🚀 go-pugleaf

This post is largely thinking-out-loud about various
related topics, leading up to some hypotheses about the
early evolution of the nervous system.

I'm not sure if the info-theorists have done much with
the theory of _handles_-- small info-chunks that are
offered to a listener who can then use them to access
larger info-chunks.

Chip design has quickly evolved from 8-bit to 16-bit
to 32-bit handles, with a sort of critical mass being
achieved at that point, especially in the 80386 and
its successors.

Phonetic alphabets reduced the bits-per-letter from
8 or ten to 5 or 6, with average words needing only
about 32 bits.  The number of words in a typical
vocabulary is around 64k, I think, so 16 bit handles
will be barely adequate.

Bitmapped screens become usable with a mouse at
something like a million bits, and icons work best
at 1000 or so (ie 32*32).  Amino acids can be
identified with 5 bits.

Links in hypertext are handles, and I've argued that
in most circumstances one-word 'text buttons' are
the most efficient form of link.  MS-DOS burdened
early versions of Windows with 8-letter file-handles.

But hierarchical info-systems like Roget's thesaurus
must be seen as using a progressive bit-length-- the
first tree diagram was Porphyry's, which progressed
ideally from one bit to two, then three, then four
(spirit-body, mineral-living, plant-animal, beast-
human).  (The I Ching had used an idealised 6-bit
system in its divinations a millennium earlier.)

As nervous systems evolved, such a progressive
notation scheme must have been worked out by trial
and error, exploring neural 'ratios' between inputs
and outputs.

The organism itself was embedded in a world of
_stories_, and from its point-of-view some were
comedies (helping it thrive) and some tragedies
(causing suffering and death).

Neural mutations changed the overall balance of
comedies and tragedies, and were selected if they
increased the comedy-score more than the tragedy-
score.

Coarse 'ratios' were fine-tuned by additional
mutations, but sometimes these complexes would
become too confused, and great simplifications
would have been needed, replacing Rube-Goldberg
adaptions with something closer to the natural
order.

Where possible, neural 'homonyms' might have been
exploited, in which similar patterns in different
sense modalities would take advantage of similar
output 'ratios'.

But all of this had to be built on a foundation
of semantic sand, in which patterns didn't have
any 'positive' definition, but only a series of
mixed positive-and-negative effects...


500k timeline of knowledge representation:
 http://www.robotwisdom.com/ai/timeline/0000.html