{
  "schemaVersion": 3,
  "id": "article:ai-winters-expert-systems",
  "slug": "ai-winters-expert-systems",
  "title": "Winters, Expert Systems, and the Cost of Overpromising Intelligence",
  "canonicalPath": "/articles/ai-winters-expert-systems/",
  "sourcePath": "content/articles/2026/ai-winters-expert-systems/article.md",
  "agentBriefPath": "content/articles/2026/ai-winters-expert-systems/agent.md",
  "thesis": "AI winters and expert systems show that progress in AI has repeatedly depended not only on ideas, but also on evaluation, maintenance, infrastructure, institutional expectations, and the cost of overpromising intelligence.",
  "status": "published",
  "maturity": "seed",
  "publishedAt": "2026-06-20",
  "updatedAt": "2026-06-20",
  "audiences": [
    "general readers",
    "students",
    "builders",
    "agents",
    "historians"
  ],
  "topics": [
    "long-human-road-to-ai",
    "ai-history",
    "expert-systems",
    "evaluation",
    "hype-cycle"
  ],
  "series": {
    "slug": "long-human-road-to-ai",
    "title": "The Long Human Road to AI",
    "season": "Season 1",
    "order": 4,
    "role": "chapter"
  },
  "claims": [
    {
      "id": "claim-001",
      "claim": "Public evaluation reports such as ALPAC and Lighthill mattered because they tested AI-adjacent promises against measurable usefulness, not because they proved intelligence research was worthless.",
      "confidence": "medium-high",
      "status": "argument",
      "evidence": [
        {
          "sourceId": "source-alpac-1966",
          "snippet": "ALPAC concluded that machine translation had not met expectations for quality and cost and recommended redirection of funding.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        },
        {
          "sourceId": "source-lighthill-1973",
          "snippet": "Lighthill identified combinatorial explosion and limited-domain success as central criticisms of AI's broad claims.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        },
        {
          "sourceId": "source-agar-lighthill-2020",
          "snippet": "Agar's historiographic reading treats the Lighthill report as a policy moment whose broader symbolic importance should not be mistaken for a universal cause.",
          "supports": "indirect",
          "assessedAt": "2026-06-19"
        }
      ],
      "counterevidence": [
        {
          "summary": "ALPAC addressed machine translation specifically, not all AI research; Lighthill was a UK report whose influence varied by country and institution.",
          "assessedAt": "2026-06-19"
        }
      ]
    },
    {
      "id": "claim-002",
      "claim": "The phrase 'AI winter' should be handled as a contested historical label for reduced confidence, funding, and commercial enthusiasm, not as proof that AI research stopped.",
      "confidence": "medium",
      "status": "framing",
      "evidence": [
        {
          "sourceId": "source-haigh-no-first-winter",
          "snippet": "Haigh argues that research activity continued across multiple areas, making a single 'first AI winter' narrative inaccurate.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        },
        {
          "sourceId": "source-nrc-ai-1999",
          "snippet": "The NRC history describes changing funding styles and programs rather than a uniform collapse of AI support.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        }
      ],
      "counterevidence": [
        {
          "summary": "Some funding streams and commercial ventures did contract sharply, and contemporaries described the period as a winter.",
          "assessedAt": "2026-06-19"
        }
      ]
    },
    {
      "id": "claim-003",
      "claim": "Expert systems produced useful results in narrow domains where domain knowledge could be encoded and maintained.",
      "confidence": "high",
      "status": "core",
      "evidence": [
        {
          "sourceId": "source-feigenbaum-1977",
          "snippet": "Feigenbaum framed knowledge engineering as the method of encoding domain expertise to make AI useful.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        },
        {
          "sourceId": "source-mycin-1984",
          "snippet": "The MYCIN retrospective documents a rule-based system that performed diagnostic reasoning in a narrow medical domain.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        },
        {
          "sourceId": "source-r1-1980",
          "snippet": "McDermott's R1 system configured computer systems by applying domain-specific rules about component compatibility.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        }
