CNS Information

Suppose I monitor a major sensory nerve, of type unknown. Would I be able to determine the information in that nerve’s activity without knowing the source or destination of that activity? If line labeling is the correct theory of information content and transport through the specific nervous system then I can not. Perhaps the quantity of excitation or some temporal pattern variations will prove to be a clue, to determine, e.g., whether the signal is auditory or visual. But I can’t know much unless I know the source or the destination of the sigmal.

IMPD again, there is no “you” inside your head, and the above consideration is one of the reasons. We don’t know and can’t control and certainly can’t monitor our own cognitive processes. They just happen, we know the input and the output, nothing in between

The Theory of Propositions (Judgment)

Traditional (medieval) theories of propositions resolve the issue into two components: meaning and truth. We resolve prepositional knowledge into two different components: information content and evaluation.

Declarative behavior,. e.g., propositional performance (saying, writing, thinking…) has an informational content independent of meaning, it tells us about the performer independent of any consideration of meaning. The evaluative element resolves itself into additional issues: Truth/Falsity, Informative/Uninformative, Interesting/Uninteresting, Provable/Unprovable, any and all characteristics or attributes of declarative statements and related behaviors. We include not only the traditional epistemic issues but also those evaluations that consider the behavior in a wider context, e.g., social or historical.

All declarative behavior can be divided into two categories: internally and externally determined according to whether the behavior is determined by the performers current sensory situation or not. Judgments involving past sensory experiences are internally determined even though the behavior is ultimately related to sensory experience. The point of this internal/external distinction is that the totality of internally based driving functions can be delineated scientifically. We can thus say what it is possible to say when the declarative behavior is internally determined. This is what epistemology is really all about, saying what can be said, and then selecting among the possibilities. But according to what criteria?

While it is impossible to generate a theory of possible experience, or then a theory of possible externally determined prepositional behavior, (I might see a Martian tomorrow, or develop a new scientific theory, a theory of possible experience is a theory of history); we can generate theories of possible internally determined performance, this is cognitive psychology, evaluating them is philosophy.

What is Recognition?

It seems to be assumed that sensory recognition requires feature extraction- the “bends and wrinkles” theories of identification. Feature extractors have been found in the visual cortex at least. So what? What is recognition actually?

Recognition is essentially knowledge of prior experience, memory at its most basic level. ( Knowing that you have seen or heard something before without any further identification.) Recognition is deja vue at its most basic level. On top of this we may have more elaborate responses which are associated with the stimulus: naming, running away, shooting at it etc., but these are not required for recognition.

Higher level recognition processes still don’t necessarily require feature extraction, they don’t require resolution of sensory aspects or variations. What is required is a unique result in the output space, a response unique down to some semi-arbitrary level of resolution of the stimulus. Recognition requires discrimination, not feature extraction or identification. Universals require resolution, counter intuitively, otherwise everything is a particular. The essence of universals is “close enough”.

How do you generate unique outputs without feature extraction? Time could be an important parameter. Time tags on each element in the output space. But what about similarities? Knowing this is like that requires knowing how they are similar, what features they have in common, or does it? Can stimuli, different at some level of information contact or resolution, wind there way through the sensory cortex and wind up at the output side in similar locations without feature extraction?

(Imagine a future dictionary that not only gives verbal definitions but samples, i.e., sensory stimuli to explain or illustrate the word. For red you get the definition plus a flash of red light. For pineapple you get the whole works including the olfactory and taste sensations. What subset of nouns would benefit here?).

Really Smart Things (RSTs)

Mnemonic RSTs (MRSTs) work on information processing alone, by reduction and comparisons, no calculations allowed, except for statistical data processing and related manipulations. They might do this digitally or by analogue mechanisms, e.g., they might use brains or similar biological systems. They lack concepts like causation, causal process, natural law, functional relationship…Chiefly they remember and compare very large and intricate data sets. They use time lines in effect, but have no concept of time or temporal process.

They work like this : Given a problem: “Determine what happens next….” for a particular system they predict the next element(s) in the data set by referencing only the previous elements, and data sets like the current one but acquired on different occasions. The old data sets could have been generated by the same system (defined, say in terms of system components) or by other “similar” systems. They use old data to produce the new data set, using a variety of mathematical techniques to relate system states only along parametric flow lines, old data produces new data, no theories of “Why” or “How” involved. No “Scientific” laws or theories involved. (This is beginning to sound a lot like a chartist approach to trading stocks- “technical analysis”, but they assume causation even though they don’t know or care what it is.)

The contrast between MRSTs and RSTs is really the difference between high and low information levels. Today’ computational analysis uses low information itinerative analysis (e.g., Multiphysics software) and physical laws, it’s RST type of analysis.