CN113962353A

CN113962353A - Method for establishing strong artificial intelligence

Info

Publication number: CN113962353A
Application number: CN202010695466.3A
Authority: CN
Inventors: 陈永聪; 曾婷; 其他发明人请求不公开姓名
Original assignee: Individual
Current assignee: Individual
Priority date: 2020-07-20
Filing date: 2020-07-20
Publication date: 2022-01-21

Abstract

The method for realizing the strong artificial intelligence provided by the invention adopts the mapping from the input data to the local common characteristics on the perception layer and the mapping from the local common characteristics to the concept on the cognition layer. The establishment of the concept is realized through a relationship network and a memory and forgetting mechanism. The basis of the relational network is realized by multi-resolution features and associative activation. The decision of the machine is realized by adopting the principle of trending toward interest and avoiding harm and the method of iteratively using prediction, decision and response through the connection relation between the machine demand and emotional motivation and motivation state and specific things established in the relation network. Through the method provided by the invention, the machine can gradually obtain simple to complex responses to the input information, and has similar motivation and emotional expression to human beings, which all show that the machine learning method provided by the invention has great difference with the existing machine learning method in the industry, and no similar method exists in the industry at present.

Description

Method for establishing strong artificial intelligence

Technical Field

The invention relates to the field of artificial intelligence, in particular to how to establish strong artificial intelligence.

Background

Current artificial intelligence is typically designed for a specific task, and there is no general artificial intelligence that can accomplish multiple uncertain tasks. The biggest obstacle to implementing the general artificial intelligence is how to establish a cognitive network similar to common human knowledge among various complex things. Only if the machine has common sense like a human, the machine may produce mental activities like a human. The result of the current deep learning is an exquisite characteristic mapping method which is greatly different from the learning process of human beings, so the achievement of the deep learning is difficult to generalize and use. At present, knowledge engineering, expert systems or knowledge maps all adopt a coding mode to organize human knowledge by a method which can be identified by a computer. However, these systems make it difficult for machines to learn and generalize autonomously, so that in the face of differentiated scenarios, machines cannot autonomously generate new strategies and methods. So these systems have so far been applicable only to a certain area and range and have not been able to generate human-like intelligence.

The invention is based on the patent invention application with the application number of 202010400739.7 and the name of 'a method for realizing general machine intelligence by imitating human memory' of the same applicant, and further explains the details of the implementation method.

Disclosure of Invention

In patent application No. 202010400739.7 entitled "a method for modeling human memory to implement general machine intelligence", a method for building a network of relationships between things through memory is disclosed. In the present application, we further deepen how to build strong artificial intelligence (general artificial intelligence) by building a relationship network in memory.

In the present patent application, we propose a conceptual diagram of the composition of a machine. In fig. 1, the S101 module is a sensor module of the machine. In order for a machine to produce a similar cognitive pattern as a human, the S101 module needs to employ one or more generic sensors: visual sensors, auditory sensors, taste and smell sensors, touch sensors, gravity direction sensors, attitude information sensor information and the like, and sensors for specific applications (for example, infrared sensors, laser radar sensors and the like can be added for automatic driving). Machines also require the use of sensors for monitoring their own condition, which sensors are also part of the machine's sensory information. S101 is a module mainly composed of sensor hardware and software corresponding to the sensor, and aims to sense information outside the machine and information of the machine through the sensor. The difference and size of these sensor types do not affect the claims of the present application. Since all sensor data are processed in the same way in the present application.

In fig. 1, the S102 module is a simplified module of the machine inputting information to the sensor. The simplification of the machine to the input information mainly refers to the extraction of bottom-layer characteristics of the machine to the input information. It may employ any existing feature extraction method including, but not limited to, convolutional neural networks, image segmentation, contour extraction, downsampling feature extraction, etc. Any existing machine image recognition algorithm may be used in the S102 module.

However, the difference between the S102 module and the currently mainstream machine algorithm is that: 1, the S102 module is not intended to identify a particular thing. In the currently popular neural networks, a machine performs data processing on input data layer by layer, and then optimizes data processing parameters through error back propagation, with the goal of achieving minimum errors under large sample statistics. The algorithm implements a mapping of data space to label space. In the present application, however, the goal of the S102 module is to extract local common features in the input data. In the patent application No. 202010400739.7 entitled "a method for modeling human memory to implement general machine intelligence", we propose a method for repeatedly extracting input data using sampling windows of different sizes and different resolutions, and associating the data by placing them adjacent to the memory. And through a memory and forgetting mechanism, the ubiquitous connections are strengthened, and the accidental connections are weakened. Therefore, in the present application, the purpose of the S102 module is to find those local common features that exist widely, rather than finding the mapping relationship between a specific sample space to a label space. In the present application, the same input sample space may contain a large number of "local common features", which are extracted at different resolutions, respectively. It should be noted that the combination of the local features of things is also a local feature. The local feature and size have no relation, but refer to a part of information of things extracted at different resolutions. Some images may have local features as large as the image itself, but have low resolution and only contain partial information of the original image, for example, it may contain only the composition of other local features of the original image. For example, in an image, the local features may include bottom layer geometric features such as contours, straight lines, curves, textures, vertices, vertical, parallel, and curvatures at different resolutions, as well as features such as colors and brightness at different resolutions, and may also include motion patterns at different resolutions, as well as a combination topology of the bottom layer geometric features. The current popular deep convolutional neural network is to find the mapping relation between the same input sample space and a small number of specific labels. In block S102, a deep convolutional neural network may be used as an applied algorithm to implement mapping between input data and local common features. Such an algorithm does not belong to the claims of the present invention, but it is within the scope of the claims of the present invention to find multi-resolution local common features from the input data and use these multi-resolution local common features to establish a connection relationship between things.

In the S102 module, extracting multi-resolution dynamic features is further included. Similar to the image feature extraction, the S102 module also extracts local common dynamic features. Local common dynamics here refers to basic motion patterns such as wobble, circle, line, curve, wave, etc. that are widely present in our world. It does not map between a large amount of motion sample space and labels that specifically represent dynamics (such as dance, running, parade, binge, etc. label space), but rather input sample space to similar underlying dynamics features that exist extensively in our world.

In particular, dynamic features are the basis for knowledge generalization. Human analogy applications (generalization) of knowledge must be associations established based on some similarity. The similarity may be static similarity (such as appearance similarity or abstract feature similarity) or dynamic similarity (such as motion similarity or abstract feature variation similarity). While the dynamic features themselves may be represented by abstract particles or volumes, the motion features may serve as a bridge between empirical generalizations of different things.

In block S102, the machine also processes other sensor input data in a similar manner, including extracting static multiresolution features and dynamic multiresolution features. For example, for speech, the basic speech and speech rate part can be used as a static feature, and the change of audio frequency, tone and speech rate is a dynamic feature. The machine sliding samples the speech according to time windows of different lengths, which corresponds to different time resolutions. Machines need to extract static and dynamic features at different temporal resolutions and different detail resolutions. In the strong artificial intelligence implementation method, the extraction of dynamic features of objects by adopting multiple resolutions is a crucial part. The multi-resolution extraction method of dynamic features is described in patent application No. 202010400739.7 entitled "a method for simulating human memory to implement general machine intelligence", and is not repeated here.

The S102 module is a sensing layer feature extraction and is important for multi-resolution extraction of static and dynamic features of input data. Because of the connections between things, are different at different resolutions. The similarity between things is also different at different resolutions. The machine needs to build a relational network of things at different resolutions. Two things that do not meet in daily cognition may have similar attributes at different resolutions. These properties are the bridges for generalization of related knowledge.

The S102 module is a sensing layer local feature extraction, but it does not extract all multi-resolution local features every time. But rather the use of those resolutions and extracted intervals of emphasis is determined according to the search target of the machine for the sensor data. And the machine's search for sensor data is targeted from the anticipated targets that the machine generated in previous activities.

S102 is a processing layer of machine perception layer information and is a software layer for simplifying sensor input data. The input of S102 is data collected by the sensor, and parameters sent from the machine cognitive layer. These parameters are the range of resolutions that the cognitive layer needs to adopt and the range that needs to extract the emphasis. These parameters are determined by the machine from the size and attributes of objects that have been processed by the machine from previous information and are further identified as needed, and are part of the machine's response to the input information.

