Modeling human reasoning 3 CONNECTIONS.

1. Any "knowledge" is a "path" through the connections of images.
The system is binary and therefore everything has only two states - Yes / No (1/0, true / false)
All the "forks" paths have only two options forks - Yes / No (1/0, true / false). In this case, the "negative zone" is not stored.
Denote the fork as a "switch".

Therefore: the switch of possible states is 3.
1. I do not know
2. Yes
3. No

YES - this is the way somewhere along the links.
NO - end of the road.
I DON'T KNOW - this is the lack of connections. Absolutely normal (natural) state of the switch.

The "limiter" works on the "filling" of the cells and therefore the "stupid" thoughts can not arise in principle. For example, the "handler" when "recognizing" the word "brick" stumbles upon "inedible". Everything. The end of the road. The result is sent to the "ball of thought", there is no further way. And accordingly, for example, such a thought: "What is tastier than fried brick or boiled?" it can not be. Another thing is if the handler stumbles upon a switch in the "I don't know" state.
There are such words: "hypothesis", "suppose", "suppose" - this is a "miscalculation" according to the variant of the "state" of the switch. If we are dealing with a “branch” which has reached an impasse, the neutral state of the switch (I don’t know) has two options, interrupt further calculations and wait for new data to arrive. Or try to "figure out" the branch yourself. We decided to "think" (to dream) Let the switch in the state "1" then ...
Immersion "deep" thoughts. Controller "protection from deep care" (by timer). The "farther away" we plunge along the "unexplored" branch, the more it gets in the way of the switches in the state "I don't know", the more "branching". Protection from side drift - priority controller.
What is a lie? This is a deliberate distortion of the already calculated data, to obtain the desired result.
(By the way, when smoking weed, the priority controller is cut down. The result of such a breakdown is a spontaneous switching of the direction between the branches. It looks like this: Thoughts jump from one to another and constantly go to the side Until you forget about the subject of conversation "in general. The reason is an overflow of the buffer (thought ball) and the dumping of" useless "data. When drunk, there is a shortening of the" depth of immersion "(all do not care, a lot of houses were built, trees were grown etc.) ) besides, “wrong” is happening "Estimate Values" - noise interference in data.)
Thus, we have "image-object" - the brick of which the building of "knowledge" is built. All that is "below" "image_object" - the data from which image "consists of image-object, these data do not have" horizontal connections "since these are data from the" lower level "of" atoms "(data from the senses). The main difficulty lies in that "image-like object" and "image-like-combination of images" (concept) can be "connected" at any level of "immersion". I called these connections "horizontal".
A little example: What color is the grass? This is a journey inland. A: Than I hammer a nail? - This is a trip to the "neighboring branches."
In general, “going aside” is moving to the next branch along horizontal links between the branches, “leaving” is very far (if the task is not solved quickly) as well as the controller of the “depth of immersion” (by timer).
How is the selection of "pattern"? Suppose we have a "task" to find a "connection", but we have very few "known parameters", but then we know the parameter of the output connection - "color" (for example, red)
With "reflections" will be immediately "screened out" any options links having unspecified this bit. That is, such concepts as "sky" and "water" may be red under certain conditions (sunset sky, sparkling water), but for example, "blue sky" and "tarragon" will be eliminated immediately (as the color is known). At any level of "immersion" the miscalculation first of all goes "deep into" and it turns out that the color is known. And then the remaining parameters are checked.
If the exact “data” about an object is unknown, this is an absolutely normal state of “knowledge”, we are simply dealing with an “object” with a limited number of properties. At the slightest opportunity (if in the process of communication, for example, such data is obtained) the corresponding cells are filled (this is the "learning" / storing / filling in the database. If, in the "miscalculation" of any data, the "assumption" that received "proof" in In the form of a horizontal intersection with another “positive connection”, then this “switch” is set to “1.” In case some data received a “refutation”, the corresponding “switch” is set to “0” or “I don’t know” independent of the lengths s branch (immersion depth).
Data checking. When receiving data from the communicator, they are checked for "consistency and correctness." An example has already been given of the "fried brick" - this is the wrong connection, therefore the thought is marked "wrong" and no further processing is done. "Ball of thought" is reset. Data output - "nonsense." Checking for "consistency" is when the response received (during a miscalculation or from the communicator) "fits into the pattern, but is not the" correct thought "Example-anecdote: An Indian bought himself a new boomerang, and the old one gored up. ( miscalculation of the meaning: boomerang - he is thrown back; an Indian bought a boomerang (everything is fine), and the old one was bored to throw it out - the old one “many times” [you] “throw”, but if he “throw” it comes back. Not logical. is returning.
Possible examples are more serious:
"Can we build a raft of reinforcement?". - The armature does not float in water. The raft will sink. Wrong thought. Answer: NO.
Can I build a bottle house? - Answer: It is possible, but illogical. Glass bottles. Glass fragile material. Obno homes built of bricks.
"I hammered a nail into a brick." Can you hammer a nail with a brick? Can. So - the right thought. But it is illogical, as they use a hammer for hammering nails. The template does not fit.
So besides "the very possibility" (the fulfillment of the condition) must also
present the predicted probability of the event origin (approaching the pattern). Although in the latter case, the idea is not disputed, still "I know" is not installed, because this is not “knowledge”, although it is “new” information. The illogicality "extinguishes" the further miscalculation forward along the links. Extinguishes the "process of filling the database."

