How did expert systems rise and then fall out of favor

Codifying Human Expertise Into Explicit Rules

Expert systems were a form of AI designed around a fairly intuitive idea: capture the specific knowledge and decision-making rules of a human expert in some narrow domain, encode that knowledge explicitly into a program’s rule base, and let the resulting system apply that codified expertise to help solve problems within that domain, such as medical diagnosis or industrial equipment troubleshooting.

Their Rise to Commercial Prominence in the 1980s

By the early-to-mid 1980s, expert systems had achieved significant commercial success and generated substantial investment and enthusiasm, as businesses across various industries adopted them for specific, well-defined problem domains where they could demonstrate clear, measurable value by capturing scarce specialized expertise in a reusable form.

Why Maintenance Became a Bigger Problem Than Expected

In practice, many organizations discovered that keeping an expert system’s encoded knowledge accurate and current required substantial, ongoing effort, since real-world domains continue to change and expert knowledge itself evolves — maintaining these systems proved considerably more expensive and labor-intensive on an ongoing basis than many adopters had initially anticipated when first investing in the technology.

Why Brittleness Outside Their Narrow Scope Was a Persistent Limitation

Expert systems also tended to be brittle when encountering situations outside the specific, narrow scope their rules had been designed for, often failing in ways that revealed they lacked any deeper, more general understanding of the underlying domain beyond their explicitly encoded rules — a limitation that became more apparent and more costly as organizations tried to expand these systems’ scope over time.

How This Disappointment Fed Into the Late-1980s AI Winter

As these practical limitations became more widely apparent through the 1980s, commercial enthusiasm for expert systems declined substantially, contributing significantly to the broader AI funding collapse of the late 1980s and early 1990s, alongside the related collapse of the specialized computer hardware market that had grown up to support expert-system-style AI development.

Their Legacy in Later AI Research

Despite their commercial decline, expert systems left a lasting influence on the field, contributing ideas about how to structure and represent specialized domain knowledge that would later inform other approaches, even as the field’s center of gravity shifted decisively toward data-driven machine learning methods that don’t require this kind of explicit, hand-crafted knowledge encoding.

Bottom Line

Expert systems rose to commercial prominence in the early-to-mid 1980s by codifying human expert knowledge into explicit rules for narrow problem domains, but fell out of favor by the late 1980s and early 1990s once their high maintenance costs and brittleness outside their original narrow scope became clear, contributing significantly to that era’s broader AI funding collapse.

Look Up AI Terms

Search plain-English definitions of AI and machine learning terms in our free AI Glossary.

Go deeper

• How Do You Prompt an AI to Stick to Only the Information You Give It, Without Adding Outside Knowledge?