Cycorp Overview
How much of what you know is common sense? You can burn yourself if you touch a hot light bulb. A cold shower can wake you up. A mountain is bigger than a mole-hill. Is 50% of knowledge common sense? 90%? 99.99%? According to Push Singh at MIT, several attempts to benchmark the scope of common sense put it in the order of hundreds of millions of rules.
Hard to believe that you could keep several hundred of anything in your brain, let alone several hundred million. Yet, there they are, right behind your eyes. The rules that you live by. So what will it take for a computer to learn those rules? Sure, itÂs the stuff of science fiction [think HAL or C-3PO], but it is also the stuff of science fact for a couple of below the radar software companies and research projects.
Cycorp, based in Austin, Texas is a 20-year-old research project turned start up turned government funded technology vendor. Cycorp was started as a research project in 1984 by Doug Lenat, then a Stanford professor. Lenat moved to Austin and his project took up residence at MCC. In 1994, Lenat spun out of MCC and formed Cycorp as a for-profit venture. CycÂs original goal was to Âcodify in machine readable format the rules that make up common sense. We will give Cyc a report card at the end of this article
Rules, rules, rules
Cyc has three main technology concepts that merit understanding. The first is the knowledge base of rules  or assertions - that have been entered into Cyc. The rules that Cyc knows about have generally been hand entered, and are grouped around specific key words. Several hundred thousand keywords each have 10 plus assertions entered about them. For example, the key word dog might have assertions such as dogs are mammals. Dogs can be pets. Dogs are color-blind etc. Assertions are furbundleddeled into related concepts called microtheories.
Twenty years, and a reported $60 million in government and venture backing later, Cyc appears to be making some progress. According to company press releases, Cyc has assembled a knowledge base of over 3 million rules of thumb. While this is only of sliver of human common sense, what is important, is that this knowledge has been successfully pointed at solving business problems.
Hard to believe that you could keep several hundred of anything in your brain, let alone several hundred million. Yet, there they are, right behind your eyes. The rules that you live by. So what will it take for a computer to learn those rules? Sure, itÂs the stuff of science fiction [think HAL or C-3PO], but it is also the stuff of science fact for a couple of below the radar software companies and research projects.
Cycorp, based in Austin, Texas is a 20-year-old research project turned start up turned government funded technology vendor. Cycorp was started as a research project in 1984 by Doug Lenat, then a Stanford professor. Lenat moved to Austin and his project took up residence at MCC. In 1994, Lenat spun out of MCC and formed Cycorp as a for-profit venture. CycÂs original goal was to Âcodify in machine readable format the rules that make up common sense. We will give Cyc a report card at the end of this article
Rules, rules, rules
Cyc has three main technology concepts that merit understanding. The first is the knowledge base of rules  or assertions - that have been entered into Cyc. The rules that Cyc knows about have generally been hand entered, and are grouped around specific key words. Several hundred thousand keywords each have 10 plus assertions entered about them. For example, the key word dog might have assertions such as dogs are mammals. Dogs can be pets. Dogs are color-blind etc. Assertions are furbundleddeled into related concepts called microtheories.
Twenty years, and a reported $60 million in government and venture backing later, Cyc appears to be making some progress. According to company press releases, Cyc has assembled a knowledge base of over 3 million rules of thumb. While this is only of sliver of human common sense, what is important, is that this knowledge has been successfully pointed at solving business problems.
posted by jonas at
7:09 PM
0 Comments:
Post a Comment
<< Home