Approximate data structure diagram.

If you submit any "concept" in the form of a tree cut, turned by the butt (butt) to the observer.
That central trunk is an image. The branches from the central part will be "images_ as_object".
And the leaves at the ends of branches are references to data from the senses. While thinking, we go further.

Horizontal connections between concepts are interconnected connections between trees and branches. In this case, only the “intersection point” or “branch point” is considered - it is called the “switch”. Here is an analogy ...
In this case, if you extract only the "horizontal" connection (and not the whole tree :), then the resulting structure will be a "natural language phrase".
And how does "understanding" work? And understanding is just a "reverse process." Having a "horizontal connection" we select "a place" where it should be "littered". If there is no suitable branch or even a whole tree, then everything, the problem will not be solved ...
And the poet IR (AI) can not "understand" the phrase entered by man. Just trite "not in the database"
How to get around?
When I say that you need to create a "correct database", you need to create a "forest of trees" or set a mechanism for the spontaneous generation (appearance) of these trees. If in the process of "selection" there is no necessary "tree" - it is created, etc.
It is not necessary to implement the IR exactly this way, but there is “something” in it ...
For those who think that the process is avalanche-like and the AI ​​will very quickly choke in a small stream of information, I will answer. Far from it. Restrictions in the system are of course necessary, but the "flow of information" is not as endless as it seems. Since very quickly go "folding" (compacting) of the system. Synonyms, sentences with the same meaning, etc. Stems (concepts) at the output will be much less than you think;)
But the "templates" - can be infinitely many. It makes no sense to fill the database with "templates" and in response to one "template" there is no answer to the other "template" ... (but this is how all existing chat bots are implemented)
And so on the basis of this paranoid example, I wanted to show the difference between "understanding" and "response."

And so, we have a tree. But we obviously do not know whether we have ALL branches and WHAT and HOW MANY leaves on the ends ....
It does not matter. No matter what data is, what is not. Since at the moment we are operating with the trunk (image_of_the set of images), particulars are not important to us, and only if necessary, we go down to the leaves, behind the “concrete” information.
(hell, not a very good example with a tree brought :) but oh well ...
At what level does the information about the object appear? Answer: on any.
We have at the entrance the phrase: The earth is in the shape of a ball.
We look at the DB tree "Earth" (the concept, image_as_cumulative_brain) and stretch the ribbon to the tree "Geometric objects" - "ball". We have a "horizontal connection": Earth- (shape) ball
We have at the entrance the phrase: What color is "orange"?
We look at the database. "Orange" we go "down" on the trunk and look at the "characteristic" color. We transfer data for preparation of the answer.
We have in the phrase: Monkey eating a banana.
We look at the database. "Monkey" know, and the word "banana" is unknown to us. We start a banana tree at the trunk level, all the data are switches in the state “I don’t know” at the first stage everything. We look at the "priority controller" how important the information is.
If this is “just information” :) then we do nothing, we will do it when we need it ...
If this is a question, for example, and an answer to the question is required, we display the appropriate message. (Well, I don’t know what a banana is, I’ve never seen it. Or. And what is a banana? Or, before I can answer your question, explain the meaning of the word "banana" to me, I’m ashamed to admit it, but I don’t know what it is)

1. How many and what leaves can there be?
The number of "small branches" is limited by the number of sense organs and the number of leaves by the number of receptors.
(sorry for blasphemy)
2. How many and what branches can a tree have?
The number of branches depends on the number of available "properties"
3. Is the concept and tree the same?
Yes, just one of the branches of the tree is responsible for the "instructions"
etc. If there are more questions on the data structure ask questions.
If the example with a tree is more or less clear, then let's talk about the database.
Initially, for the speed of the system, we do not need ALL the information on the object, but only a label (label, ID, trunk section) is needed, and only if necessary we’ll "get the necessary data from the desired cell. Going down or up the trunk.

