Cyc

What Cyc is

Cyc is a knowledge base and ontology — an ontology being a structured collection of concepts, terms and the relations among them. It is meant to cover the basic concepts and rules about how the world works, with the emphasis on implicit common-sense knowledge: the background people rely on without ever stating it, such as the fact that a cup turned upside down spills what is in it, or that a dropped object falls. The aim is to assemble such concepts and rules comprehensively rather than for one narrow task.^[1]

Origins

Cyc began in July 1984, when Douglas Lenat started the project at the Microelectronics and Computer Technology Corporation (MCC).^[1] MCC was a research consortium launched by two corporations based in the United States, set up to counter the Japanese "fifth-generation" artificial intelligence effort.^[1] Lenat was a Principal Scientist at MCC, and he called the work he began there in 1984 ontological engineering.^[2] In 1980 he had been one of the co-founders of AAAI and of Teknowledge.^[2] He received the biennial IJCAI Computers and Thought Award and was named the first Fellow of the Association for the Advancement of Artificial Intelligence.^[2]

From MCC to Cycorp

At the end of 1994 the project spun out of MCC and became Cycorp, Inc., with Lenat as its chief executive.^[2] From January 1995 Cyc was under active development at Cycorp, where Lenat was CEO.^[1] The undertaking has been expensive: by 2002 Cyc was described as having consumed 60 million dollars and 600 person-years of effort.^[1]

Knowledge written by hand

Most of the knowledge base was created by hand, as axioms — statements the system treats as true and can reason from. In 1994 it held about 1 million axioms; as of 2017, about 24.5 million.^[1] The ontology, the vocabulary of terms those axioms are written in, grew from about 100,000 terms in 1994 to about 1,500,000 terms as of 2017.^[1] The two figures move together: adding a term means adding the definitions and rules attached to it, and over that period both grew by more than a factor of ten.

OpenCyc

A public edition, OpenCyc, made part of the knowledge available outside the company. The first version was released in spring 2002 and contained 6,000 concepts and 60,000 facts.^[1] OpenCyc 4.0, released in June 2012, contained 239,000 concepts and 2,093,000 facts, mainly taxonomic assertions — statements that place a concept in a hierarchy beneath a broader one.^[1] Even at that size, the open edition is a small fraction of the full knowledge base.

Place in the development of AI

Cyc stands for one bet about how machine intelligence should be built: that reasoning needs a large body of background knowledge written down explicitly, and that people can write that knowledge. The hard part, on this view, is not the reasoning machinery but the missing common sense that a system needs before it can interpret an ordinary sentence or situation. Cyc is also a record of what hand-built knowledge engineering costs — sustained work since 1984, and by 2002 already 600 person-years and 60 million dollars.^[1] By 2017 the knowledge was still being accumulated largely through hand-written axioms, with the ontology and the axiom count both far larger than in the project's first decade.^[1]