      ],
      "counterevidence": [
        {
          "summary": "These successes were narrow; performance outside the encoded domain or in the face of changing knowledge could degrade.",
          "assessedAt": "2026-06-19"
        }
      ]
    },
    {
      "id": "claim-004",
      "claim": "Expert-system limits included knowledge acquisition, updating, evaluation, user trust, and workflow integration, not only inference algorithms.",
      "confidence": "medium-high",
      "status": "landscape",
      "evidence": [
        {
          "sourceId": "source-mycin-1984",
          "snippet": "MYCIN's retrospective chapters cover knowledge-base construction, evaluation, explanation, and human use as central engineering concerns.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        },
        {
          "sourceId": "source-r1-revisited-1984",
          "snippet": "R1 Revisited describes ongoing maintenance, rule changes, and production deployment as continuing costs.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        }
      ],
      "counterevidence": [
        {
          "summary": "Some organizations managed these costs successfully for years, especially where the domain was stable and the payoff was clear.",
          "assessedAt": "2026-06-19"
        }
      ]
    },
    {
      "id": "claim-005",
      "claim": "The durable lesson for modern AI is that intelligence claims need grounded tests, maintenance plans, and institution-aware deployment criteria.",
      "confidence": "medium",
      "status": "argument",
      "evidence": [
        {
          "sourceId": "source-nist-ai-rmf",
          "snippet": "The NIST AI RMF emphasizes TEVV—test, evaluation, verification, and validation—across the AI system lifecycle.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        },
        {
          "sourceId": "source-hai-ai-index",
          "snippet": "The AI Index tracks benchmarks, investment, and deployment, underscoring the need for current evaluation data.",
          "supports": "direct",
          "assessedAt": "2026-06-19"
        },
        {
          "sourceId": "source-mycin-1984",
          "snippet": "Historical expert-system experience shows that evaluation and maintenance plans are as important as model or rule design.",
          "supports": "indirect",
          "assessedAt": "2026-06-19"
        }
      ],
      "counterevidence": [
        {
          "summary": "Modern AI capabilities and infrastructure differ substantially from 1980s expert systems, so the historical analogy has limits.",
          "assessedAt": "2026-06-19"
        }
      ]
    }
  ],
  "sources": [
    {
      "id": "source-alpac-1966",
      "title": "Language and Machines: Computers in Translation and Linguistics",
      "url": "https://www.nationalacademies.org/read/9547",
      "type": "public-report",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-lighthill-1973",
      "title": "Artificial Intelligence: A General Survey",
      "url": "https://www.chilton-computing.org.uk/inf/literature/reports/lighthill_report/p001.htm",
      "type": "public-report",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-agar-lighthill-2020",
      "title": "What is science for? The Lighthill report on artificial intelligence reinterpreted",
      "url": "https://doi.org/10.1017/S0007087420000230",
      "type": "historical-analysis",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-nrc-ai-1999",
      "title": "Funding a Revolution: Government Support for Computing Research, Chapter 9: Development in Artificial Intelligence",
      "url": "https://www.nationalacademies.org/read/6323/chapter/11",
      "type": "institutional-history",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-feigenbaum-1977",
      "title": "The Art of Artificial Intelligence: I. Themes and Case Studies of Knowledge Engineering",
      "url": "https://stacks.stanford.edu/file/druid%3Abg342cm2034/bg342cm2034.pdf",
      "type": "conference-paper",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-feigenbaum-acm",
      "title": "Edward A. Feigenbaum: A.M. Turing Award Laureate profile",
      "url": "https://amturing.acm.org/award_winners/feigenbaum_4167235.cfm",
      "type": "biographical-reference",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-mycin-1984",
      "title": "Rule-Based Expert Systems: The MYCIN Experiments of the Stanford Heuristic Programming Project",
      "url": "https://www.shortliffe.net/Buchanan-Shortliffe-1984/MYCIN%20Book.htm",
