Reading 3 min
Absorbed Skill, Breakdown, and Dreyfus's Heideggerian Critique of AI
Skilled activity hides its tools. When you type, you don't notice the keyboard. You notice the sentence you're writing. On a familiar drive, your feet work the pedals while your attention is on the traffic, or on nothing in particular. The keyboard shows up when a key sticks. The brake pedal shows up when it sinks farther than it should.
Martin Heidegger made this ordinary pattern the basis of an argument about what a world is (Heidegger [1927] 2008, §§15–16). On his account, we first meet things not as objects with properties but as equipment in use, things for doing something. Their properties come into view mainly when the doing is disrupted. If that's right, a description of skilled activity that starts from objects and their properties has started in the wrong place.
The claim has had a long research life, largely through the philosopher Hubert Dreyfus. With his brother Stuart, an engineer, he developed a model of skill acquisition in which a novice follows explicit rules, while an expert responds directly to the situation, often without being able to state the rules at all (Dreyfus and Dreyfus 1986). Expertise, on this model, isn't rule-following done faster. It's a different kind of engagement.
Dreyfus drew a sharper conclusion about artificial intelligence. Symbolic AI represented the world as facts and rules for manipulating them. Dreyfus argued that such systems couldn't capture the background of practical understanding that lets a skilled agent see what's relevant in a situation (Dreyfus 1992). He later argued that even approaches calling themselves "Heideggerian," which tried to replace representations with situated behavior, had failed. Fixing them, he claimed, would mean taking embodiment and absorbed coping more seriously still (Dreyfus 2007). His critiques were controversial, and AI has changed a great deal since, but the questions they raised about skill and relevance haven't gone away.
The same analysis matters for work on skill learning and for the design of tools and interfaces. In skill learning, it asks what changes when a learner's attention moves from the rules to the situation, and why an expert who stops to think about technique often performs worse. In design, a good tool, on this view, is one you don't notice while working, and a design flaw is often a breakdown waiting to happen. In each field, the distinction between a tool in use and a tool as an object gives you a precise way to describe absorbed, skilled activity, and to see what a rules-and-representations account leaves out.
References
- Dreyfus, Hubert L. 1992. What Computers Still Can't Do: A Critique of Artificial Reason. Cambridge, MA: MIT Press.
- Free: None for the book; the original statement of the critique is free: Dreyfus, "Alchemy and Artificial Intelligence" (RAND, 1965 0
- Publisher: MIT Press
- Dreyfus, Hubert L. 2007. "Why Heideggerian AI Failed and How Fixing It Would Require Making It More Heideggerian." Philosophical Psychology 20 (2): 247–68.
- Free: None.
- Publisher: doi:10.1080/09515080701239510
- Dreyfus, Hubert L., and Stuart E. Dreyfus. 1986. Mind over Machine: The Power of Human Intuition and Expertise in the Era of the Computer. New York: Free Press.
- Free: None for the book; the skill model's first statement is free: Dreyfus and Dreyfus, "A Five-Stage Model of the Mental Activities Involved in Directed Skill Acquisition" (1980), DTIC
- Publisher: Simon & Schuster
- Heidegger, Martin. (1927) 2008. Being and Time. Translated by John Macquarrie and Edward Robinson. Harper Perennial Modern Thought. New York: Harper Perennial.
- Free: None. Cite by section and German page (H.), which both English translations print in the margin.
- Publisher: HarperCollins (Harper Perennial Modern Thought; ISBN 9780061575594 0 · alternative translation: SUNY Press (Stambaugh, rev. Schmidt)
- Free companion: SEP, "Martin Heidegger" (Wrathall), §2 · IEP, "Heidegger, Martin"