Scalable GI as comparison-first pixels-up hierarchical connectivity clustering

The Northwest AGI Forum welcomes Boris Kazachenko. Boris has been a fixture in the independent AGI community for over a decade.

MAIN TALKING POINTS

- General intelligence as unsupervised learning, AKA pattern discovery. Conventional definitions in terms of ability to perform undefined tasks and cargo-culting human mind or behavior are not constructive.

- Effective generality is perception/action patterns, they must be learned bottom-up vs. designed top-down as in “integrative” schemes.

- Consistent implementation: pipelined pattern composition hierarchy. Each level cross-compares inputs, then clusters them by the resulting derivatives. The derivatives include match, defined as a measure of compression. Projected match (predictive value) is a fitness function of the system. Clusters of each level are incrementally composed generalized representations, which become higher-level inputs.

- Contrast with conventional clustering, ANN and BNN, complex but efficient intelligent design vs. evolved brute-forcing

Click here for a detailed introduction and comparison to NN/DL:

http://www.cognitivealgorithm.info/

Github:
Wiki
System Diagram