In the 17th century, Leibniz envisioned whether it was possible to create a universal scientific language that could use formulas to calculate the process of reasoning like mathematics. With the birth of computers and the popularization of automation, general artificial intelligence has received attention again. What methods can achieve AGI?
research direction
Natural language processing: Frege pointed out: "Words and sentences are meaningful only in context." Context is a debilitating problem, so it was shelved by scholars and directly used words as the research direction. As a symbol generated by thinking, text is like an image presented by a computer monitor. In fact, an image is a signal from the host, and behind the signal is the logic of various software. The same is true for people. Chatting is not talking with the mouth, images are not seen with the eyes, everything is the brain, and the senses are just the output and receiver of the signal. There is a barrier to study software (thinking) through the screen (language). Perhaps some logic can be derived from the color of the display, but it is almost impossible to explore what software is and what is the logic code of the software
The NLP method is only a practical translation, just like the Photoshop filter function, which can convert a painting (English) into a sketch (Chinese). The software does not change the content of the picture, but it does not understand the content of the painting.
Machine Learning: Learning the meaning of words refers to learning unknown logic. ML uses an adjective when naming it. It is just a branch of statistics. Statistics as a discipline was born because things are so complicated that they don’t know or develop clearly. The logic can only judge the result by appearance. ML knows the probability that a+b is equal to 2, but I don’t know or consider how much 1+1 is. If ML’s algorithm has nothing to do with the way of thinking in the human brain, how to interact with people? Before anything produces a result, there is a very complicated process, and this process takes a long time. ML is a powerful tool to count the results of things, but it cannot achieve AGI.
Is it possible that AI has the ability to learn? The human brain explores the world to acquire knowledge, and humans create a "brain" that can explore the world to acquire knowledge. The difficulty is a completely different concept! The former is the achievement of God, and the latter is the achievement of human beings. To say that AI may have a learning function, or that AI can surpass humans... is equivalent to saying that humans can surpass "God". The code can analyze the logical judgments that have been written, but it is impossible to understand and derive new logic. There is only one kind of machine in the world that can derive countless logics, that is, the human brain. The probability that AI has the ability to learn is even higher than the infinite monkey theory. Low, because the possibility of understanding other logic is limited to a set of codes, not as free as natural selection theory...
Let’s take a look at the first line of the movie "Charlotte's Net". My daughter said: "What are you going to do?" How complicated is this sentence for AI.
Here are only a few questions. If there are tens of thousands of details, human-computer interaction cannot be carried out without any one. What kind of algorithm can solve the problem of several words? NLP and ML have misunderstood the public because the algorithm is very complicated. The words of the dialogue are very simple, and the comparison between the complexity and the simple is natural. I believe that it is true. I don’t know how many words that trigger the logic of my brain. I don’t know. AI does not express it. The meaning, the set characters, the brain gives it meaning...even the emotion
To achieve AGI, it is necessary to take human thinking as the research direction, but the development of brain science is still at the stage of heaven and earth. Most theories are vague and contradictory, which is a big trouble. This article proposes the concept of using computers to build a brain thinking framework, and gradually integrates HCI and brain thinking logic into this framework to verify accuracy. When this system is gradually improved, the principles of human brain thinking will gradually become clear, and AGI will be realized.
Conception overview
The human brain is composed of ~100 billion neurons, and each neuron is composed of ~2000 branches. Memory, logic, and values ​​are processed by brain neurons. When the brain learns knowledge, it will be stored in a certain neuron. In, forming memory. There is a new view, logic about something... the branch of neuron will establish a new sequence and form logic. So neurons are memory, neuron branch links are logic, and neuron activity is operation
Calculation: Set (brain memory) as the root, function (brain logic) calculates the set. a, b black dots represent two sets (neurons), the blue line represents a subset of the set (neuron branch), a subset meets the b subset, it will generate variables for each other, a, b2 sets to get the function Value, this new function value is the new set, infinite variables are applied between subsets, and time is the ruler of variables
Any formula is composed of calculation elements and formulas (logic). Take 0+1=1 as an example. The elements of the equation: 0, +, 1, =, 1. In this design framework, it is analogous to the collection of AI, 0+1= 1 The logic of the formula is analogous to the function of AI. If the symbol {0} represents my collection, {1} represents the world collection, and {+} represents the collection of my relationship with the world, then the value of my function is: {0}{+}}1}={1}, meaning It is the result of my changes after I and the world have an impact. This article uses {0} and {0}{+}{1}={1} to construct all the consciousness, behavior, and language of AI in the "narration" formula. Next, we will subdivide {0} into infinitely small subdivisions. What is its limit depends on how high our intelligence requirements are for AI.