In S102, various specific algorithms such as convolutional neural network, cyclic neural network, image filtering processing, etc., which are currently popular, may be employed, but their output targets are local common features, rather than being directly mapped to a specific classification space. And local common features to a specific classification space are accomplished by the cognitive layer.

In the application of the invention, the assumption that the information of multi-resolution feature extraction and time adjacent input has a connection relation with each other is a key step for establishing a cognitive layer. In the S103 module, information connection relation (relationship network) in memory is optimized by associating activation, memory and forgetting mechanisms, which is to establish a cognitive layer.

In the present application, we propose a basic assumption: "the sensor groups have a connection relationship between information inputted adjacently in time". This is the key assumption we propose for cognitive layer establishment.

The machine stores the input information in the order of input. Such information includes motivational data such as external sensor data, internal sensor data, demand and mood data. External sensor data is the machine's perception of outside information. Internal sensors are the machine's own items of monitored information. The demand is an incentive and state of motivation for which the machine is preset. Mood is also a motivation and state of motivation for which the machine is preset. The needs and emotions are part of the preset motivation for the machine.

In the present application, the machine may be given various motivations, which are sources of power to drive the machine in response to input information. They are the control mechanisms behind machine behavior. In the present application, we take the example of giving needs and emotions to a machine to illustrate how the machine decides its behavior according to these incentives. The motivation that the machine may be endowed with includes not only needs and emotions, but may also include other types of motivation. The difference and magnitude of these motivation types do not affect the claims of the present application. Since all types of motivational data are treated the same in the present application.

In the methods presented herein, machines employ symbols to represent the various underlying requirements that various humans impose on the machine. For example, the safety requirements of the machine itself, pursuit of pleasure, desire to obtain human approval, desire to obtain human respect, further, for example, self-reward (sense of achievement) brought by self-goal achievement (goal achievement) of the machine, for example, unknown curiosity of the machine for exploration, and the like. These requirements can be represented by a symbol, and the symbol can be assigned a value to represent the state. The difference and magnitude of the demand type do not affect the claims of the present invention. Since all requirements are treated the same way in the present application.

In the method presented in this application, the machine employs symbols to represent the underlying emotions that various humans impart to the machine. The machine's emotions can be varied and each type of emotion can be represented by a symbol that can be assigned a value by the machine to indicate the state in which it is. The difference and magnitude of these mood types do not affect the claims of the present patent application. Since all emotions are treated in the same way in the present application.

In the method provided by the application, the relation between the emotion of the machine and the demand state of the machine can be related through a preset program. The parameters of the preset programs can be self-adjusted by the machine in the learning process of the machine according to the principle of 'tendency towards interest and avoidance from harm'.

In the method provided by the application, the emotional state of the machine and the explicit expression mode of the machine emotion can be connected through a preset program. The parameters of the preset programs can be self-adjusted by the machine in the learning process of the machine according to the principle of 'tendency towards interest and avoidance from harm'.

In the S103 module, it is necessary to establish a connection relationship between external information, memory information, machine internal sensor information, and motivation-related symbols and states of the machine at different resolutions. This connection needs to reflect the common knowledge of the world we are in, and is called a relationship network in the present application.

In the world where people are located, relationships among things are complicated, and it is very difficult to establish various relationships among things manually, and it is also difficult to quantify and flexibly apply the relationships. In the present application, we extract the relationship between things by memory. First, we use the concept of mirror space to store the extracted multi-resolution information. Mirror space means that we store information in the temporal order of the input information, and for simultaneously input information (such as images), the information is stored in the corresponding original spatial organization. The specific method comprises the following steps: when the machine extracts multi-resolution information features from the input, the machine needs to use these features to create a mirror space. The machine firstly adjusts the position, angle and size of the bottom layer features according to the position, angle and size with the highest similarity with the original data by scaling and rotating the extracted features, and places the extracted features and the original data in an overlapping mode, so that the relative positions of the bottom layer features in time and space can be reserved, and a mirror image space is established. The machine stores in memory multi-resolution feature data organized in the most similar way to the original data, which we call the mirror space.

The information stored in the memory has the memory value of the information. The new memory stored in the memory, including the memory value of the demand symbol and the memory value of the emotion symbol, are related to the activation value possessed by the corresponding information (symbol) when the storage occurs, usually in a positive correlation, either in a linear or non-linear relationship.

In order to establish a relationship network in memory, in the present application, we propose a basic assumption: "information input by the sensor group adjacently in time has a connection relationship with each other". This is a key assumption for our establishment of a relational network. Meanwhile, we propose three other assumptions for the optimization of the relational network: the "proximity relation" assumption, the "similarity relation" assumption, and the "memory strength relation" assumption.

The "proximity relation" assumes: let us assume that the memory information inputted adjacent in time are related to each other in the memory. Adjacent refers to temporal adjacency in memory storage. The "similarity relationship" assumes: in memory, similar memory information is also related to each other. The "memory strength relationship" hypothesis: among the memories, those with high memory values are more easily activated.

When a message in memory is activated, it activates other messages using the "close activation" principle, the "similar activation" principle and the "strong memory activation" principle.

"proximity activation" refers to the activation of a particular message in memory that activates the message in proximity to it (i.e., a message that has a proximity relationship).

"similar activation" refers to a specific feature in memory, and when receiving activation signals from other features, the receiving ability and the similarity between the features are positively correlated. This is a directional reception capability. So that a similar memory will send its own activation signal after it is activated and may easily further activate other similar memories. This is because similar memories have a strong ability to receive each other's activation signal. In the present application, a simple activation value transfer relationship may be that the transfer coefficient is proportional to the similarity. Of course, other transfer functions may be used, but the principle must be that the transfer coefficients are positively correlated with the degree of similarity.

"strong memory activation" means that the higher the memory value, the stronger the ability to receive activation signals from other features. Those deeply remembered information are more easily activated. In the present application, each memory information is assigned a memory value for indicating the time that can exist in the memory. Those with high memory values may exist for a long time and have a strong ability to receive activation signals from other features. This is how many synapses mimic the human brain represent memory strength, whereas memory that assumes many synapses is more readily activated by acquiring more activation energy from the surrounding environment.

The three activation modes are collectively called associative activation.

In the application of the invention, a memory and forgetting mechanism is adopted to maintain the memory value of the information in the memory base. The memory and forgetting mechanism is a relationship extraction mechanism widely used in the application of the present invention. In the application of the invention, the information in the memory is considered to be used once every time the information is activated, so that the memory value is increased according to the memory curve of the memory bank in which the information is positioned. Meanwhile, all memories decrease the memory value according to the forgetting curve of the memory bank in which the memory bank is positioned. The memory function means that some data increases as the number of repetitions increases. The specific increasing mode can be represented by a function, and the function is a memory function. It is noted that different memory functions may be employed for different types of data. The forgetting function means that some data decreases with increasing time. The specific reduction mode can be represented by a function, and the function is a forgetting function. It is noted that different forgetting functions may be employed for different types of data. The memory and forgetting mechanism refers to using a memory function and a forgetting function for memory information.

If we consider the memory as a three-dimensional space containing innumerable information, then the relationship network is the context in this space. These veins appear because of memory and forgetting mechanisms, relationships that cannot be repeatedly activated are forgotten, and relationships that can be repeatedly activated are strengthened. Multi-resolution information that are connected by coarse context of relationships constitute the concept. It links the image, voice, text or any other expression form of the same kind of information. Because these expressions appear frequently together and frequently translate into each other, the connections between them are tighter. Since human beings use languages very frequently in the future, the number of activations of a language is likely to be the largest and the memory value of the language is the highest in a concept. Meanwhile, since the language is usually connected with all attributes of a concept, the language is a mutually activated bridge between the attributes. So it behaves as if language were the center of our conceptual thinking. However, in the patent application of the present invention, it is considered that the language needs to be reconstructed by the language to obtain its true corresponding meaning, so that the activated information stream after being reconstructed by the language is the information really bearing our thinking.

