rules about the world, a machine could replicate human intelligence. Dreyfus identified this as a philosophical error stretching back to Plato and Descartes, who viewed knowledge as detached, theoretical, and explicit. To counter this, Dreyfus introduced Martin Heidegger’s concept of Dasein, or "being-in-the-world." Heidegger argued that humans are not detached observers calculating facts about their environment; rather, we are fundamentally embedded in and constantly interacting with a meaningful world.

To explain how we navigate the world without rules, Dreyfus utilized Heidegger’s distinction between two modes of engagement: "ready-to-hand" (zuhanden) and "present-at-hand" (vorhanden). When a skilled carpenter uses a hammer, they do not calculate its weight, trajectory, or kinetic energy. The hammer is "ready-to-hand"—it becomes a transparent extension of the carpenter's body, guided by fluid intuition. Only when the hammer breaks or goes missing does the carpenter step back and analyze it as an isolated object with specific properties, making it "present-at-hand." Dreyfus pointed out that symbolic AI operates exclusively in this detached "present-at-hand" mode. By forcing machines to build models of the world out of discrete data points and rules, AI researchers completely missed the primary, intuitive way humans actually function.

Furthermore, Dreyfus argued that human intuition relies on a vast "holistic background" of shared cultural practices, bodily skills, and common sense that cannot be formalized. When an AI tries to understand a simple situation—like ordering food at a restaurant—it must rely on millions of explicit rules about human behavior, physics, and social norms. However, if you try to write rules for how to apply those rules, you fall into an infinite regress. Humans avoid this trap because our physical bodies, biological needs, and cultural upbringing automatically filter what is relevant in any given moment. Because this background coping is lived rather than thought, it cannot be broken down into the context-free symbols required by a computer program.

Dreyfus solidified this critique in his Five-Stage Model of Skill Acquisition, which demonstrates the exact opposite of what symbolic AI assumes. When humans learn a new skill, like driving a car or playing chess, they start as novices who rely heavily on strict, formal rules. However, as they gain experience and reach true expertise, they abandon those rules entirely. A grandmaster chess player does not calculate millions of symbolic move trees; they intuitively recognize the holistic "gestalt" of the board and immediately sense the right move. Therefore, Dreyfus concluded that symbolic AI is not modeling true human intelligence at all—at best, it is only capable of mimicking the clunky, rule-bound behavior of a human novice.