Re: Putnam's kind of realism

Eliezer S. Yudkowsky (
Wed, 03 Nov 1999 12:54:13 -0600

It sounds to me like Putnam, or the person explaining Putnam, or someone, is failing to clearly distinguish between the question "What is truth?" and "What is rational?" The truth precedes us, generated us, and acts according to laws of physics which we cannot specify. There is an objective answer to the question "What is truth?". Rationality is the process whereby we attempt to arrive at the truth. There is no objective answer to the question "What is rational?", not *here*, not without direct access to objective reality. Rather, "How should rationality work?" is an engineering question about how to create systems that can model and manipulate reality - or rather, how to create parts of reality whose internal patterns mirror the whole, to the point that the internal process can predict the external processes in advance.

As for AI, it seems to me that the concept of an internalist mental model is a confusion between "is" and "should". A reference to "green"
*is* the cognitive concept of "green", but what it *should* be, what the
system tries to make it converge to, is the external reality of green.
If you have a wholly internalist system, then the concepts don't converge to anything. If the only definition of correctness is the thought itself, there's no way to correct malfunctions. The system is meta-unstable. The AI thinks "I can't possibly be wrong; anything I think is by definition correct," and then it gets silly, just like subjectivist humans. Shades of Greg Egan's _Quarantine_.

           Eliezer S. Yudkowsky
Running on BeOS           Typing in Dvorak          Programming with Patterns
Voting for Libertarians   Heading for Singularity   There Is A Better Way