The connection relationships in the relationship network are linked by "proximity activation", "similar activation", and "strong memory activation". The tightest local connection relationship forms a concept (comprising a static feature map and a language thereof under multi-resolution, a dynamic feature map and a language thereof); a bit looser than the concept is experience. Experience is the cognitive relationships that frequently recur. Just as it can be repeated, the memory connections between each other can be increased step by step. The common experience that is common in human cognition is common knowledge. Memory is more incompact than experience.

In the present application, we propose a memory organization form that can simply express a relationship network is: the information is stored in the order of the input time. Those "input temporally adjacent relationships" are expressed with "storage locations spatially adjacent". The adjacency of information in the storage space may be the adjacency in physical location: that is, time-adjacent information is stored on adjacent memory cells. The information may also be logically adjacent in storage space: it is stored with logical locations contiguous and a specific physical storage unit location is represented by a mapping table between logical locations and physical locations. Another method is that each stored message has its own stored time coordinate, and the machine determines the neighboring messages by searching the neighboring time coordinates. There are of course other storage means, but they must be able to express temporally adjacent information.

In the present application, we propose a way to search for similar information: one approach is to use a special similarity-contrast calculation unit to handle the task of finding a memory from memory that is similar to the input. The similarity-contrast calculation unit may be implemented using hardware or may be implemented using software. Either as a single module or as a module integrated into the whole arithmetic unit. The similarity comparison is a well-established algorithm and will not be described in detail here.

When activation is approached, the information activation value transfer coefficients close to each other are large.

When the similar activation is carried out, the similar information transmits an activation value through the similar activation, and the transmission coefficient and the similarity are in positive correlation.

When the memory is activated, the information with high memory value is easier to be activated by obtaining large activation value from the adjacent activation. Therefore, one particular instance of activation is strongly remembered as being imminent.

In the above-mentioned "close activation", "similar activation" and "strong memory activation", the transfer function of the activation value needs to be determined by practice, but the basic principle is that the transfer coefficient of the activation value of the "close activation" is inversely related to the time distance when the two information stores occur. The time interval can be considered a medium that attenuates the propagation of the activation values. And the activation value transfer coefficient of "similarity activation" is positively correlated with the similarity between two pieces of information. In the case of "strong memory activation" which is activation proximity, the activation value transfer coefficient needs to take into account the memory value of the information receiving side in addition to the attenuation of the time interval. The memorized value of the party receiving the information can be considered as the size reflecting the ability to receive the activation value.

It should be noted that a same object may have a large number of features at different resolutions to represent the attributes of the object. The similarity between two things at different resolutions is not the same. When a feature (attribute) of a thing at one resolution is activated, the feature can activate the feature of the thing at other resolutions through adjacent activation, and can also activate other features similar to the thing at the current resolution through similarity, so that other attributes of other things are further activated through adjacent activation. Therefore, whether or not similarity is established on the basis of the specified resolution. This is why we need to extract multi-resolution features simultaneously on the input information.

In the present application, the specific activation value transfer function does not affect the basic working principle of the machine. However, the optimal activation value transfer function for each type of activation needs to be determined by practice, but all of them must comply with the above-mentioned principle.

It is further noted that the activation values for information acquisition in the relational network fade over time. The fade out curve needs to be optimized by practice. The length of the fade-out time needs to be balanced between the connection between the pre-and post-activation information and the activation status of the newly entered information. If the fading time is too long, the activation state brought by the new input information is easily masked by the original activation state, and the connection relationship between the new information and other information cannot be clearly expressed. If the fading time is too short, the connection relationship between the activation information before and after is easily ignored.

In relational networks, static concepts are analogous to the small parts widely used by machines, while those dynamic profiles (including concepts representing relations) are analogous to the connectors widely used. While those represent a class of processes that are organized in a certain temporal and spatial order of a plurality of small parts (static objects) and connections (dynamic features). They can often be used for empirical simulation. Dynamic profiles (including concepts representing relationships) can often be used as a tool for empirical generalization because they play a role as abstract relationships that abstract things change, being a common attribute across concrete things. The essence of the generalization process is the process of applying an existing analogy to experience through common attributes. The common attributes may include certain features at multiple resolutions (e.g., certain similar features at low resolution, or one or more similar features at high resolution), and may also include similar attributes (e.g., language, analog dynamic patterns) connected by a cognitive layer as a bridge. However, these generalization relationships and bridges can all be embodied in the association activation process of the relationship network, and are generalized and applied through the prediction, decision and execution system, so that how generalization is implemented is not specifically described here.

In the memory space, external information, internal information, demand and demand states, mood and emotional states, as well as specific motivation and corresponding motivational states that may be added for other applications of the machine, are stored in memory directly maintaining a temporally adjacent order. The memory values of the two are maintained according to a memory and forgetting mechanism, and the associative activation is carried out according to a near activation principle, a similar activation principle and a strong memory activation principle. So in the present application, the processing of such information is similar. Their type, size and data format do not affect the claims set forth in this application.

In the present application, another feature of the memory storage is to store motivation and motivation status data (such as demand and demand status data, emotion and emotion status data) of the machine into the memory, so that the information is identical with other information and an activation value transmission path is established between the information and the other information. That is, when a memory message is activated, it may deliver activation values to many of the demand and emotion symbols in the associated memory through chain activation. The magnitude of the transfer coefficient is the direct or indirect connection strength in the relationship network after the optimization through a memory and forgetting mechanism. The nature of the relationship network is a causal network that reflects the causal relationship between two pieces of information. The information adjacent in time usually has causal relationship, and the similar connection connects the causal relationship connected in different time in series to form a large causal relationship network.

In the S104 module of fig. 1, we need to establish an application layer that utilizes a relational network. The application layer comprises:

1, an initial activation value assignment system.

The input data of the sensor is first simplified in step S102. The machine then uses these simplified information features to assign initial activation values to these input information using an initial activation value assignment routine. The initial activation value assignment program is a preset program, and input parameters of the program comprise input information, motivational states such as the demand and emotion of the machine at the moment. Its output includes an initial activation value assigned to the input information.

The machine may also repeatedly assign activation values to the input data after simple processing of the data. This step is one of predicting, deciding, and executing the response result output by the system.

2, associative activation.

After the machine gives an initial activation value to input information, the machine carries out chain activation by adopting a 'near activation' principle, a 'similar activation' principle and a 'strong memory activation' principle. The machine implements the associative function by chain activation, so we also refer to associative activation. The machine looks for (a) experience associated with the input information by associating activation. (b) There are connected motivations and states of motivation (including demand and demand states, emotions and emotional states) with these related information. (c) And other types of information that may be relevant to the presence of the input information in memory.

The information activated by the machine through association is usually memorized with a higher memory value. These memories are usually higher memories obtained by a memory and forgetting mechanism because they can be activated repeatedly. These memories can be activated repeatedly because such information relationships can be reproduced continuously in our lives. They are the "experience" acquired by the machine. Therefore, the memory and forgetting mechanism is an optimization mechanism of the relationship network and is the basis of intelligence generation.

The machine activates the memory based on the input information, which may include all types of memory in the memory. For example, when speech is entered, the machine may activate images, text, feelings, emotions, related other speech, or a past memory that is closely related to the languages based on similar speech in the memory. The specific activation depends on: (a) the machine learns the relationship network obtained by the experience and learning the parameter settings. (b) Chain activation parameter setting of the machine (this is equivalent to setting the associative way of the machine). (c) The initial activation value assigned to the input information by the machine. The greater the initial activation value, the more content the machine can activate.

And 3, activating information reconstruction.

Through associative activation, the machine obtains experience associated with the input information. These experiences require further processing to which the machine can respond to the input information.

And (3) environment reconstruction: after a machine enters an environment, specific things, scenes and processes are identified by extracting images, languages and other underlying features of the sensor input. And the same thing characteristics, scene characteristics and process characteristics found in the memory are overlapped with the similar parts in reality, so that the machine can presume the parts which are not seen by the current thing, scene and process temporarily.

Since the size of the object is one of the low resolution features, the machine also uses the size of the particular object in the field of view to compare to the normal size of the object in the feature map to assist the machine in establishing depth of field in the environment. Machines typically resize the reconstructed environment based on their target (typically the conceptually related image with the highest activation value).