Types of links.

VERTICAL - directed "from top to down"

HORIZONTAL - depends on "connection type"

at the bottom "level" - directed from the "object" to the "properties"
Orange has the shape of a ball. ORANGE - BALL.

> only by the arrow, and you cannot go against the arrow.
on the “top” - (classes) - (“concept” - “concept”) are bidirectional.
> do 2 arrows, one there, the other from there
> 2) both concepts are absolutely equal, it is possible to walk back and forth completely by communication in both directions. And it cannot be said that the connection comes from somewhere and enters somewhere, it simply combines 2 equal concepts.
Snowflakes fall.
Spinning wheel.

Vertical links are descriptions and specifications. MEMORY. KNOWLEDGE. (within the system) DB.
Horizontal is COMMUNICATION. SPEECH

We communicate only with "a chain of horizontal connections"

When the chain enters the inside, ALL THE TREE is completely completed. (even if the tree consists of 1-2 branches)

1. Speech is a chain of horizontal connections.
2. Trees combine the scattered data into a single whole (forest) - an internal representation in memory.
(sorted by concepts)
3. Each tree - corresponds to the concept.
4. A fork is a branch to certain "data", and both the branch "vertical" and "horizontal" are the same.
(here we need examples) How to distinguish them? No They are equal and independent. The only division into "upper level" and "lower level" low level ends with "data from the senses" (whether it is smell, color, shape, etc. it does not matter) "upper level" is just an interlacing with nothing ending - a maze. chaos. Ordered by what is already there. "I know that - that I ALREADY know, what I do not know - then I WILL LEARN over time."
5. "Path through connections" is the process of working with the database. (algorithm, executable part of the program when rendering)
Just the data can not be extracted "just so" (give me the whole sample on this topic!) Since there can be SO data on the topic that there is not enough RAM. Data is retrieved only "along the way" along links (more precisely along the path).
Perhaps a somewhat limited way, because before some of the data will have to go "curved devious paths," but that is how it is arranged in the UE. It is more important not to have access to all at once, than to drown in chaos and an overabundance of information. Self-restraint. Anti-hang protection.

Branches are only from the trunk and from the branches. There are also no "double meanings"
(Square circle)
The switch plays the role of not only the filter, but also the controller-rechecker of the correctness of the recorded data. The scan is performed all the time, (just like Kaspersky checks the files in the folder when you enter the folder).

> "chain of horizontal connections" - directed or not?
NOT. Examples:
car wheel
car wheel
snow goes
snowing
wind is blowing
the wind blows

Just a connection. Data. Direction appears when they are ORDERED in a structure (tree).
The main thing is the hierarchy of the tree.
The principle of the interaction of "horizontal links" without "vertical" is simply missing. Horizontal links are a set of phrases. Porridge. Chaos. Each connection makes sense (meaningful) Each phrase is related to another / other. But to understand this intricacies will be IMPOSSIBLE.
Vertical links ORDER structure.

Objects and processes.

Any object is an object. I have no other definitions here.
Table, house, man, glass, bottle - objects.
The process is the interaction of the object / objects without the objects as such, with "variables" in the form of objects.
(but this is not a connection!)
Examples: Run, jump, fly, move, build, break, program, etc.

Approximately (but not exactly) in the Russian language objects - this name is substantive / pronoun
The process is verbs. Once again I’ll focus on something, since we’re not talking about words ...

The process of thinking is not in words but in images. IMAGE NOT = WORD. Match only approximate.
Words-synonyms and images-synonyms DIFFERENT things, although they are very closely related and the definitions are similar.
What is the difference?
Words MUST be separated from IMAGES (thoughts) because the binding of one to another is made at INPUT and AT THE EXIT of the system (we have a line of text at the entrance, a line of text at the output, but INSIDE THE SYSTEM, thinking is IMAGING)
In addition, one and the same words (sentences) can be correlated with DIFFERENT IMAGES (depending on the context) and vice versa, one and the same IMAGES can be expressed in DIFFERENT WORDS.
Why so stupid? Because a thought is a thought and it DOES NOT HAVE a relationship to WORDS. Speech (words) is the second signal, then the binding of images / links is THEN, secondarily. Thinking is NOT WORDS.

