Douglas B. Lenat

CYC: a large-scale investment in knowledge infrastructure

Since 1984, a person-century of effort has gone into building CYC, a universal schema of roughly 105 general concepts spanning human reality. Most of the time has been spent codifying knowledge about these concepts; approximately 106 commonsense axioms have been handcrafted for and entered into CYCs knowledge base, and millions more have been inferred and cached by CYC. This article examines the fundamental assumptions of doing such a large-scale project, reviews the technical lessons learned by the developers, and surveys the range of applications that are or soon will be enabled by the technology.

On the thresholds of knowledge

Three major findings in the domain of artificial intelligence are articulated. The first is the knowledge principle, which states that if a program is to perform a complex task well, it must know a great deal about the world in which it operates. The second is a plausible extension of that principle, called the breadth hypothesis, which states that there are two additional abilities necessary for intelligent behavior in unexpected situations: falling back on increasingly general knowledge, and analogizing to specific but far-flung knowledge. The third finding is a concept of AI as an empirical inquiry system requiring the experimental testing of ideas on large problems. It is concluded that together these concepts can determine a direction for future AI research.<<ETX>>

The nature of heuristics

Abstract Builders of expert rule-based systems attribute the impressive performance of their programs to the corpus of knowledge they embody: a large network of facts to provide breadth of scope, and a large array of informal judgmental rules (heuristics) which guide the system toward plausible paths to follow and away from implausible ones. Yet what is the nature of heuristics? What is the source of their power? How do they originate and evolve? By examining two case studies, the am and eurisko programs, we are led to some tentative hypotheses: Heuristics are compiled hindsight, and draw their power from the various kinds of regularity and continuity in the world; they arise through specialization, generalization, and—surprisingly often—analogy. Forty years ago, Polya introduced Heuretics as a separable field worthy of study. Today, we are finally able to carry out the kind of computation-intensive experiments which make such study possible.

Enabling agents to work together

l Paradigm 1: Competence emerges from a large number of relatively simple agents integrated by some cleverly engineered architecture. The choice of architecture is the make-or-break theoretical part of this; the detailed characteristics of the implementation of the architecture (and the algorithms that crawl around it) are the make-orbreak pragmatic parts. The archetype of this paradigm is SOAR [61; its forerunners were the early “pure production systems.”

The Role of Heuristics in Learning by Discovery: Three Case Studies

As artificial intelligence (AI) programs are called upon to exhibit increasingly complex behaviors, their builders are faced with the growing task of inserting more and more knowledge into the machine. One long-range solution is for the program, by itself, to learn via discovery. The first case study presented, am, demonstrates that new domains of knowledge can be developed mechanically by using heuristics. Yet as new domain concepts, facts, and conjectures emerge, specific new heuristics, or informal judgmental rules, are needed. They in turn can be discovered by using a body of heuristics for guidance. The second case study, eurisko, has already achieved some promising results in this endeavor. If this process—using heuristics to guide “learning by discovery”—is so powerful and simple, one wonders why, for instance, nature has not adopted an analogous mechanism to guide evolution. Indeed, the final part of the article is a speculation that evolution does function in that manner. In place of the conventional Darwinian process of random mutation, we hypothesize a more powerful plausible generation scheme.

Beings: knowledge as interacting experts

Knowledge may be organized as a community of intfiacting modules Each module is granted a complex stiucture, to simulate a particular expert in some small domain An extended analogy is drawn to a group of cooperating human specialists Based on this, an internal constraint is imposed on the modules. Then structure must be standard over the entire community Some advantages of a uniform formalism are thereby preserved. An experimental community was implemented for the task domain of automatic programming. It has managed to synthesize a few inductive inference LISP programs, nonformally. from specific restricted dialogues with a human user.

Cyc: a mid-term report

After explicating the need for a large commonsense knowledge base spanning human consensus knowledge, we report on many of the lessons learned over the first five years of attempting its construction. We have come a long way in terms of methodology, representation language, techniques for efficient inferencing, the ontology of the knowledge base, and the environment and infrastructure in which the knowledge base is being built. We describe the evolution of Cyc and its current state and close with a look at our plans and expectations for the coming five years, including an argument for how and why the project might conclude at the end of this time.

Common sense reasoning – from cyc to intelligent assistant

Semi-formally represented knowledge, such as the use of standardized keywords, is a traditional and valuable mechanism for helping people to access information. Extending that mechanism to include formally represented knowledge (based on a shared ontology) presents a more effective way of sharing large bodies of knowledge between groups; reasoning systems that draw on that knowledge are the logical counterparts to tools that perform well on a single, rigidly defined task. The underlying philosophy of the Cyc Project is that software will never reach its full potential until it can react flexibly to a variety of challenges. Furthermore, systems should not only handle tasks automatically, but also actively anticipate the need to perform them. A system that rests on a large, general-purpose knowledge base can potentially manage tasks that require world knowledge, or “common sense” – the knowledge that every person assumes his neighbors also possess. Until that knowledge is fully represented and integrated, tools will continue to be, at best,idiots savants. Accordingly, this paper will in part present progress made in the overall Cyc Project during its twenty-year lifespan – its vision, its achievements thus far, and the work that remains to be done. We will also describe how these capabilities can be brought together into a useful ambient assistant application. Ultimately, intelligent software assistants should dramatically reduce the time and cognitive effort spent on infrastructure tasks. Software assistants should be ambient systems – a user works within an environment in which agents are actively trying to classify the users activities, predict useful subtasks and expected future tasks, and, proactively, perform those tasks or at least the sub-tasks that can be performed automatically. This in turn requires a variety of necessary technologies (including script and plan recognition, abductive reasoning, integration of external knowledge sources, facilitating appropriate knowledge entry and hypothesis formation), which must be integrated into the Cyc reasoning system and Knowledge Base to be fully effective.

CYC, WordNet, and EDR: critiques and responses

I applaud Millers WordNet project and feel that there is much in common in our approaches, even though there are fundamental differences in the two expressions of that spirit. Here, I list the four differences I noted, closing with a crucial observation concerning the common spirit in our work.

PRINCIPLES OF PATTERN-DIRECTED INFERENCE SYSTEMS

The general class of pattern-directed inference systems (PDISs) is defined and its general properties are investigated. Within a taxonomy of PDISs, the special properties of rule-based systems and two subclasses, production systems and transformation systems, are considered in detail. A comparison between these PDIS properties and contemporary information processing conceptions of human cognition suggests that PDISs provide an excellent basis for cognitive modeling. Principles for knowledge representation and system architecture are developed that suggest guidelines for controlling the combinatorial explosion that will accompany efforts to implement intelligent systems with extensive amounts of knowledge.