The three-dimensional environment is reconstructed by a plurality of local information (including approximate frames and different details of a kind of things at different resolutions), which is a technology that is mature at present and is not described in detail here.

Self-reconstruction: by overlapping a plurality of pieces of memory about self information (including memory of self at different resolutions, some memory about the whole frame and some memory about details), similar parts are overlapped to construct a stereo image about self, and the self-mirror reconstruction is performed. Because of the memory of the machine about the self information, some memories are visual, some are tactile, some are emotional, and some are requirements, and when the common parts of the memories are overlapped, the whole self whole perception reconstruction is formed. The information is integrated into the whole image by overlapping similar parts. We created a self-image in mind as if we were able to see the action of the self-image. This self-image is part of our self-awareness. It is the basis on which machines distinguish between self and non-self.

After an external mirror image space and a machine mirror image are established, the machine is aware that the machine distinguishes the machine from the outside, and a behavior mode of interaction between the machine and the outside is determined according to a mode of 'driving towards interest and avoiding harm'. The nature of consciousness is a way of behavior. It is just because of the differentiation between self and non-self that the machine only generates consciousness and links self with the external things. The essence of this connection is the relationship between the external things and their own "interest" and "harm". This relationship is built up step by step during the learning process of the machine.

After the input information activates many of the relevant memories in the memory, the machine uses these memories to create the mirror space and the machine's own mirror, and uses the mirror space and the machine's own mirror as a way to combine these experiences. The machine views these memories from a third person's perspective. The memories contain relevant information such as emotion, demand and result of the machine, and the machine uses the information to plan the response of the machine to the input information according to the mode of trend interest avoidance and by taking reference to the experience, and can evaluate the possible results of the response for many times.

4, bottom layer motivation of the machine.

The source of power to drive the machine is the motive of the machine, which may be summarized as "hedging". The part of the 'benefit' and 'harm' is preset; some are established in acquired learning because they are related to the needs of the machine itself. By analogy with humans, for example, "water", "milk", "food" is originally preset "interest", and then we have learned the association between "test score", "banknote" and our innate needs, and then we have found that the subject of the operation can also be something without entity such as "love" and "time", and even we have pursued dominance in the population, which is an extension of the dominant interest avoidance in the underlying motivation in our genes.

In a similar way, we can also give machines the motivation that humans want them to possess. Because in the relational network, all memories are stored, and the requirement symbols and the corresponding memory values of the machines at that time are simultaneously stored. These memory values are positively correlated with the state values of the demand symbols at that time. For example, if a machine is responsible after some action. Because the responsibility is a loss (the cognition can be preset, can be expressed by the language of the trainer, and can be realized by directly modifying the relationship network), and the degree of responsibility (such as a word representing the degree in the language) brings different loss values to the machine. The stronger the accountability, the higher the memory value that the machine assigns to the symbols lost in this memory. Then in this memory, since the memory value of the loss symbol is higher, all other feature maps with higher memory values in this memory frame have stronger connection with the loss symbol. If similar actions send out an object or receive an object in a similar environment and similar responsible behaviors occur again, the characteristic diagram and the loss symbol which bring loss in the memory frame are repeated, and the memory values of the characteristic diagram and the loss symbol are increased according to the memory curve in the memory frame, so that the relationship between the characteristic diagram and the loss symbol which bring loss is increased. Through repeated times, the relationship between the characteristic diagram actually bringing loss and the loss symbol is selected according to a memory and forgetting mechanism. It is unclear from the outset as to why a subject is being blamed, and it is clear to the latter as to what has brought about the cursory consequences to the subject. This process is similar to the learning process of human children.

Even if the machine is not fed back in time when the action takes place. The trainer may also at a later stage point out the behavior itself and send feedback, thus linking the behavior and the result in a single memory frame. Even the trainer does not need to indicate which behavior is good or bad, and the machine can gradually establish the connection relationship between the correct behavior and the required value by memorizing and forgetting only by receiving correct feedback each time.

Therefore, only basic underlying motivation needs to be given to the machine, and the machine can establish the relationship between the physical and color objects and the underlying motivation according to external feedback. These relationships are the basis for the decision making by the machine. The underlying motivation of the machine is relatively simple and can be in a preset manner. Such as giving the machine motivation to learn and comply with human laws, asking the machine for human approval, giving the machine motivation to avoid danger, giving the owner motivation to protect safety, etc.

The motivation and motivational state of the machine are closely related to the emotion. In the application of the invention, the emotion is determined by using motivation and motivation states through a preset program, and the relationship between the emotion and emotion appearance is realized through the preset program. That is, the emotion of the machine is expressed by the action (expression and body language) and the language (language output manner) using a preset program. But at the same time, mood or expression of mood can also be adjusted by the motivation of "driving towards and avoiding harm". The machine, through learning and feedback, gains in what circumstances, what emotions can bring in revenue and loss, and in turn adjusts the emotions or emotional expressions. Just because of the close relationship between emotions and motivations of machines (including motivational states), motivations of machines can also be expressed uniformly in terms of emotional needs. The machine may make the selection and respond with pursuit of emotional needs.

It is specifically noted that the storage of any input information by the machine stores both motivation and motivational states (e.g., demand and demand states, emotional and emotional states, etc.) of the machine. In memory, motivation is represented by a symbol, and the memorized value is positively correlated with the activation value at the time of memory. Those incentives to obtain high activation values may be long-term memory, typically because of the high memory values obtained when their storage occurs. These long-term memories may be activated one time at a time, thereby affecting the decision-making and behavior of the machine over the long term. And the information related to the motivational state with high activation value (such as strong emotion, large income or loss and the like) can be reversely activated by the high activation value of the motivational symbol with high activation value because of being adjacent to the motivational state with high activation value in time, so that the activation value of the user when the user stores the information is improved, and the memory value obtained when the user stores the information is also improved.

And 5, reconstructing language information.

Language plays an important role in machine intelligence. Language is a set of symbols that humans create for better communication experience. Each symbol represents a particular thing, process, and scene. When a language is entered, the associated memory represented by the language is activated. These memories may have both information about the language itself and memory related to the use of the language activated (e.g., speech emphasis or text emphasis to emphasize an emphasis such as an irreconcilable voice or a jeopardy tone, etc.). The activated information constitutes an activation information stream. To balance the context of the language and the current semantic recognition, the activation value of the activated information may decay over time. The parameters of the decline are related to the motivation and state of the machine (such as demand and demand state, mood and emotional state).

The chained activation of the languages enables context-dependent recognition of all input information. The input information here includes both the environmental information and the activated memory information. The mutual activation and assignment of the information embody the context association. This correlation is more extensive than statistically generated semantic library content. It does not relate to only speech, but also to all sensory inputs and associated memory. So through the multi-resolution information formed in S102, the common sense network formed in S103, the machine can achieve the connection of language to static and dynamic images, feelings, demands and emotions, as well as the connection of language to related language and memories. When such a connection is incorporated into the understanding of the language input by the machine and is responsive to the understanding of the language, and associated experience, it is apparent that the machine actually understands what the input language really is.

The language input constitutes an input information stream and the corresponding activation memory also constitutes an activation information stream. The machine needs to reconstruct this active information stream when understanding the language. The reconstruction is to reconstruct the environment information therein, and an imagination process is formed by overlapping the same parts of the activated environment-related information (such as images, sounds, feelings and the like).

Similarly, the machine also needs to integrate information about feelings, emotions, vision, actions, limb states, etc. activated by the input language with the existence of the machine itself. This information will also typically activate similar experience of the machine itself. The machine can thus realize information related to feelings such as feelings, sight, emotion, movement, or body state, etc., which are brought about by these languages.

6, Generation of machine prediction capability

The nature of the prediction is a statistical behavior. The prediction of the machine is to estimate various possibilities and corresponding probabilities of the development of the object or various possibilities and corresponding probabilities of the behavior of another person based on past experience or similar experience.

