Saturday, March 31, 2007

Some thoughts:
what is the smallest atomic definitions for an intelligence?
What are the lowest forms or least abstract forms of data?
What are the most specific forms of data?

These all seem like different ways of asking what are the basic rules that could be built upon to create intelligence. The ultimate goal of artificial intelligence is learning. I think this definition works well because as a program can learn beyond the original intent/ability of its author lies the true nature of intelligence.

Something that is intelligently designed works well and can often be used for purposes beyond its initial intent. Maybe this is just lucky.

It seems like the most concrete thing in any computer program is a binary instruction. An opcode. This is the only thing the computer can do without having to translate. Sort of like a neuron firing in a brain. It can do it or not. Binary, like the universe. There are a few different things a program can do, depending on the hardware. But the opcodes seem to be the ultimate atomic unit.

What is the next step up? Really the sky is the limit since any modern cpu is really a touring complete computational machine. So maybe the question as stated is not very useful. How about what would a simple yet powerful next step up be?

Things like math and reading come to mind. But do we as humans break things like math and language down into smaller commands before they get down to the opcodes of reality? Is there a difference between math and language?

What is the most fundamental definition of language? What are the basic operators of language?

Same for math (see philosophy of mathematics and foundation of mathematics).. of these seems like First Order Logic (FOL) and set theory are the most promising although at first glance they seem to suffer from the same problem of any self referencing system.

See: Synopsis: Wittgenstein's Logic of Language []

See: Bound versus Free variables in context of FOL, lambda calculus