Artificial Experts (1990)
Artificial Experts: Social Knowledge and Intelligent Machines (MIT Press, 1990) is Collins's first major book on artificial intelligence. The title's irony is central to its argument: machines could in some sense be 'expert' — they could reproduce expert outputs under constrained conditions — but the book's sustained argument is that the expertise so reproduced is not the expertise that human communities actually produce and maintain. The book established the sociological framework that Collins would refine across the subsequent three decades: expertise is a property of social groups maintaining practices, not of individuals processing information, and projects to build artificial experts by building artificial heads are therefore incomplete in principle, not merely technically limited.
In the AI Story
The book was written in response to the expert systems era of 1980s AI, when rule-based systems promised to capture domain expertise in computational form. Collins's ethnographic observations of how expertise actually worked in scientific communities revealed the gap between the rule-based model and the reality of expert practice. The book anticipated the failures of expert systems in the late 1980s and early 1990s — failures that confirmed Collins's sociological diagnosis.
The framework developed in Artificial Experts would prove remarkably durable. The fundamental claim — that expertise lives in social practice, and that machines trained on the outputs of that practice absorb only its explicit residue — applies to transformer-based language models with the same force it applied to rule-based expert systems, even though the technical capabilities of the two paradigms differ enormously. The invariance is the point: Collins's framework identifies a structural feature of human knowledge that is independent of the specific computational approach used to try to replicate it.
Origin
Published by MIT Press in 1990, the book drew on Collins's decades of ethnographic work on laser builders, gravitational wave physicists, and other scientific communities. The sociological approach was consolidated alongside Collins's parallel theoretical work in the sociology of scientific knowledge (SSK).
Key Ideas
Expertise is social. The book's foundational claim: expertise is a property of communities, not individuals.
Expert systems will fail. Collins predicted the limitations of rule-based AI because the rules could not capture what the community actually knew.
The framework invariance. The sociological critique applies to any AI paradigm that attempts to capture expertise through its explicit residue — including modern LLMs.
Empirical grounding. The book is built on ethnographic studies, not philosophical argument alone.
Appears in the Orange Pill Cycle
Further reading
- Harry Collins, Artificial Experts: Social Knowledge and Intelligent Machines (MIT Press, 1990)
- Harry Collins, Tacit and Explicit Knowledge (University of Chicago Press, 2010)
- Harry Collins and Martin Kusch, The Shape of Actions (MIT Press, 1998)