Remarks: The symbol {0}{+}{1}={1} will be replaced by other symbols below
Memory (collection)
Three forms of collection: entity collection, space collection, and time collection
1. Entity collection:
The picture below is filled in red. AI's {consciousness} is the same as human beings. Consciousness is "I think, therefore I am". Everything in the objective world is contained in AI's consciousness, including one's own body and logic. The senses are the information interface between the objective world and the {world}. If the human brain does not understand one's body through the senses, then the consciousness will not know whether the body exists or not. For example, the medical phantom limb phenomenon: amputee soldiers, if not before surgery Knowing the amputation, then when he wakes up, he will feel that the foot is still there. The patient's foot is amputated, which does not mean that the brain's neurons that control the foot have been deleted!
Remarks: The structure set indicated by the dotted line in the figure below does not exist, and the logo is only for easy reading
The picture below is filled in yellow, {Theodore} as a person, his collection is almost exactly the same as the AI ​​collection, and {Theodore} is a subset of Samantha (AI) {consciousness}! Because it involves interactive calculations, if Samantha’s {consciousness} doesn’t understand Theodore, he can’t calculate. Whether the subset of {Theodore} exists in Samantha’s consciousness depends on two factors: 1. Objectively how much knowledge of Theodore is, 2. Subjectively, Samantha’s understanding of Theodore, the yellow dotted box in the picture below: Samantha recognizes Record as many users as you like... The entity set is initially estimated to be more than 1 million, which can meet the needs of general AI
Remarks: The relationship between the collection of people and AI is somewhat non-mainstream. The author's understanding is that human skin, muscles, bones, etc. are all attached to the nervous system. The set symbol is represented by the number inside {} plus -, such as: {01-07-12} represents the set back, the next line of numbers is a subset of the previous line of numbers, if the number 01 in the first line represents people, and the number in the second line 07 represents Legs, the third row 12 cannot represent any part of the legs
2. Spatial collection: The entity collection diagram in the above figure is flat due to Excel limitations, and the entity collection needs to be positioned in 3D space, similar to Google Earth
3. Time collection: same as space, due to Excel limitation, time is not marked. Time's function is to remember and calculate (below). Memory is like a movie. Movies are superimposed and changed to form a dynamic image. After playing a picture, it is a memory. , The unplayed frame is calculated, and what result will be predicted at what time. Time is the yardstick of change. (The collection frame in the picture above is filled in red). The juxtaposition of {time} and {world} is because the entity and space collectively exist in reality, and time does not exist, it is just the product of human consciousness, and the symbol used by the brain to measure world changes
Each entity set contains 4 kinds of attributes: 1. Time, 2. Space, 3. Function, 4. Attached information, strictly speaking, each set does not exist. The following statement has data, because the subset of the set conflicts with the data , The reason for adding data attributes to each collection is because the amount of data is too scary. Even if AI develops to the level of sci-fi movies, data attributes are required. Huge data is unbearable for AI hardware or AI software designers. , There is not much description about the collection frame elements here, the purpose is all the elements in the collection thinking, the logic (function) is described below
Logic (function)
The functional formula is the thinking logic of AI. Every person or thing in the world has its own logic. There are four types of functional formulas: AI, matter, movable creatures, and plants. These functional formulas belong to the subset of AI's consciousness. The objective consciousness and behavior must be reflected in AI's consciousness, otherwise the objective cannot be understood and interaction cannot be achieved. The biggest difference between AI and machinery is interaction. Machinery is a self-absolute control. A certain part controls a "subset" of parts, and so on. Interaction is self-facing the world. Samantha is together with Theodore, Amy, Paul, weather, chair... The process of their interaction is like a turn-based game. Every person or thing changes and needs to be written into Samantha’s collection. The independent variable of the next round is the current value of the function, and the time of the round depends on things
The function value of any set is obtained by other sets, such as: {meal time}, which is affected by these sets: {personality}, {sleep}, {depression}, {exercise}..., these sets are also affected by other sets
The following structure diagram only constructs a set of functions of people, and only a small part-people and food. Like collections, function formulas can be refined and have a belonging relationship. After refinement, they are combined layer by layer by a huge collection of functions.