When information is entered, the machine does not need to exhaustively predict all possible outcomes, which is also an unfulfilled task. The machine need only evaluate those results that are activated and that may occur in the experience associated with the input information. This is equivalent to using common sense to define the range of possible outcomes. Within this limited scope, the machine may employ any artificial intelligence prediction method to infer the probability that the current thing has progressed to every possible probability, such as monte carlo search, decision trees, bayesian estimation, machine-based inference, etc.

In past experience, each possible outcome has been linked through a relationship network to a demand state and an emotional state, which represent the "interest" and "harm" that such outcomes may have on the machine. Thus, by combining the probability of occurrence with the associated demand and emotional state, the machine can infer the magnitude, type, and probability of occurrence of "interest" and "harm" that each outcome of the development of the thing may bring to itself. Therefore, the machine needs to synchronously take the probability of possible occurrence into consideration when predicting the 'benefit' and 'harm' brought to the machine by the possible occurrence. Therefore, the machine should evaluate both "interest" and "harm" and also evaluate their corresponding probabilities, and combine both to make a decision.

Decision and response system for machines

After the machine inputs information, a related decision range is limited according to the memorized activation state in an associative chain activation mode. The machine's evaluation and response to the input information is searched, evaluated and responded to based on the input information and the range defined by the activated memory. This is equivalent to using common sense to define the scope of the search, evaluation and response required.

Within this context, there may be one or more pieces of memory associated with the input information. The machine may simulate these past experiences by means of a piecewise simulation to establish a possible response process. Then, within this limited range, the machine can adopt any current artificial intelligence prediction method, such as Monte Carlo search, decision tree, Bayesian estimation, machine inference based on rules, etc., to select its own decision and response according to the predicted probability of "interest" and "harm".

Because the purpose of the machine is 'interest and harm avoidance', the basic starting point of the response of the machine to the input information is to make the response of the machine according to the past experience, so that the occurrence probability of the things generating 'interest' is increased as much as possible, especially the situation that a high profit value can be obtained. But to reduce the probability of occurrences of "harm", especially in situations where significant loss is incurred. Therefore, the machine combines own response according to experience under the motivation of balancing advantages and disadvantages to achieve the aim of 'driving toward interest and avoiding harm'.

The decision of the machine is a path planning method based on the prediction capability of the machine. The goal of the path is to minimize losses, if at all. With predictive capabilities, the machine transforms decision making and response, a fully open problem, into a string of relatively closed problems of how to increase or decrease the probability of a range of things occurring. Since the knowledge is established in the previous steps, the conditions related to each event (which is the cause of the cause and effect relationship) can be obtained through the relationship network when the event occurs. The causality relationships with strong association are strongly connected in the relationship network because of repeated occurrence. The relationship network can express the causal relationship layer by layer.

The goal of each step of decision making is to make the development direction of the things "favor and avoid harm". This may be a process of interaction with the outside world. The interaction itself is a method to promote the development direction of things to be beneficial and harmful according to past experience. The probability of the event with high profit value is continuously improved through the information and the behavior obtained by interaction, and the probability of the event with high loss value is continuously reduced. This is an iterative process. But the manner in which each step is processed is the same. And the machine increases the probability of the occurrence of the events leading to high profit values layer by layer on the basis of the causal chain. This is similar to the chain activation process, activating those events that lead to high revenue paths step by step, while carefully avoiding those events that may lead to high loss values.

Since the probability of causal connection between paths is expressed by a relationship network, the response planning problem of the whole machine becomes the problem of finding the optimal path in the causal chain network, which is the problem that the current machine intelligence has solved. By way of example, the machine can determine the prior probability of an event (such as an event that brings a high value of gain or a high value of loss) by searching the memory. The causal strength (a posteriori probability) between a certain condition and the event can then be determined by means of a relationship network. The connection strength between different conditions in the relationship network can reflect whether the different conditions are independent or not. The machine can predict the probability of the event by only selecting a plurality of relatively independent conditions through a naive Bayes algorithm. The machine may determine its own response based on the calculated probability. These responses may take various forms, such as: the probability of the occurrence of the event is improved, or the probability of the occurrence of the event is reduced, or the probability of the occurrence of the event is not influenced. Depending on the value of the return and loss to the machine. And increasing or decreasing the probability of the occurrence of the event, and further planning to increase or decrease the probability of the occurrence of the condition related to the occurrence probability of the event. This process is essentially an iterative probabilistic path search problem.

For example, the following steps are carried out: if the response of the machine is further to determine possible gains and losses. First, the machine takes as a priori probabilities, in memory, probabilities of various possible outcomes, under conditions similar to the current situation. Various revenue values and loss value occurrence probabilities are then calculated based on the conditions associated with each outcome and the posterior probabilities between the outcomes. The machine then generates the next target and further determines the probability of each condition occurring. For example, the response of the machine at this time may be (a) searching for and counting the posterior probability between each condition and the corresponding occurrence of the profit and loss values. And then used to update the overall gain and loss assessment. This can be done by searching for the connection strength between the relationship networks. (b) Further updating the probability of a certain condition occurring at present. Such as by directly querying the information source as to the probability that a condition has occurred or is likely to occur, based on mimicking past experience. Or by obtaining in other ways the probability of whether a certain condition has occurred or is likely to occur. Depending on the behavioral mimic memory obtained during machine learning. (c) According to the principle of driving toward interest and avoiding harm, certain conditions closely related to income and loss are used for promoting the occurrence of the disease or avoiding the occurrence of the disease as a new target. Under the drive of a new target, the same evaluation process is adopted to respond. With such iterative responses, the ultimate goal remains to gain revenue and avoid losses.

Therefore, when information is input, the causal connection determined by the relationship network, through the principle of driving toward interest and avoiding harm, and through the relationship between the events and the 'interest' and 'harm' established in the relationship network by the machine, the response of the machine which seems to be completely open to the information input can be changed into a multi-level target. These goals are all served by increasing the probability of certain events occurring, or decreasing the probability of certain events occurring. Therefore, through the causal relationship of the relationship network, the machine can convert the interest and risk-averted target into a series of targets which are associated with each other in a specific situation. These goals constitute a realization path for the machine to maximize revenue and minimize losses.

In this process, the machine may respond by continuously searching for new information, or continuously passively obtaining new information, and updating the target path with the posterior probability between the new information and the result. The possible external feedback prediction of the machine after responding to the machine also comprises the activation of two types of motivational state memory. One is the need and emotional state of oneself in recurrent memory, which comes from various feelings and emotions about oneself in activated memory. One is the demand and emotional state of the person himself when viewing a similar scene from the observer angle, which comes from the activated memory observing the various feelings and emotions produced by the machine of others under similar scene. Therefore, when predicting the 'profit' and 'harm', the machine simultaneously evaluates the 'profit' and 'harm' brought to the machine by an event from the view angle of the machine and the view angle of other people.

The predictive capabilities of a machine include not only predicting the "profit" and "harm" that an event may bring. It is also desirable to predict the responses that one or other might take driven by "interest" and "harm" and the effects that others might have on one's own "interest" and "harm" in response. These are obtained by counting the motivational state values of the related needs and emotional states in the relationship network. Therefore, the evaluation results of the machine are dynamically changed with more input information. The decision and response process of the machine is a dynamic path planning process. It is jointly driven based on empirical response and probability calculations based on profit and loss.

By the method, the machine can decompose an abstract profit-and-harm-avoidance target into a large number of tasks for improving or reducing the probability of certain specific events layer by layer in a layer-by-layer iterative decomposition mode under a specific input condition. These tasks can be subdivided layer by layer into very specific target tasks, for example, up to the underlying drive capabilities of the machine. This process is the decision and response system of the machine.

8, simulation ability

The mimic ability is the ability of a human to exist in a gene. For example, for a child to say an calandering, if we say "you are back" with his (her) callout every time he (she) comes home. After a few times, when he (she) comes home again, he (she) will actively say "you come back". This indicates that he (she) has begun to imitate others for learning without understanding the meaning of the information. Similarly, we let machine learning use the same approach. Therefore, the machine needs to have the emulation built into the machine as a kind of underlying motive. The machine is willing to imitate the behaviors of other people (other machines), and the machine is continuously improved according to the evaluation of the machine or the external feedback information, so that the coordination and consistency abilities of various senses, limbs, languages and actions of the machine are continuously exercised, and the learning efficiency is improved. At different stages of machine learning, we can give the machine different strengths of motivation for simulation. For example, when the machine learns the language and the action output, the machine can be directly endowed with a strong imitation motivation, and at other stages, the machine can be endowed with a normal imitation motivation.

