Douglas Lenat on Cyc: The Decades-Long Quest for Common Sense Reasoning in AI

This podcast episode features Douglas Lenat, creator of Cyc, a project launched in 1984 with the ambitious goal of solving the core problem of artificial intelligence: acquiring and utilizing common sense knowledge. Lenat explains that early AI systems, despite impressive initial successes, consistently hit a "brick wall" because they lacked general world knowledge and true understanding, performing tricks without grasping their underlying meaning. He likens understanding to a deep, layered foundation of knowledge that allows for robust reasoning, especially in novel or unexpected situations, contrasting it with brittle, rule-based systems that fail outside their narrow domains.

The discussion delves into the technical approach of representing common sense knowledge using formal languages like predicate logic, which enables algorithmic inference. A pivotal 1984 meeting with AI luminaries like Marvin Minsky and Alan Newell estimated the required common sense assertions to be around one million. However, this estimate proved to be an order of magnitude too low, with Cyc ultimately requiring tens of millions of such rules. Lenat recounts the historical context of the Japanese Fifth Generation Computing effort and the subsequent US response (National Cooperative Research Act, MCC) that provided the necessary long-term funding and person-centuries of effort for Cyc's development.

Lenat details the unique knowledge acquisition techniques employed by Cyc, which involve analyzing the "white space" in text—the unstated assumptions a writer expects a reader to infer—and filling in the logical gaps between sentences. Other methods include introspecting on why humorous or contradictory headlines are unbelievable, thereby revealing underlying common sense principles. A crucial architectural innovation for Cyc was the abandonment of global consistency in favor of local consistency, using "contexts" (analogous to tectonic plates) that are individually consistent but allow for inconsistencies at their boundaries. These contexts are first-class objects, arranged in a graph structure, enabling flexible and nuanced reasoning.

Ultimately, the knowledge encoded in Cyc is not merely factual data but rather "rules of thumb" or "platonic forms"—general principles that are usually true and enable deeper understanding and inference. After decades of development, Cyc has amassed its vast common sense knowledge base and is now being applied to domain-specific problems, providing a robust foundation that prevents the brittleness characteristic of earlier expert systems. This long-term endeavor highlights the profound complexity of replicating human common sense and its critical role in achieving true artificial general intelligence.

Psych is a project launched by you in 1984 and still is active today whose goal is to assemble a knowledge base that spans the basic concepts and rules about how the world Works in other words it hopes to capture Common Sense knowledge which is a lot harder than it sounds

the brick wall was the programs didn't have what we would call Common Sense they didn't have General World Knowledge they didn't really understand what they were doing what they were saying what they were being asked

I think of understanding more like a um uh think of it more like the ground you stand on which um could be very shaky could be very unsafe um but most of the time is not because underneath it is more ground

there's almost no telling what little bits of knowledge about the world you might actually need um in some situations which were unforeseen

there's a way of representing things in this formal language um which enables a mechanical procedure to sort of grind through and algorithmically produce all of the same logical entailments all the same logical conclusions that you or I would from that same set of pieces of information that are represented that way

amazingly everyone got an answer which was around a million

all of us were off by an order of magnitude um that it turns out what you need are tens of millions of these um pieces of knowledge about um every day

one of the hardest lessons for us to learn it took us about five years to to Really grit our teeth and um learn to love it um is we had to give up Global consistency

the psych knowledge base is divided up into almost like tectonic plates which are individual contexts and each context is more or less consistent but there can be small inconsistencies at the boundary between one context than the next one

what we're representing the things that we need a small number of tens of millions of are more like rules of thumb rules of good guessing things which are usually true and which help you to make sense of the facts that are on sort of sitting off in some database or some other more static story