Hubert Dreyfus is a philosopher at MIT and an outspoken critic of artificial intelligence development. Dreyfus's primary interests included phenomenology, existentialism, psychology, and artificial intelligence's philosophical implications. 

Dreyfus gave a negative appraisal of AI's progress and a critique of the field's philosophical roots in a series of papers and books, including 

• Alchemy and AI (1965), 
• What Computers Can't Do (1972; 1979; 1992), and 
• Mind over Machine (1986). 

Dreyfus' criticisms are in the majority of introductions to AI philosophy, including 

• Russell & Norvig (2003), the classic AI textbook, and 
• Fearn (2007), a survey of modern philosophy.

Dreyfus's objective

Dreyfus contended that human intelligence and expertise are determined mainly by unconscious processes rather than conscious symbolic manipulation and that formal rules can entirely capture these unconscious abilities. His critique was on the concepts of modern continental philosophers such as Merleau-Ponty and Heidegger. In the initial wave of AI research, the researchers addressed it, which attempted to reduce intelligence to symbol manipulation by using high-level formal symbols to represent reality.

Dreyfus' critique

Dreyfus detailed the history of AI in "Alchemy and AI (1965)" and "What Computers Can't Do (1972)", mocking the field's boundless optimism. For instance, Herbert A. Simon anticipated in 1957, following the success of his software General Problem Solver, that by 1967: 

• In chess, a computer would be the global champion.
• A computer would uncover and establish a critical new mathematical theorem.
• The majority of psychological theories will be as computer programs.

Dreyfus' four AI research assumptions

Dreyfus established four philosophical assumptions that supported early AI researchers' belief that human intelligence was on symbol manipulation in Alchemy and AI and What Computers Can't Do. "In each case," Dreyfus says, "workers in [AI] regard the assumption as an axiom, promising outcomes, whereas it is, in fact, one hypothesis among several that the success of such effort must evaluate."

• The biological assumption

The brain processes information in distinct actions through biological on/off switches.

• Psychological assumption

The mind can be a device that operates according to formal rules on bits of information.

• The epistemological assumption

All knowledge is formalizable. This assumption is a philosophical question concerning epistemology or the study of knowledge. 

• The ontological assumption

The world is of self-contained facts that self-contained symbols can represent.

Dreyfus' Knowing-how vs knowing-that: 

Dreyfus examined the distinction between human expertise and the computers that purported to capture it in Mind Over Machine (1986), written during the heyday of expert systems. This notion expanded on concepts from What Computers Can't Do. He made a similar argument against Allen Newell and Herbert A. Simon's 1960s "cognitive simulation" school of artificial intelligence research.

Dreyfus maintained that human problem solving and expertise are dependent on our innate sense of context, of what is significant and fascinating in a given circumstance, rather than on the process of combing through options to locate what we need. In 1986, Dreyfus defined it as the divide between "knowing-that" and "knowing-how," referring to Heidegger's distinction between present-at-hand and ready-to-hand. 

That is our capacity for conscious, step-by-step problem resolution. We use these abilities when confronted with a challenging problem that demands us to pause, take a step back, and search through potential solutions one at a time. The concepts become extremely precise and simple: they become context-free symbols that we may control using logic and language. These are the abilities demonstrated by Newell and Simon through psychology experiments and computer programs. Dreyfus concurred that their systems accurately replicated the capabilities he refers to as "knowing-that."

On the other hand, knowing-how is how we generally deal with things. We do activities without employing conscious symbolic reasoning, such as recognizing a face, driving ourselves to work, or locating the appropriate word to say. Instead, we appear to leap to the appropriate response without considering alternate responses. Dreyfus said the core of expertise is when our intuitions educate to the point where we disregard norms and "measure up the situation" and react.

Conclusion

AI research has advanced significantly over the last half-century and is no longer based on the primitive intelligence concepts implicit in GOFAI. Indeed, the more sophisticated modern views by Dreyfus' first critique. For example, the notion that AI will fully mirror human thought and intelligence are largely lacking. The frame issue has become highly technical in recent years, yet it remains a crucial issue. It is one of the few disciplines in which philosophers and scientists have successfully worked.