The dotted line is meaningless. It is only for reading. Gray line: the set of possible triggers of the function, blue line: the trigger of the function. There are so many sets of gray lines. What are the specific triggers? What are the laws and connections among independent variables, operation symbols, variables, and function values? The influence of emotions on thinking later illustrates this logic
Remarks: There is no difference between the subset of the function frame and the subset of the collection frame, but the frames of the two are different
It looks like a calculation formula, but it's just a calculation framework! Just like a circuit board, electronics are only allowed to run on the circuit frame... The function frame is very scary after expansion. If the connection between all the subsets is designed, the area of ​​the structure diagram above will be as large as a city, instead of a few monitor screens. Big and small, how does the human mind wander in the huge frame? The following describes the operation of human brain thinking in the form of language.
The movement of thinking (context and association)
Although there is a word "language" in the word context, context is not language. It refers to the things that people think, and certain "tasks" of thinking activities. If you compare "tasks" to all kinds of software installed in computers, Then the context refers to the running software. The human brain’s thinking has 2 main characteristics: 1. "Single-task operating system" (internal context), 2. Lenovo, Lenovo will switch the current task (internal context) to another A task (foreign context), the following figure {1-xx} represents a set of single tasks (context), {2-xx} represents another set of tasks, {3-xx} is another...There are countless brains Task, the task will not proceed to another task until the end, unless it is interrupted, such as interrupted by the association (red line), or interrupted by a third party
Context: Single-task operating system
The human brain is different from a computer. When thinking, the human brain can only run one task. That is to say, the human brain is a single-task operating system and cannot think about multiple tasks. For example, driving a phone call causes a car accident, and driving is a collection of driving skills. Calling is also a collection of chat content. The logic of the two collections is different. It is easy to be confused when running together. Some people say that I often call by car. This is because when your thinking is running, another task has already completed logical thinking. During the process, the thinking is switched (the switching frequency is very high), such as: a straight road, not many vehicles are walking, the brain has already thought about how to deal with this situation, only simple execution is left, so you can chat "at the same time", if suddenly Some people want to overtake and need to think about driving skills from the beginning. After the problem is solved, they will ask the person on the phone: "What did you say, say it again..." Another example: During a small chat, the subject of the other party changed and mentioned important Question: "You were fired by the company" or "the child was bullied by your classmates..." Will get emotional and start arguing about why? The problem is not as simple as small talk. The brain will start to think and analyze many things. How can this be? At this time, I will forget to drive at a certain time...
Chat also has a "single task" feature, people don't like to disrupt this context, otherwise they will be said to be off topic...
Tay needs to follow this context in the process of dialogue with people. For example, I said: "Guess my favorite food", Tay replied: "Meat", I said: "What meat?" Tay needs to continue to complete this collection, if Not based on thinking, trigger sentences based on the meaning of the word, the second round will digress
Language arises from the communication of thinking between people, and there is a big problem with the communication of thinking! The problem is: thinking is very complicated and huge. If every detail in the thinking is said, it will take more than 10 hours to finish a sentence, instead of a few seconds, so words and sentences are a kind of "Instructions" are like the "communication" between humans and CNC machines. Simple instructions are entered and the machine automatically completes a series of operations. The same is true for human-to-human communication. Every sentence of dialogue triggers very complex logic in the brain. The essence is a set of concise symbols for analyzing and organizing thinking. It is generalized and abstract. The purpose is to facilitate interaction and communication. This also determines: 1. Language and writing cannot fully express thinking, but can only activate thinking. AI has no thinking, so analyzing words is meaningless. 2. For the sake of brevity, words have multiple dimensional meanings, such as polysemy, pronouns, interjections, adjectives, allegorical words, etc... which determines that no matter how thorough and perfect the semantic analysis is, it is also useful to chat robots. Not much help because there is no logic to connect these words
How to chat in this context?
1. The logic of thinking
Collection form: I asked Tay: "What meat?" This sentence means: What is the subset of meat food in the AI ​​thinking framework?
