Learn how is it that your brain can recognize features, this as this and that as that, without any effort.

Intelligent operation involves feature detection and feature extraction.

One is the dominant mode of intelligent operation. Which one is that?

And what does it mean for something to be the same thing — when a feature is detected or extracted?

 

— 〈  1  〉—

``The appropriate and suitable for some end can occur by a process which neither has information or knowledge of the end which it achieves nor to achieve that end, and which is not occurring in order to achieve any end'' (SCH51).

Much feature detection and even feature extraction can occur as a side product of the behavior of many decentralized processes, merely by the replaces of one the processes by another and nothing changing, which can be sniffed by other processes as an invariant. Each process may inherit and carry parameters from its environment, allowing recognition of an invariant, with high probability, in the environment.

The other fact is that often an image is not recognized as such and such: as an apple or an orange. Or as the same probably as another image. There may not be enough information there in the image and it's ambiguous.

Often we interact in an object, play with it, in a Piagetian way, use it for some purposes. And identify it as such and such. And classify it as such and such. And then move away, and merely associate the various resulting blurry images by plain habit with the object as identified (PRI71, PRI13).

There may really not be enough information in the image itself to identify it unambiguously yet we people are able to achieve such identification, and bots are not. Most bots are trained with images, or programmed to look for all the information to identify an image in the image itself. Very likely this will not work for identification in the same style as human can achieve.

While bots can now recognize images as such and such with less error than people, most people outperform bots in recognizing some other images, especially low resolution or noisy or abstract images, and the bot does not and never will, if all they do is work with images, coincide in identification or see the same world, as human actors.

Many separate processes in the identifying system having to do with operating or experiencing the object for other reasons may encapsulate and carry parameters from the object and go about their business in the system doing other things than plain identification and one gets replaced by another without any detectable effect, which is itself a detectable effect to another process among them, for example, or two groups overlap with no effect, which is an effect, and lo! an invariant is extracted. Apparently some feature of the universe is extracted; and a world is piece by piece constructed in the system, an abstract or classificatory model of its environment.

Our brains and learning systems we construct may not necessarily aim to detect reality. They have many little tasks, problems. Autonomous processes operate on reality to solve these problems. But by interacting with reality they detect it anyway. And they identify some things as the same thing.

Let's discuss some of this and start unpacking it.

 

— 〈  2  〉—

The question in detecting regularities and ``the same event'' isn't feature detection but really feature extraction one may argue and this view has a long and interesting history.

The multiplicity view of existence was also James Hutton saw the world (HUT94).

Difference is source of existence, and = is an abstraction with the method of abstraction encapsulated. Therefore hidden and not inherited. Or else sameness is by combination AB = CD but A,B,C,D ∈ ア, and ∀ (V,W) ∈ ア×ア, characteristic measurement result 「SAME」(n,m)=0.

Notation:

B(A) = A.B.

Alternatively, considering things pragmatically, like our brain mostly does: N =_(f,X) M if (N.f = X = M.f) ∧ (N ≠ M) (PRI71).

This is exactly the sense in which the brain is a computer. The is here encapsulates a pragmatic method of abstraction: studying the one we learn about the other (MIL67). Eliminate the layer of abstraction and of course they are very different (NEU58, MIL67).

So the uniformity of natural law in time and space was not despite difference over time but it was rather because of all the precise differences of things as they exist in the world (WAT72).

Meanwhile the differences were sources of multiplicity and existence of primaries. Things were basically abstract equivalences or else combinations. Invariants. A 「desk」 is not a desk.
Not exactly. Long strings of = could lead to error. In this manners. A = B, B = C, A ≠ C.

In some cases X ≠ X. (Remember can be fundamentally abstract.)

Consider the case A =_G B with the equality made explicit.

G might be A < T and B < T therefore G(A) = 1 = G(B).

If combinatorial identity A =_G B then a block must be specified, G, where this holds.

For example, we might unencapsulate:

(∀ X ∈ G) (A⊗X = B⊗X) ∧ (A = B),

(∀ X ∉ G) (A⊗X ≠ B⊗X) ∧ (A ≠ B).

Then (A = B) ∈ 「BLOCK」G.

Outside the block G, which has a mist of other operations X, this may not hold.

 

— 〈  2  〉—

We can view G as a small commutative category.

Two things labeled by the same word cannot in every context replace each other.

Why is that? For example, consider the following case:

AB = CD but A and C can have different side effects on E and E does not participate in the context of AB but E does participate in the context of CD. AB and CD are abstractly equal.

Differences in behavior and therefore relations of each one to the rest. If all are related to the others in the same way. That is if the same pattern of measurements result is streamed when each measures the others, none would be primitively distinguished and having no behavior would not exist in the world. Or else the measurement was occurring at some abstract layer where differences were presumably abstracted away (WAT72).

Therefore the question in detecting regularities and the same event isn't feature detection but really feature extraction (PRI13).

And this can be generalized to processes as actors.

Identity is disaggregated anyways and contextual and infinitely many definitions can be made.

Not all are useful for feature extraction. The question is to view features as behaviors.

 

— 〈  3  〉—

Behaviors can be structured and context sensitive so we treat them as actors. They interact and the interactions and their results is the system (MIN86).

{"Actor1'":
|   Operation11   Operation12   Operation13 ...   |
Operation11 ≡ ...
Operation12 ≡ ...
...}

{"Actor2":
|   Operation21   Operation22   Operation23 ...   |
Operation21 ≡ ...
Operation22 ≡ ...
...}

...

Suppose a mist is a large group of actors working concurrently. Messaging.

A mist is an object in a block. Suppose you have mist C and mist D in the same block G.

Mists can also partly overlap when joined. Hence the name: mist.

Updating/Assigning mist C ↦ Actor1 ∪ mist C doesn't send you into another category.

Updating/Assigning mist D ↦ Actor2 ∪ mist D doesn't send you into another category.

And (C ↦ Actor1 ∪ C) ∧ (D ↦ Actor2 ∪ D) doesn't send you into another category.

And if so, and this for all pairs of objects in the category, and if also Actor1 can be replaced with Actor2 and Actor2 with Actor1, then (Actor1 = Actor2) ∈ 「BLOCK」G might be meaningful.

Then for (Actor1 = Actor2) ∈ 「COMPLEX」(「Mist」C, 𝔹) to be true analogous relations must hold for invariants of the mists for all the various overlaps of mists for some set 𝔹 of blocks.

Will write more about this. The question is about useful definitions for us pragmatists (JAM07).

Identity is disaggregated anyways and contextual. Infinitely many definitions can be made. Not all are useful.

©2018 tibra.
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License  . . .   . . .   . . .    Text and images: @tibra. @communicate on minds.com