There is nothing terrible here. But you just need 2 things:
1. This is a TRANSLATION from EXTERNAL (speech, words, sentences) LANGUAGE into INNER (images / links-chains of images) and vice versa.
2. This is not to confuse the word "ELEPHANT" and the image of "ELEPHANT". That's why I offered to somehow mark them at least like this:
(ELEPHANT)

The analysis of ANY thought, whether it is “born” inside the system, or received “outside”, begins with the “definition of meaning”. That is, the incoming information is first analyzed. Then compared with the existing one. (context) Then builds up in the "current thought" buffer. The meaning is determined (so what did he mean by this?) (There are SEVERAL semantic layers, it depends on the "depth of immersion") And only then is the "binding_ to_" made.
Binding_to_It is a reaction of the IR as a person, to the received information.
I am a mediator between the interlocutor and my own database.
This is just another test, "on the need for" storage of this idea.
The idea is placed in the "virtual buffer" before it is entered into the database. Basically, due to distortion when translated from "external" into "internal" language. And because of mistrust to the interlocutor. (the interlocutor may be mistaken / not know, lie, fantasize, mock, joke, and finally just turn out to be a "moron"). Different "systems of coordinates" among the interlocutors (here you have the presence / lack of knowledge, here you have vocabulary, here you have the coefficient of intelligence and the ability to choose words for your thoughts and much more). Finally, just noise / interference / distortion from a misunderstanding of each other due to a heap of various reasons (ranging from racial to religious beliefs and mental normality).

After all the "modifications", the thought "seems to be" becomes clear, that is, it is given "in order" and "is what was transmitted."Now it is necessary to “appreciate” the need for this thought. Does she even need it? Does it have any value (priority) for me as an individual? Or is it useless and you can forget about it? Or vice versa is harmful (wrong)

So. What is the "own opinion"?
Life experience. Knowledge. Emotional experiences. Findings.

What is communication? This is an exchange of all of the above.
Consequently, on the basis of these data, we will try to consider the types of possible reactions of the individual to the information received. We classify information by type.

1. Useless information. (low coefficient of confidence in the interlocutor, respectively, the conversation is not remembered for a long time and most of the data is destroyed, the more so nothing is recorded in the "long-term memory").
2. little information. (same)
3. General (introductory) information.
4. Information on a specific topic.
5. Behavioral information (instruction).
6. Unknown information (important! Always important is that unknown. Unknown can be dangerous).
7. Clarifying information. (Boris, you're wrong!)
8. Disputes information. (correction of connections)
9. Fixing information (I already know this, but repetition is the mother of learning, additional confidence in knowledge will not hurt)
10. Own guesses (unverified connections).
eleven. ??? (can you add?)

Hence the reaction of the individual, after the classification:
1. talked, forgotten.
2. talked, forgot, but maybe some thought like it, you can remember it.
(for example, you can memorize a joke, anecdote, etc.)
3. Usually it is simply remembered, without processing, with notes to clarify.
4. Remember.
5. Remember.
6. MUST remember! and mark the priority of processing and clarification.
7. This is to point 3. And to point 10.
8. FIX IMMEDIATELY! BUT!Before correction, ALL is rechecked and re-recalculated. It is always more difficult to fix than to memorize a new one, since the "wrong connection" has already managed to acquire a bunch of other "correct" connections. Accordingly, it is necessary to fix them ALL! And this is quite a lot of work and a large sample of links to be rethought and destruction (if necessary) or corrected.
9. Just correct the coefficient of "confidence".
10. Point 7 will help us out.

So “binding to the self” is simply the definition of the “coefficient of necessity” in these data and the corresponding reaction to this coefficient.
The problem is that the data received at the input of the "communicator" do not have any "coefficients". And the IR, should set these coefficients by itself "by eye";)

For chat bot, everything can be done and easier ...

