,
10 tweets,
2 min read
Read on Twitter
The entire history of computer science can be framed from the perspective of scaling specification into robust implementation.
Attempts using analog circuitry had difficulty scaling due to accumulative errors in their iterative operations.
This was solved through the transition of digital based computation, however the original sequential behavior of algorithms remained.
Imperative computer languages were invented to allow humans to more easily express sequential algorithms.
Declarative computer languages were later invented to allow humans to express specifications without requiring knowledge of the underlying algorithms.
Expert systems were invented that leveraged declarative specifications. Unfortunately, there were scalability problems with logic rules interfering with other logic rules.
Neural networks are massively parallel rule based systems, but rather than using human-engineered algorithms to define the rules, algorithm reminiscent of analog computing is used to learn the rules.
The reason it works is that we are treating populations of simple linear approximations as ensembles resembling continuous systems. The math seems to work but it ignores the reality that underneath real neural networks are more complex processing neurons.
Nature through eons of evolution has developed advanced building blocks called cells that have the instructions to generate anything in the body. medium.com/intuitionmachi… .
How does coarse behavior program nano-intentional behavior? There are insufficient signals in the environment to guide the 'programming' of nano-intentional components. What is the information 'ratchet' that allows more complex behavior to accumulate?