When the machine obtains external voice or action input, the voice or action can activate the relevant memory of the machine. These memories may be a similar pronunciation, or a basic action fragment. These memories further activate sensory, demand and emotional information, language or action memories associated with these memories. The machine, driven by the simulated motivation, will make similar speech output or motion output by adjusting the underlying driving parameters in the experience through the decision making system based on these activated memories. And the bottom layer driving means outputting bottom layer experience by voice or outputting bottom layer experience by action. They are muscle-driven commands corresponding to a particular voice or action, where the parameters are updated through acquired learning and continuously through feedback.

Humans may preset the machine with some of the most basic speech or motion (including expression and body language) capabilities. The optimization of the parameters can be realized by subsequent learning and training, the results of the parameters and behaviors are combined by memory, and are continuously adjusted by an emotion and demand system (influenced by self or external feedback), and finally, under the drive of a bottom layer motivation, the machine obtains the relationship between different parameters under the excitation of different external information through a memory and forgetting mechanism to form memory. These memories are all the knowledge and skills of a machine in the face of external information input. They include behavioral habits of language, motion, expression, limb movements, and the like.

Humans may also give the machine a preset conditioned reflex system. The role of these systems is that humans expect the machine to respond under certain input conditions. Such as evasive action of the machine in case of emergency, or specific output action of the machine under specific information input (for example, these conditioned reflex systems can achieve the purpose of self-checking the machine, or emergency shutdown, or adjusting the working state of the machine, etc.).

9, carrying out the process

After having the above various basic capabilities, the machine can perform the response specifically according to its own decision. Such as speech output, motion output (including expression and body speech output), or other forms of output (such as output data streams, images, etc.). The execute response step is a process of translating the plan into actual output.

If the machine selects the voice output in the steps of selecting various possible responses, the machine is simple, and only the image feature map to be output needs to be converted into voice, and then the relationship between languages in the relational network (grammar knowledge existing in the relational network) is used for organizing a language output sequence and calling pronunciation experience for implementation.

It should be noted that machines may choose from experience (either self or others) to express the dynamics of an entire sentence (e.g., using different movement patterns of tone, or variation in stress to express questions, jeers, distrust, emphasis, etc., which are commonly used in humans. Because the machine learns these expressions from human life, any expression of a human can be learned by the machine theoretically.

The problem becomes much more complicated if the machine chooses to output motion, or a mixture of speech and motion. This corresponds to the tissue moving about. In response planning of a machine, there may be only major steps and final goals, the rest requiring random strain in practice.

The machine needs to respond to the sequence target to be output, and the sequence target is divided in time and space according to different time and space related to the targets, so that the execution efficiency of the machine is coordinated. The approach taken is by selecting as a group closely related targets in time and closely related targets in space. Because the dynamic characteristic diagram and the static characteristic diagram are combined to form an information combination, and the environment space of the related memory is provided with time and space information, the classification method can be adopted in the step. This step corresponds to a change from the general scenario to the sub scenario.

The machine needs to expand the intermediate targets in each link layer by adopting a segmented simulation method in combination with the real environment again. Response plans developed by machines at the top level are usually composed only using highly generalized process features and highly generalized static concepts (since these highly generalized processes find many similar memories, responses built from them are highly generalized). Below the total output response, such as "business trip," the "go airport" is an intermediate link target. But this goal is still very abstract and the machine cannot perform emulation.

Therefore, the machine needs to be divided according to time and space, and links needing to be executed in the current time and space are taken as the current targets. And temporarily putting other time and space targets to one side as inheritance targets. After the machine targets the middle link, the machine still needs to further subdivide the time and space (write the lower level script again). This is a process of increasing temporal and spatial resolution. The process of converting one target into a plurality of intermediate link targets by the machine is still a process of analyzing various possible results and possible occurrence probabilities by using decision-making capability and selecting own response according to the principle of 'benefiting and avoiding'. The above process is iterative, and the process of dividing each target into a plurality of intermediate targets is a completely similar processing flow. Until the underlying experience of the machine is resolved. The underlying experience is that for language it is the muscles that are mobilized to make syllables. For an action, it is decomposed into the issuing of drive commands to the relevant "muscles". This is a tower-like decomposition structure. The machine starts from the top level target and decomposes one target into a plurality of intermediate link targets. This process is to create virtual intermediate process targets that are retained if they are "on demand". If "not compliant," it is recreated. This process expands layer by layer, eventually creating a machine rich response.

In this process, the machine may be exposed to new information at any time, resulting in the need for the machine to process a variety of information, and these original goals become legacy motives. This is equivalent to the situation that new situations are encountered continuously in the process of organizing activities, and the problems need to be solved immediately, and the activities cannot be organized. The director then calls off other activities to resolve the problem that was encountered in front. After resolution, the activity continues. In another case, the director suddenly receives a new task during the process, and then decides to suspend the activity after the director balances the interest and the disadvantage, and processes the new task preferentially.

The machine breaks down other objects to more detailed objects while performing the emulation tasks that can be performed. The machine is thought at the same time. This is because the reality is very different and it is impossible for the machine to know the external situation in advance and make a plan. It is the process of an environment and machine interaction to accomplish an objective.

Thus, the machine can complete understanding and responding to the input information once by utilizing the capabilities. This process acts as a minimum period for the machine to interact with the environment. The machine is continuously repeated to use the process to achieve a larger target, which is represented by a continuous interaction process of the machine and the outside world and is represented by machine intelligence.

Experience summary of the machine 10

The experience of the machine, not only through the memory and forgetting mechanism to form connections in the relational network, can also actively strengthen such connections. Such actively reinforced connections can take many forms: such as learning the experience of others through language. The activation information stream formed by the language, together with the language, constitutes the experience of the others who have learned. This experience is stored in the memory as new input information, which are part of the memory. As another example, the machine actively changes those memories into long-term experience by repeating the memories by linking closely related information to those in "interest" and "harm" relationships.

The machine can adopt a preset algorithm, the information which can repeatedly appear and can greatly influence the memory of the relationship between 'benefit' and 'harm' is memorized, the information of the corresponding events is sequentially and virtually input, the virtual processing process is repeated, the related connection in the relationship network is strengthened, and the experience is enhanced. In this enhancement, the connections between those common parts of similar experience may be progressively enhanced, so that experience becomes increasingly concise and generic, eventually forming some sort of machine-self-summarized rule. Therefore, the self-summary implementation of the machine can be realized by adopting a preset program and adjustable parameters. These adjustable parameters can be adjusted by the machine's connection relationship between ' interest ' and ' harm ' and specific things, which are gradually summarized in learning, and can also be adjusted according to the intensity and frequency of the machine being stimulated.

The S105 module of fig. 1 is a communication connection module of the machine. The machine can exchange data with other external machines (including computers) through the communication module according to a preset protocol. The machines may implement distributed computing through data exchange. For example, the machine in the field may process part of the data and may also transmit part or all of the information to the central brain, with the central brain's powerful information storage and processing capabilities to make decisions. The machines may also share memory through data exchange. For example, the experience obtained by one machine can be transmitted to other machines, and the memories of a plurality of machines can be fused, so that more complete experience is formed. Computing power may also be shared among machines, constituting distributed thinking capability. Through the communication connection module of the machines, cognitive sharing, decision sharing and behavior coordination capabilities can be realized among the machines, so that super artificial intelligence is realized.

Drawings

Fig. 1 is a schematic view of one possible component of the machine.

Detailed Description

The invention is further described in the following with reference to the figures and the specific examples. It should be appreciated that the present application text mainly proposes the main steps to implement general artificial intelligence. Each of these main steps may be implemented using presently known structures and techniques. The present document therefore focuses on these steps and their components and is not limited to the details of implementing each step using known techniques. The description of these embodiments is merely exemplary in nature and is in no way intended to limit the scope of the present disclosure. In the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the focus of the present application text. All other embodiments obtained by a person skilled in the art without making any inventive step are intended to be within the scope of protection of the present application.

