zlacker

Before you try to encapsulate common sense in a collection of obvious facts, you might want to start with a simpler task: Collect a database of elementary-school math problems and their associated answers until the database is so large that it coalesces into a fully functioning CPU.

If this seems like a bad idea, it's because it is.

>>goodsi+(OP)
I'm not convinced by Cyc either, but this argument is a bit off. A system like this would need to define axioms and inference rules. But it wouldn't need to collect upfront a database of every possible inference which might be made from them.

(In mathematics you don't technically need very many axioms or inference rules, although you do need a large body of heuristics and hints if you want a proof system which confines itself to proving inferences that are actually useful in some sense of the word, rather than proving a combinatorial explosion of trivial theorems. Dealing with the body of elementary arithmetic problems, however, wouldn't necessarily be particularly intractable -- last time I checked software proof systems can already deal with proving theorems in areas of mathematics like this and a fair bit further.)

>>goodsi+(OP)
You wouldn't believe it, but principal Cyc developer Douglas Lenat started almost exactly with that. Eurisco, a predecessor to Cyc, was a deduction machine that operated with minimal human intervention over minimal sets of number theory axioms. It rediscovered many theorems on its own.

Douglas Lenat: http://en.wikipedia.org/wiki/Douglas_Lenat Eurisco: http://en.wikipedia.org/wiki/Eurisko

>>mjw+V2
Maths is a self-consistent and closed system. From that point of view, the nature of axioms is irrelevant. You can have an unlimited kinds of axiomatic systems, each consistent and each explanatory in its own way. What is a theorem in one system can be treated as an axiom in the next as long as the system itself leads to no contradictions.

Given that human knowledge encompasses more than one axiomatic system, it would be foolish to endow a system designed to replicate human knowledge with an immutable set of axioms.

Please watch this presentation of Richard Feynman on the nature of maths and physics: http://www.feynmanphysicslectures.com/relation-of-mathematic...

>>thesz+37
Lenat never released the source code to Eurisko. There has always been a question as to how much he did and how much the program did.

Ideas that Lenat promoted with Eurisko and Cyc have made him successful by most criteria. But it would be better had he published a proven finished product, including it's innards. My feeling is that, in publishing, one should show the code or it never happened.

>>goodsi+(OP)
Your argument has a fundamental flaw; the information content of a large set of elementary math problems is minimal, the information content of obvious facts is much, much higher. A few hundred bytes in any decent programming language can generate the first on demand in a fraction of a second, if you can do that for the second set you deserve every accolade you will receive. It is not at all obvious that any property of the first will apply to the second.

I think Cyc is a joke, too, but your argument doesn't hold.

>>jerf+le
The information content of elementary school math problems is quite high. They contain lots of names of hypothetical multi-ethnic children, statements about adding and taking away apples, amounts of money needed to purchase carpeting for rooms of particular dimensions, etc. Excluding this information because you know in advance it's useless for building a CPU is cheating, since it uses your knowledge of what a functioning adding machine looks like.