Moving!!

I'm currently in the process of combining my blogs into a single blog and moving to my own domain name. This site will exist for a while, but further updates can be found at www.mattsjourney.com. All posts dealing with artificial intelligence can be found under the label 'Artificial Intelligence'.

Axiomatic System Followup

I spent some more time thinking about complex adaptive systems being, at their base level, axiomatic systems.  As axiomatic systems are defined, this cannot be true.  There is no internal application within axiomatic systems.  Even though mathematics is n axiomatic system, all applications come from outside the system.  Another issue is that axiomatic systems are not adaptive at all.  Kinda hurts that I missed out on that big clue.

So a new question:  can we enhance axiomatic systems (change their definition) to make them adaptive and make them externally applicable, yet keep their fundamental concepts intact?  Can we take them to a new level for both theory and application?

Bottom-Up AI

A recent commenter (yes, people do comment here, on occasion) said that building AI from the bottom up is the only way to go.  I spent some time thinking about that, and I have to agree.  Then I started expanding my thinking to include all artificial complex adaptive systems (ACAS).  If I wanted to build an ACAS, I would HAVE to start from the bottom, then test from the top down. 

This sounds like an axiomic system.  Create a set of axioms (building blocks) and see if you can reach a set of higher-level principles with those axioms.  If not, you might have bad axioms, bad operators to use on those axioms, or just a bad goal in general.  Even reaching the principle might not prove anything, for the same reasons.

The downside (the only one I can see so far) is Godel's Incompleteness Theorem.  The mind is self-referential.  Godel shows use that axiomic systems run into problems when they become self-referential.

Overall, this sounds like a promising avenue to explore.  Can any complex adaptive system be translated into an axiomic system?  If so, how does Godel come into play?  Is it even relevant?

AI-Completeness Hypothesis - Second Corollary

Hypothesis:

Any true artificial intelligence technology is AI-complete.

First Corollary:

Any approximation of a true AI technology is not necessarily AI-complete as it is not true AI.

Second Corollary:

The total set of AI-complete technologies will be the necessary and sufficient components of a complete AGI.

Open Versus Closed Mind

Let's look at two extremes.  Mind A is a completely open mind; it weighs all input as equal no matter what logic, experience, or anything else says, leading to extreme cognitive dissonance.  Mind B is a completely closed mind; unless new input conforms to what the mind already knows, it will ignore it, leading to no cognitive dissonance.

I have never heard of anyone at either extreme, but I have met people very close.  Let's create a range of -1 to 1, where a score of -1 relates to a completely closed mind and a score of 1 relates to a completely open mind.  We'll call this the mind's openness score.  Those in the lower 15% or so we'll call dogmatics; they are set in their ways and nothing will ever change them.  Those in the higher 15% or so we'll call sheeple; they will change their mind every time someone gives them new information.

Question 1:  Is the openness score fixed for a given mind?
Question 2:  If the score is fixed, which value is optimal for a mind?  What is the balance needed to allow a mind to stick to their values, beliefs, and goals but open enough to change them given the right input?
Question 3:  If the score is dynamic, what causes a change?  Is it age?  Is it the input itself?  What type of input causes larger or smaller changes?

This strikes me a bit like a bias in an artificial neural network. 

A Balance to Complexity

I was reading an email from Ben Goertzel, a major proponent and researcher in artificial general intelligence (AGI), and something that he said stuck out to me. 

"The central problem [of AI/AGI] is coordinating various components together in a synergetic way, so that they can help each other avoid combinatorial explosion, rather than exacerbating each others' problems via 'error magnification'"

The issue with many AI projects, such as creating a common sense engine, is the fact that you need a massive number of interacting objects (such as rules) to make anything realistic.  Earlier today, I saw a TED Talk about simplicity, and how few scholars are looking into it, while much was known about complexity.

When putting those two together, I started thinking about balancing the rules of complexity with rules of simplicity.  There needs to be a balance to allow for a massive number of interactions while not creating a combinatorial explosion.

AI Completeness Hypothesis

I've read up a bit on AI-completeness and want to put forth a hypothesis:

Any true artificial intelligence technology is AI-complete.

This means that any truly intelligent/conscious feature of an AI will only be developed when a true AI is developed.  It also means that chess (the way current programs 'play' it) is not true AI.  I also wonder how this would affect the Turing Test. 

A corollary would be:

Any approximation of a true AI technology is not necessarily AI-complete as it is not true AI.

An example is, again, how chess is currently handled.  A computer can be 'great' at chess simply because it can test possible plays faster than a human can.  This is an approximation of playing chess intelligently.

How would we go about proving this?  The first step is to create formal definitions of AI-complete, true AI, artificial intelligence technology, and an approximation of a true AI.