Что такое (например) ЯБЛОКО?
1. ОПРЕДЕЛЕНИЕ.
Проще всего, загоняем определение (для коммуникатора) в текстовой форме (например: толковые словари, википедия, энциклопедии и т.д.)
1.1 Для процесса (КАК?) - описание процесса.
2. ПРЕДСТАВЛЕНИЕ. см схему. по стрелочке и вниз, где визуальноорганолептическое представление в архиве от органов чувств. Для чат-бота, думаю вполне достаточно... Для ИР придется делать еще и МОДЕЛЬ-представление.
2.1 Для процесса (КАК?) точно так же - КАК выглядит процесс? (если это например"забивать", то визуализация/представление процесса забивания)
3. ИНСТРУКЦИЯ. (если она есть) Это описана РЕАКЦИЯ ИР НА "ПОНЯТИЕ".
"Понятие" это ни что иное, как обыкновенная связь. Просто ВЫДЕЛЕННАЯ от остальных и помеченная ЯРЛЫКОМ для упрощения использования.
Например: "написать письмо", "съездить на дачу с шашлыками", "пригласить девушку в кино", "сходить в ресторан" и т.д.
Самая большая сложность здесь - это большое количество информации в самой инструкции. Например "пригласить девушку в кино" подразумевает целую бездну ЗНАНИЙ, ПОСТУПКОВ и линий поведения, в зависимости от реакции девушки.
(А если откажет?/как добится своего, а если не может?/перенести на другой день, а если фильм не интересный?/поискать другой кинотеатр, А что ПОСЛЕ сеанса?/смотреть по обстановке, то ли прощальный поцелуй... то ли "пригласить попить чаю"/главное что б в постели не курила :)
Наши задачи (пока) не настолько глобальны, и все же...
При поступлении информации, в первую очередь отыскивается "понятие" и выстраивается "линия поведения" извлечением имеющейся "инструкции".
На следующем этапе - этапе обучения, при обнаружении "инструкции" в "чистом виде" она ЗАПИСЫВАЕТСЯ в раздел "инструкции".
Например: "Перебивать собеседника невежливо", "если с тобой поздоровались - нужно поздороваться в ответ", "с незнакомым человеком нужно общаться на Вы".
Это самые обыкновенные инструкции, после перевода их на "внутренний язык", их можно еще и перевести на "язык инструкций" и записать в необходимую ячейку БД.
В конце - концов, чат-бот всегда может спросить/уточнить что-либо...
For example, ask, "how should he behave" in a given situation.
(and this is also an instruction :) If you don’t know something, ask ...

1. And so we have some information at the entrance.
Information is analyzed, meaning is highlighted based on context.
No instructions were found for the DB of instructions for using this information. Information is simply memorized. Snapping_to:
- I understand you.
- Thanks, I remembered.

1.1 Same as in p1. but the information is already known:
- Yes, I already know that.
“I already know that.”
- You did not tell me anything new.

2. У нас есть какая-то информация на входе.
Информация проанализирована, выделен смысл на основании контекста.
По БД инструкций не найдено ни каких "рекомендаций" по использованию данной информации. Но информация признана ЛОЖНОЙ, противоречащей той информации, что уже имеется в памяти. При чем коэффициент "достоверности" источника выше, чем у собеседника. Как поступить в этом случае? Ведь реакция на ложь может быть разной. От откровенного "послать вруна подальше", до "а может он прав, а я просто неправильно информирован?". Оставляю решение этого вопроса за вами :)
ИМХО я бы использовал "контролер приоритета" (насколько этот вопрос важен?)
и Общее Эмоциональное Состояние (ОЭС) как "направление" для дальнейшей беседы.
(соответственно - для разных ОЭС, уже должны быть заложены основные "линии поведения"-"инструкции")
2.1 Для ИР все гораздо сложнее, так как ложь и "искажение правды", вымысел и фантазия, все эти "понятия" попросту не имеют между собой четких границ.
И для реализации "житейский логики" ИР должен не только уметь "распознавать ложь" но и "уметь лгать". Ложь, ведь это всего лишь "умышленное" искажение одной/нескольких связей внутри БД, для получения "необходимого результата" (как "линии поведения" в "нестандартных ситуациях")

3. Information received and cannot be interpreted. What to do in this case? Maybe it is: qwerty, or maybe the text is in English, or maybe just an unknown section of knowledge (for example, chemistry / physics, mathematical formulas, etc.).
What to do?

In general, something in this spirit ...
In any case, first you need to do something simpler, but leave the opportunity "to build the database" in the future.