AI Computers are being built on modern (read: ancient) architecture whereby processing is considered as a single master function. Supercomputers used for AI are built as all computers are built: parallel processing of all functions in a coordinated architecture controlled by an effective single processor.

The solution is simple: AI computers do NOT rely on processing. They are dependent more upon memory. Just like our human minds, our decision making is actually a side effect of our stored memories. So processing is less important than memory. Memory must therefore be compartmentalized and processed separately and independently.

This means a completely new approach to processing as it will require separating out the functions for memory recall and setting up a single processor which is dedicated to organizing memory (I am able to know how this works because my center for organizing memory was damaged) and which merely organizes separate processors which in turn organize their own categories of memories.

For the system I envision in these initial thoughts on the subject (I just conceived the idea about 5 minutes ago) I would have a hierarchy of types of memory organized by need and logical category (i.e. critical knowledge, social knowledge, spatial, conceptual, etc.) with each category being represented by its own processor.

What this does is turn computer programming completely on its side. Instead of organizing programs to operate with hardware, we are organizing hardware to work with programs. Processes are not dedicated to single processors but spread across multiple processors. For example, if the AI is asked to speculate on the nature of a video of what turns out to be a rocket launch, the main processor would instruct the appropriate processors to make data available relevant to answering the question and other processors to assemble and analyze that information as appropriate in order to provide the main processor with "expertise" on the categories and types of memories needed to answer the question.

This example necessitates that there also be processors for doing "thinking" such as three dimensional assembly and disassembly for "visualization" - in the example above a dedicated math processor would provide hard numbers for scenarios of potential altitude/speed models which the main processor would then be able to use in its analysis of other data such as observed speed, brightness, and what those would be of likely objects and situations.

Basically, you have to design computers to work how the actual human brain works, not how you people are trained to use yours. You all think in terms of a single central identity that your conscious self controls. You fail to realize you are a bunch of separate processes working together (most of the time anyway.)

Comments are not available for this entry.