Faildex 失樂園 网站标志FAILDEX失樂園
中文

Cycorp (Cyc General-AI Vision)

The business, the turning points, and the decisions behind them.

Founded1984
Event year1994

01 / WHAT IT BUILT

What the company did

Cyc is a common-sense knowledge engineering and symbolic reasoning system started by Doug Lenat at MCC in 1984. It was developed within the MCC research consortium until the end of 1994, when it was spun out; development and commercialisation continued under independent Cycorp from 1995. Founder Doug Lenat died in 2023, but Cycorp and the Cyc inference platform continued operating as of 2026.

02 / Cycorp (Cyc General-AI Vision) BUSINESS MODEL AND OUTCOME ANALYSIS

Unfinished original AI ambition: company still active

Scope of review:Cyc’s original ambition of broad common-sense reasoning through human-curated knowledge, and its transition away from the original MCC consortium support to independent Cycorp at the end of 1994. Cyc was not shut down; 2026 is not a failure year.

Cyc began with consortium research support at MCC in 1984. At the end of 1994, Cyc left the original MCC funding framework and continued development through independent Cycorp. That transition did not mean its researchers stopped or the project closed. Manually encoding and checking vast amounts of common-sense rules could not exhaust real-world knowledge, contexts and exceptions, making ongoing maintenance costly and the original general-common-sense AI ambition difficult to realise. Doug Lenat died in 2023, while Cycorp and Cyc remained active as of 2026.

Original Funding Transition

Cyc spun out of MCC at the end of 1994, shifting away from the original consortium funding framework to independent Cycorp; this was not a loss of all human support.

Knowledge Coverage

Open-world knowledge, context, exceptions and constantly changing facts cannot be exhaustively hand-enumerated.

Knowledge Maintenance

Growing ontologies and rules require costly consistency checks and ongoing knowledge updates.

Generalisation Gap

Useful reasoning within defined domains is not equivalent to universal common-sense intelligence.

Continuing Operation

Cycorp still lists the Cyc knowledge base, inference platform and healthcare products, so no project closure is established.

03 / EVENT HISTORY

Events in the structured record

  1. Cyc research starts

    Doug Lenat began large-scale common-sense knowledge engineering at MCC.

  2. Spun out of MCC research funding

    Cyc spun out of the original MCC consortium funding framework at the end of 1994, and development continued under independent Cycorp in 1995; not a project shutdown.

  3. Founder dies

    Doug Lenat died in 2023; this did not mean Cyc ceased development.

  4. Still operating in 2026 (status only)

    Cycorp continues to market the knowledge base, inference engine and healthcare products.

04 / CONTEMPORARY REBUILD ASSESSMENT

Difficulty of rebuilding this project today

A rules-based scenario score drawn from the recorded business model and case information, not a fresh independent company investigation, investment advice or a prediction of success. Review the assessed scope and validation gates below.

Rebuild difficulty5/5Very high

An open-world common-sense engine is exceptionally hard because of ontologies, context, versions and validation.

  • Assessed scopeExamines the limitations of Cyc's original broad common-sense AI ambition, including knowledge coverage, maintenance, reasoning and economics. Cycorp and Cyc remain active in 2026; this is not a business-shutdown case.
  • Technical delivery constraintsAn open-world common-sense engine is exceptionally hard because of ontologies, context, versions and validation.
  • Minimum viable testStart with one auditable domain task and specify provenance, exception handling, adversarial tests and human approval.
Scalability2.5/5Limited

Vertical healthcare or supply-chain reasoning is more reusable, though expert reviews and updates increase marginal cost.

  • Reusable scope and scale constraintsVertical healthcare or supply-chain reasoning is more reusable, though expert reviews and updates increase marginal cost.
  • Validation before scalingStart with one auditable domain task and specify provenance, exception handling, adversarial tests and human approval.

Overall assessment: This assesses Cyc’s original broad common-sense AI ambition, not a defunct company. Start with one auditable domain task and specify provenance, exception handling, adversarial tests and human approval. Scores are scenario estimates of a 2026 rebuild, not historical financial facts or success probabilities.

Assessment date · 2026-11-15

05 / CAUSAL CHAIN

How the factors connect

  1. 01

    The end-of-1994 milestone concerns the end of original MCC support and Cycorp continuation; it is not a 2026 failure or a loss of all human support.

  2. 02

    Distinguish an unfinished general-AI ambition from an operating specialist reasoning platform.

  3. 03

    Bounded, high-value, verifiable domain knowledge is more tractable than exhaustive common sense.

  4. 04

    RAG, LLMs, graphs and symbolic validation may complement each other, but generated content must not silently become trusted facts.