Thursday, January 12, 2012

In the beginning, there was no meaning, only variation...

What if, in the beginning, there was only "1" and "0"? And there was no meaning. What if, as luck (and a lot of time) would have it, there was "remembering"? Then, it was possible to have a relationship between 1 and 0 and observation at time "t" and time "t+epsilon." Relationships that occurred "more often" (repetitions) and/or "close together" (small epsilon) were remembered/refreshed/survived more than relatinoships that occurred "less often" and/or "far apart." In this system, remembered patterns survive. I will postulate that "meaning" and "purpose" are derived only from these surviving patterns.


So...


To build a super-brain, or something that does what the human brain does, only better and faster, requires the same development process, the same meaning and purpose, as the system that developed the human brain. Note: it is also possible to develop other forms of intelligence using other systems of development, but it is unlikely the human brain would understand or be capable of finding meaning or purpose in such a foreign nature of intelligence.


So...


What does a computer program look like that uses the same system of development?


I need something that can distinguish between 1 and 0. I will make the input for this development system the Internet. Note: I could build other machines, other programs, that used other inputs, but the challenge is how intelligent, how far, could these other systems develop? They would be limited by the variety (see Ashby) of the input. The Internet is the first "sense organ" with sufficient variety to develop intelligence in a reasonable amount of time and over a sufficiently large domain of space.


So...


I connect my computer to the Internet and deliver some input to my program. Note: the choice of what input I deliver to the program is already limiting the evolution of the pattern memory (meaning and purpose) of the program. But it is not such a limitation as I might first think. I will have several (maybe hundreds or thousands) of programs, each specializing in observing a particular type of information on the Internet, similar to the human senses collecting different perspectives on the environment of the body. Then, I will need to connect the outputs of these different agents (a la Minsky), through more layers of pattern survival, to provide a super-intelligence that brings everything together.


So...


What domains of intelligence do I want to develop for this super-intelligence?


1. Ability to communicate with humans (or else, how will I communicate with it?)
2. Ability to search for answers
3. Ability to remember in a retrievable way
4. Ability to float hypotheses over "long" periods of time (especially longer than human lifespan, but perhaps even longer than organization lifespan).
5. Ability to learn new things (and forget old things?)


more to come...


So...


Some ideas for Internet domains:


1. Blog web sites
2. Economic web sites (allocation of goods and services)
3. Published research web sites (domain of new and formative ideas)
4. Educational web sites (domain of old and accepted ideas)
5. Image web sites
6. Sound web sites
7. Rapidly changing web sites (news web sites?)
8. Personal web sites
9. Organizational/institutional web sites


more to come...


Now, more on the mechanism of learning pattern recognition...


A. Observe a "1" at moment "t(a)"
B. Remember a "1" paired with a "t(a)"
C. Observe a "1"
D. Remember a "1" paired with a "t(b)"
E. Remember a "1" "1" paired with a t(a,b)
F. The pattern "1" "1" decays at a rate that allows for forgetting (non-survival) of ideas.


How long until the system recognizes the distribution of ASCII characters (ie-has found meaning in the observed input)?


How long until the system recognizes repeated combinations of ASCII characters (words)?


How long until the system recognizes repeated combinations of words?


Etc...


More to come...