Establishment of the sensor system: in fig. 1, S101 a sensor module of a module machine. In order for a machine to produce a similar cognitive pattern as a human, the S101 module needs to employ one or more generic sensors: visual sensors, auditory sensors, taste and smell sensors, touch sensors, gravity direction sensors, attitude information sensor information and the like, and sensors for specific applications (for example, infrared sensors, laser radar sensors and the like can be added for automatic driving). Machines also require the use of sensors for monitoring their own condition, which sensors are also part of the machine's sensory information. S101 is a module mainly composed of sensor hardware and software corresponding to the sensor, and aims to sense information outside the machine and information of the machine through the sensor. The data generated by the sensors of the machine, which are input to the processing unit of the machine, are referred to as input data.

The communication between the sensor data of the machine and the processing unit of the machine may be any existing communication system, or a customized communication system, and the specific communication form does not affect the implementation of the present patent application.

Establishment of a preset motivation system: machines have been represented symbolically as a type of motivation. Such as directly using language symbols or some sensory information data, or directly artificially specifying a symbol to represent some kind of motivation for the machine. For example, the symbol "Sa" is used to represent the safety requirement of the machine, and for example, the numerical value of 0 to 100 is used to represent the state of the safety requirement of the machine. Where 0 represents very little security and 100 represents complete security. For example, the "danger" symbol is used to represent danger, and H, HM, M, ML, L are used to represent the degree of danger from high to low, respectively. For example, a "smiley face" is used to represent a pleasant mood, and 0 to 10 are used to represent a degree of pleasure. In the present application, how many motivation types are established, what form of symbols are used to represent motivation types, and what form of symbols are used to express the state of the symbols do not affect the claims of the present application. Since all symbols and corresponding states are handled in a completely similar manner in the present application. For example, different symbols can be used to create different emotions for machines, such as excitement, anger, heart injury, tension, anxiety, embarrassment, tiredness, coolness, confusion, aversion, jealousy, fear, joy, romance, sadness, homonymy, and satisfaction. Each mood has its own quantification space. The motivation of the machine for benefiting and avoiding the harm is to balance the advantages and disadvantages in the space established by various motivation requirements and search an acceptable space range. And then empirically push the event to a space that is acceptable to itself. This driven strategy may be to directly drive the direct target, or may be a precondition to drive the realization of the direct target, thereby increasing the probability of the direct target occurring.

Multi-resolution information feature extraction system: the multi-resolution information feature extraction can adopt the current artificial intelligence perception processing capability to extract the features of the input information under different resolutions. The method difference with the current artificial intelligence perception processing is that: current algorithms implement a mapping of data space to label space. In the present application, however, the goal of the S102 module is to extract local common features in the input data.

The existing artificial intelligence perception processing mode is to optimize a mapping network by optimizing an error function through mapping a large amount of data to a label. The mapping is from data to specific labels, and the optimized mapping relation is closely related to the specific labels, so that the algorithm has no universality. In the present application, all input data is mapped directly to locally common features. While the local common features are a base feature library established by pre-training. The basic assumptions for building these base feature libraries are: the human evolution process is developed in the direction of high efficiency of computing power. Only in this way, the energy consumption can be saved to the maximum extent under the condition of improving the algorithm complexity to deal with the complex environment, thereby increasing the survival probability. The specific embodiment of the evolution direction is that the algorithm for extracting the widely existing local features is formed by nature. Because this allows the algorithms to be reused to the maximum possible extent, the energy efficiency ratio of the calculations is improved.

The reason why multi-resolution is required is that "similarity" is a fuzzy concept, and the degree of "similarity" between two things can be described only at a certain resolution established. In the evolution of humans, humans have concluded that there may be other similarities between things that are partially similar on our planet. We need to infer other possible similarities from the known similarities. Such as inferring a likely behavior of the dog from its appearance. Multi-resolution is to establish similarities between things at different resolutions. For example, at a coarse resolution, all dogs are similar. But at further resolution we would consider that there are differences between different dogs. So by coarse resolution we can generalize some common behaviors of dogs. For example, seeing a strange dog, we can activate the relevant memory by inputting the information characteristic of the dog. First, the dog's information is activated at a coarse resolution in all memories, which further activates other features of the dog. Then more detailed features of the input dog information will activate the memory associated with those detailed features, and the final state of the activation tallies is that both the public information about the dog and the detailed features of the dog are activated, thereby obtaining a preliminary information assessment of the dog. Other detailed features about the dog that are not consistent with the detailed features of the dog will not be activated, so that the information does not interfere with cognition. Similarly, when we see an animal that is similar to a dog, from a rough resolution we will activate information about life, activate information about the animal, activate common information about the dog, and activate information about the characteristics exhibited by that particular animal. For example, if its paw is much like a cat's paw, we might activate that it might scratch our prediction. Therefore, the extraction of multi-resolution features, especially the extraction of multi-resolution motion features, is the key to the generalization of knowledge. As multi-resolution action is widely present in different things. Multi-resolution action is a ubiquitous knowledge-generalization bridge.

The specific method of the multi-resolution extraction method is to compress data with different resolutions, and then repeatedly search for local similarity by adopting data extraction windows with different sizes. In the patent application No. 202010400739.7 entitled "method for simulating human memory to implement general machine intelligence", details of how to implement the multi-resolution extraction method are not repeated herein.

So, although we can use any specific feature extraction algorithm at present, such as data coordinate basis transformation and filtering process, such as convolutional neural network, or various forms of time-delay neural network, etc., to realize multi-resolution feature extraction. However, in the present application, these algorithms are one of the steps for implementing multi-resolution feature extraction. The multi-resolution feature extraction and the existing artificial intelligence are different in algorithm target, different in output and different in required data volume. Multi-resolution feature extraction does not require a large number of samples because it is a mapping from features to concepts, and typically a concept contains limited features.

Associating a chained activation system: associative activation refers to a chained activation process that proceeds under the "close activation" principle, the "similar activation" principle, and the "strong memoization activation" principle. When information is input, the machine gives an initial activation value according to an incentive, and performs chain activation through a near activation principle, a similar activation principle and a strong memory activation principle. In a relational network, when a certain node (i) is assigned a certain activation value, if this value is greater than its preset activation threshold va (i), then the node (i) will be activated. It will pass the activation value on to other feature map nodes that have a connection relationship with it. The transfer coefficient is determined according to a 'near activation' principle, a 'similar activation' principle and a 'strong memory activation' principle. If a certain node receives the transmitted activation value and accumulates the initial activation value of the node, the total activation value is greater than the preset activation threshold value of the node, the node is activated, and the activation value is transmitted to other characteristic graphs which have connection relations with the node. This activation process is transferred in a chain until no new activation occurs and the entire activation value transfer process stops, which is called the associative activation process.

The associative activation process is a memory-related search method, so it can be replaced by other search or lookup methods that can perform similar functions. In the patent application with application number 202010400739.7 entitled "method for simulating human memory to implement general machine intelligence", a specific method for implementing the associative chain type activation system is proposed, and is not repeated here.

Establishing a preset basic response system: the preset basic response system is a machine response system realized by a preset program. These systems are the instinctive reaction to which the machine responds. These instinctive responses can be progressively optimized in the future learning.

1, basic motion system. The basic actions include giving the machine a program that can mimic human beings making basic actions, including language pronunciation and expression, and the mimicking of limb actions.

And 2, presetting instinct response. The instinct response means that the output response of the machine under specific input is realized through a preset program. For example, the machine can respond to high-temperature avoidance reaction, to fall instinctive avoidance reaction, to sudden impact avoidance reaction, and the like through a preset program. These responses are adjusted empirically in subsequent learning. For example, through a posteriori result, it is learned that the instinct response may bring serious loss under some specific information input condition. Therefore, when the input condition of the instinct response is excited, the instinct response is accompanied by specific information which can cause serious loss of the instinct response, and the machine can suppress the instinct response under the driving of interest and harm, and seek more optimal balance of profit and loss.