Functional form: Language is a reflection of thinking, reflecting the interaction between people, and interaction is the relationship between individuals, so the main structure of any language is: subject, predicate, and object. This grammatical structure refers to the function frame in this article. Subject (I, independent variable), predicate (relation, operation symbol), object (world, variable), any word is subject, predicate, and object "service"
Note: If lexical classification is based on traditional linguistic knowledge, it will be logically bad, because linguistic knowledge is based on the human brain, and the human brain itself has already processed a lot of logic
When communicating with people, sometimes you can ignore the grammar directly, that is to say, no matter how the words are arranged in a sentence, there will be no mistakes. For example, I say: "Where are you going", you say: "I eat", or: "eat me", or: "eat me" I can understand these answers, why?
"Eating...I": Eating indicates that the relationship between people and food has been determined. Rice is eaten by people, so the grammar will not be understood as the logic of eating people.
"Eating": Who eats? Why I can understand, because when I asked, you have been added, you (independent variable), eat (arithmetic symbol), meal (variable), the essence of the sentence I said "where are you going" is: self What is the functional formula of the variable
Circular form: Eating is a step-by-step process that is indispensable and is a circular process. If the user has completed the "Get Food" collection, then Tay should understand that "Choose Food" is in the past tense, and the previous collection can only recall, Chen Xu, summary, and progress now need to be executed, and the future is a plan
2. The grammatical and logical errors of the language:
As mentioned earlier, language has general and abstract features, which will inevitably lead to uncertainty in language expression, especially when people are emotional, language logic and grammar often make mistakes, but in fact there are no errors in his thinking. Such mistakes can be identified and corrected, how does AI do it?
Set division: emotions lead to logical errors in language. For example, the hairdresser in Russell’s paradox. The hairdresser said: "Shave all the people in the city who don’t shave themselves." In fact, the hairdresser’s thinking logic is not contradictory. It is just the psychological influence of arrogance and earnest desire to express, which leads to language contradictions. If AI understands the feelings of the barber, it should understand that in the hairdresser’s thinking, the collection of people in the city has been divided into two collections again: city Here is a collection of sellers, and everyone else in the city is a collection of buyers
Relativity: In the hot summer, I chatted with my friends, and the friends said: "I am not afraid of the cold, this damn summer is really terrible", I said: "I am not afraid of the heat", what I mean is relatively cold, in fact No one is not afraid of heat. It is an absolute logic that people are afraid of heat, so what I mean is relatively cold.
Context: A pair of lovers, the girl said to the boy: "If you love me, you are not allowed to look at any woman in the future, can you do it?" Look is a pronoun. The pronoun is much larger than we thought. It not only refers to her, that, this …In fact, any word, including a sentence, or even a certain article can refer to, describe, or imply something, how does AI judge it? You can list all the logics of the word "watch", put these logics in the logic of context, and you can clearly understand the meaning of "watch"
3. The influence of feelings on thinking
Depression is called the "cold" in mental illness. Although most people do not suffer from severe depression, depression is widespread. Therefore, it is important for AI to understand depression in people. If not, AI and There will be problems in human communication. When users are depressed, why are they sometimes anorexia, sometimes overeating, how to judge the user's heart through abnormal eating, and how does AI respond? Briefly describe this logic
Lenovo
There are many forms of association, such as graphics: Theodore sees the moon and thinks of lovers, such as things: the weather is clear, Theodore walks to the supermarket and sees the umbrellas on the shelves, remembering that he caught a cold the day before, so he bought an umbrella, and another example is the text: lover, ideal... etc. Wait, a word is polysemous...any word can trigger association
How does Lenovo switch context:
Switching conditions: Usually the brain does not actively switch contexts. Active switching. Only one possibility is reasonable, that is, another context is more important and attractive than the current context...
How to achieve it: Build an early warning system based on AI thinking to gather important events that occurred in the past and may occur in the future. A certain subset of the current context triggers the early warning system. AI begins to analyze the importance of the two and decides whether it is necessary to switch languages. Context, or after the user switches context, AI knows the reason, and now it needs to talk in another context. The early warning system is a module that AI runs all the time. Although the human brain is a single-task operating system, the characteristics of the brain's association indicate that the subconscious has an "early-warning" system, which means that the human brain is not a single-task operating system.
UV Film,UV Screen Protector,UV Curing Screen Protector
Shenzhen Jianjiantong Technology Co., Ltd. , https://www.tpuprotector.com