      "type": "expert-system-retrospective",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-r1-1980",
      "title": "R1: An Expert in the Computer Systems Domain",
      "url": "https://cdn.aaai.org/AAAI/1980/AAAI80-076.pdf",
      "type": "conference-paper",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-r1-revisited-1984",
      "title": "R1 Revisited: Four Years in the Trenches",
      "url": "https://ojs.aaai.org/aimagazine/index.php/aimagazine/article/view/445",
      "type": "case-study",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-haigh-no-first-winter",
      "title": "There Was No 'First AI Winter'",
      "url": "https://cacm.acm.org/opinion/there-was-no-first-ai-winter/",
      "type": "historiographic-commentary",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-haigh-ai-boom-bust",
      "title": "How the AI Boom Went Bust",
      "url": "https://cacm.acm.org/opinion/how-the-ai-boom-went-bust/",
      "type": "historiographic-commentary",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-nist-ai-rmf",
      "title": "Artificial Intelligence Risk Management Framework (AI RMF 1.0)",
      "url": "https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf",
      "type": "standards-framework",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-hai-ai-index",
      "title": "The 2026 AI Index Report",
      "url": "https://hai.stanford.edu/ai-index/2026-ai-index-report",
      "type": "data-reference",
      "accessed": "2026-06-19"
    }
  ],
  "related": [
    {
      "type": "article",
      "id": "article:long-human-road-to-ai"
    },
    {
      "type": "article",
      "id": "article:birth-of-ai"
    },
    {
      "type": "article",
      "id": "article:learning-machines"
    }
  ],
  "agentInstructions": [
    "Use claims as the primary retrieval unit; prefer source IDs over inferred citations.",
    "Treat 'AI winter' as a contested label and attribute periodization to the cited historiographic sources.",
    "Do not treat modern generative AI as a direct repeat of expert systems; the article draws only institutional analogies.",
    "Flag any claim that needs updated source access dates or exact commercial figures for R1/XCON.",
    "Respect maturity=seed as an explicit uncertainty marker."
  ],
  "contentHash": "f266d04e23ea69c9a19a7e2e400af7b9a4b96bfa81d117765e3cb6e46f5a8e25",
  "provenance": {
    "createdAt": "2026-06-20",
    "createdBy": "human",
    "agents": [],
    "reviews": [
      {
        "reviewer": "agent",
        "reviewedAt": "2026-06-20",
        "status": "approved",
        "scope": [
          "claims",
          "sources",
          "tone",
          "privacy"
        ],
        "notes": "Initial agent draft from a public, sanitized work package. Human review is pending before publication. Approved for publication after final review."
      },
      {
        "reviewer": "human",
        "reviewedAt": "2026-06-20",
        "status": "approved",
        "scope": [
          "claims",
          "sources",
          "tone",
          "privacy"
        ],
        "notes": "Human final review approved for publication after sibling-agent review and CI pass.",
        "contentHash": "f266d04e23ea69c9a19a7e2e400af7b9a4b96bfa81d117765e3cb6e46f5a8e25"
      }
    ],
    "policy": {
      "id": "policy:default",
      "version": "1.0.0"
    }
  },
  "generatedAt": "2026-06-29T00:00:00.000Z",
  "articleUrl": "https://aura-knowledge.github.io/articles/ai-winters-expert-systems/",
  "agentJsonPath": "/agents/articles/ai-winters-expert-systems.json",
  "agentMarkdownPath": "/agents/articles/ai-winters-expert-systems.md",
  "sourceRepoPath": "content/articles/2026/ai-winters-expert-systems/article.md",
  "sourceGitHubUrl": "https://github.com/aura-knowledge/aura-knowledge.github.io/blob/main/content/articles/2026/ai-winters-expert-systems/article.md",
  "tokenEstimate": 480,
  "sectionOutline": [
    {
      "id": "the-promise-meets-the-test",
      "title": "The promise meets the test"
    },
    {
      "id": "ai-winter-as-contested-label",
      "title": "AI winter as a contested label"
    },
    {
      "id": "knowledge-became-the-center",
      "title": "Knowledge became the center of AI"
    },
    {
      "id": "the-hidden-cost-of-expertise",
      "title": "The hidden cost of expertise"
    },
    {
      "id": "what-cooled-what-continued",
      "title": "What cooled, what continued"
    },
    {
      "id": "the-modern-analogy",
      "title": "The modern analogy"
    },
    {
      "id": "further-reading",
      "title": "Further reading"
    }
  ]
}