GOFAI (good old-fashioned artificial intelligence) is a term in the philosophy of artificial intelligence referring to classical symbolic AI, as opposed to other approaches such as neural networks, situated robotics, narrow symbolic AI, or neuro-symbolic AI. The term was coined by philosopher John Haugeland in his 1985 book Artificial Intelligence: The Very Idea. Haugeland introduced the term to address two central questions: whether GOFAI can produce human-level artificial intelligence in a machine, and whether GOFAI is the primary method that brains use to display intelligence.
In AI development and technology, GOFAI is used to refer to programs built with deliberate, explicit instructions for a single task, in contrast to approaches that use machine learning. Examples of GOFAI applications include AlphaGo and Apple's initial Siri design.
Historical Context and Influence
AI research in the 1950s and 1960s had an enormous influence on intellectual history. It inspired the cognitive revolution, led to the founding of the academic field of cognitive science, and was the essential example in philosophical theories of computationalism, functionalism, and cognitivism, as well as psychological theories of cognitivism and cognitive psychology. The specific aspect of AI research that led to this revolution was what Haugeland called GOFAI.
Herbert A. Simon, a founder of AI, speculated in 1963 that GOFAI could produce human-level intelligence and that it was the primary method brains use. His evidence was the performance of programs he co-wrote, such as Logic Theorist and the General Problem Solver, along with his psychological research on human problem solving.
Western Rationalism and GOFAI
Haugeland places GOFAI within the rationalist tradition in western philosophy, which holds that abstract reason is the highest faculty, separates humans from animals, and is the most essential part of intelligence. This assumption appears in Plato and Aristotle, in Shakespeare, Hobbes, Hume, and Locke, and was central to the Enlightenment, the logical positivists of the 1930s, and the computationalists and cognitivists of the 1960s.
Symbolic AI in the 1960s successfully simulated high-level reasoning, including logical deduction, algebra, geometry, spatial reasoning, and means-ends analysis, all in precise English sentences. Many observers, including philosophers, psychologists, and AI researchers, became convinced they had captured the essential features of intelligence. This conviction was entailed by rationalism; if false, it would question a large part of the western philosophical tradition.
Continental philosophy, including Nietzsche, Husserl, and Heidegger, rejected rationalism, arguing that high-level reasoning is limited and prone to error, and that most abilities come from intuitions, culture, and instinctive feel. Philosophers familiar with this tradition, such as Hubert Dreyfus and Haugeland, were the first to criticize GOFAI and its sufficiency for intelligence.
Haugeland's Definition
Critics and supporters of Haugeland's position have found it difficult to define GOFAI precisely. Drew McDermott, for example, finds Haugeland's description incoherent and argues that GOFAI is a myth. Haugeland coined the term to examine the philosophical implications of claims essential to all GOFAI theories, which he listed as: (1) our ability to deal with things intelligently is due to our capacity to think about them reasonably, including subconscious thinking; and (2) our capacity to think about things reasonably amounts to a faculty for internal automatic symbol manipulation.
This is similar to the physical symbol systems hypothesis proposed by Herbert A. Simon and Allen Newell in 1963, which states that a physical symbol system has the necessary and sufficient means for general intelligent action. It also resembles Hubert Dreyfus's psychological assumption that the mind can be viewed as a device operating on bits of information according to formal rules.
Haugeland's description refers to symbol manipulation governed by a set of instructions. The symbols are discrete physical things assigned definite semantics, like <cat> and <mat>. They do not refer to signals, unidentified numbers, or matrixes of unidentified numbers. Thus, Haugeland's GOFAI does not include techniques such as cybernetics, perceptrons, dynamic programming, control theory, or modern techniques like neural networks or support vector machines. It also does not include systems combining symbolic AI with other techniques, such as neuro-symbolic AI, or narrow symbolic AI systems designed only for specific problems.
Responses from AI Scientists
Russell and Norvig wrote that GOFAI corresponds to the simplest logical agent design and that it is difficult to capture every contingency of appropriate behavior in a set of necessary and sufficient logical rules, calling this the qualification problem. Later symbolic AI work after the 1980s incorporated more robust approaches to open-ended domains, such as probabilistic reasoning, non-monotonic reasoning, and machine learning. Currently, most AI researchers believe deep learning and, more likely, a synthesis of neural and symbolic approaches (neuro-symbolic AI), will be required for general intelligence.
Legacy and Relevance
GOFAI remains a foundational concept for understanding the history and philosophy of artificial intelligence. It highlights the shift from rule-based, explicit reasoning systems to data-driven approaches like large language models and transformers. While GOFAI's dominance waned with the rise of machine learning, its influence persists in areas like expert systems, automated reasoning, and the ongoing exploration of neuro-symbolic methods. The term continues to serve as a critical reference point for debates about the nature of intelligence and the future of AI research.