Emergency situations requiring a instinctive response are not many and can be implemented using a preset program. However, the environment accompanying instinctive response excitation may be complicated, so the machine needs to establish a relationship network through learning and posterior experience, and make different responses to different situations under the control of a prediction, planning and execution system.

And 3, initially activating the value assignment system. In step S102 of fig. 1, after the machine extracts the multi-resolution features, an initial activation value assignment program is used to assign initial activation values to the input information. The initial activation value assignment routine is a preset routine whose input parameters include input information, the machine's motivation and motivational state at the time, such as demand and demand state, and emotional state. Its output includes an initial activation value assigned to the input information. A simple method is to assign equal activation values directly to all input information based on the motivational state of the machine at the time of its anticipated information, demand and mood. Another method is to simply classify the expected information of the machine and use different initial activation values for different expected information. It is also possible to adopt a method of assigning different initial activation values to different resolution information. For example, a larger activation value is assigned to low resolution information and a lower activation value is assigned to high resolution information. The value may be assigned in reverse, depending on the machine's expectation of the input information. Whereas the machine's expectation of the input information comes from the previous information processing results.

The machine may also adjust the initial activation value assigned to it based on how often the information is entered. Such as those that are frequently given a progressively lower initial activation value. The information with low initial activation value also has low activation value obtained by other information in the initiated chain activation process. When the new memory is stored, the initial memory value obtained by the new memory is positively correlated with the activation value when the memory occurs, so that the original activation value obtained by the new memory is low, the activation value given to other information by the chain activation initiated by the new memory is also low, and the memory value obtained when the new memory is stored is also low. The memory storage is firstly put into a temporary memory bank. Only information that has obtained sufficient memory values in the temporary memory bank is moved to the other memory banks. Therefore, the common things in daily life are difficult to survive in the temporary memory bank to form long-term memory. This is also an evolutionary result of intelligence, as it is the general laws and special exceptions that an agent needs to deal with the external environment. The general rule is that by repeating the summary, the similar situation of long-term memory is not needed after the summary is finished. Those exceptional, or sporadic, strong stimuli may result in a strong chain activation process due to a high initial activation value, such that a single memory value reaches a long-term memory threshold, thereby becoming the long-term memory of the machine.

The initial activation value of the machine can also be given for a plurality of times. Such as simply assigning an initial activation value to the input information using the methods described above. Then, according to the preliminary processing result of the input information, preliminarily determining the connection tightness between the input information and the 'interest' and the 'harm', and then endowing the input information with an initial activation value again according to a preset program according to the connection tightness between the input information and the 'interest' and the 'harm'. For example, for information that may be "good" and "bad," the machine's policy may be to further analyze the information. And further analysis means that more resolutions are adopted again to extract input information, and initial activation values with higher multi-resolution information features are given again to make the whole activation range larger, so that more related memories are searched. The above process may also be iterated to form multiple initial activation value assignments.

The basic action system, the preset instinct response and the initial activation value assignment system can be completed by using a computer program which is mature at present. These specific implementations can be achieved by knowledge of the industry. The present patent application is directed to methods and approaches for implementing general machine intelligence using existing technologies, and details of implementing specific steps that can be implemented by those skilled in the art based on knowledge known in the art need not be further described in the present application.

And (3) environment reconstruction: the context information is part of the machine input information. This information will activate the machine to remember similar context information. The context information in these memories may be the memory in the memory for other parts or other angles of the same context, or the memory of a similar context. Such activated information may include visual information, auditory information, tactile information, motivational and motivational status information, and the like.

The machine needs to employ (1) overlapping similar parts in the environment for building a 3-dimensional environment. (2) Missing information is predicted using as a model features on common structures in similar environments. Through the steps (1) and (2), the environment formed by combining the machines is the current environment constructed by fusing the currently input environment information and the environment information related to memory. This is the machine-to-environment reconstruction. The reconstruction of the environment by the machine is essentially a prediction of the environment by the machine. After the environment space is established, the part which cannot be seen at present in the real space can be known according to other parts of the memory space to be referred. For example, when looking at a familiar cabinet, we can look like seeing the image inside the cabinet. This is true because we have superimposed the memory image inside the cabinet. Meanwhile, in the process of reconstructing the environments, the corresponding auditory recollections, tactile recollections, motivations, motivational states (recollections of various emotional and emotional states) and the like of the machine can be triggered through associative activation, so that the environment reconstructed by the machine is an environment with subjective feelings.

The size of the predicted range of the machine to the environment is related to the anticipated target of the machine. In the case where the target is generally expected to be large, the prediction range of the environment is also large. In the case where the target resolution is expected to be high, the prediction resolution of the environment is also high. And the intended target of the machine is the target that the machine produced during the preceding prediction, decision and execution.

Algorithms for reconstructing the whole body by local parts are many, such as GAN neural networks, such as 3-dimensional reconstruction techniques commonly used in games. The machine can use these existing techniques to combine the input multi-resolution context information with the activated multi-resolution feature information in memory for 3-dimensional reconstruction. These objects are achieved by those skilled in the art based on the knowledge known in the art, which are not in the scope of the claims of the present application and therefore need not be further explained here.

Self-reconstruction: similar to the method of environment reconstruction, machines also reconstruct self-images using the same method. In reconstructing the self-image, the machine needs to fuse the relevant information in the memory activated by the current input information, besides using the current input information: such as vision and hearing, touch, smell, taste and sensation, gravity sensation, sensation of limb state, motivation and motivational state (such as mood and emotional state, demand and demand state, etc.). The machine needs to employ (1) overlapping similar portions of these information for building a 3-dimensional avatar. (2) These information are organized as models, supplementing missing information, using common structural features in the memory of the model, such as a model of a person, or a typical image of the model itself. Through the steps (1) and (2), the information integrated by the machine is the image and the feeling established by the machine on the machine. For example, when we are doing actions with both hands on the back, we seem to see these actions. The reason is that after the nerve instruction is sent and the tactile perception is obtained, the vision connected with the similar nerve instruction in the memory is activated, the vision and the tactile sense connected with the similar body posture perception information are also activated, the vision connected with the similar tactile perception information is also activated, and the information is integrated into the integral image after the similar parts are overlapped. We created a self-image in mind as if we were able to see the action of the self-image.

The machine firstly has a specific 'self' concept through a preset software and hardware system: a limb or a vocal organ which can be driven under the command of the user, a sense organ which can transmit various information to the user, and visual and auditory information which can always integrate various senses of the user. Since these information always occur simultaneously, the relationship between them is very tight in the relationship network, thereby forming the concept of "self. The concept of "self" is formed in the same way as other concepts and is not mysterious. For example, if a person feels pain when a table is knocked, he must consider the table as part of himself.

After having a narrow "self" concept, the machine gradually obtains various relationships between "interest" and "harm" and the "self" involved in learning. These "benefits" and "hazards" are in turn tightly linked to the underlying motivation of the machine (such as demand and mood). Therefore, under the motivation of "driving toward profit and avoiding harm", the behavior pattern of the machine is likely to "possess" things that bring "profit" to the machine, and "avoid" things that bring "harm" to the machine. Therefore, with the concept of "self," there are concepts of "possession" and "avoidance. Because these concepts of "occupancy" and "avoidance" extend under the drive of the principles of revenue maximization, loss minimization. With the concepts of "possession" and "avoidance," the machine can understand the organization, laws, behaviors, and morality of our society. Because the core content of our organisation is the various expressions "occupation" and "avoidance".

The self-conscious nature of a machine is a way of behaving as a machine. The machine does not need mysterious self-consciousness to be added to the machine, and the machine is a behavior mode that the machine determines the interaction between the machine and the outside according to a mode of 'benefit avoidance' after learning various information and cognition of 'interest' and 'harm' relations of the machine through a relation network and associative activation.

The invention discloses a method for realizing strong artificial intelligence in a patent application, and aims to provide the method for realizing the strong artificial intelligence. The specific software and algorithms required for self-reconstruction of the machine will be known to those skilled in the art based on the knowledge in the art to achieve this sub-goal of achieving self-awareness creation as proposed in the present patent application, and therefore need not be further described herein.

Language reconstruction: language plays an important role in machine intelligence. Language is a set of symbols that humans create for better communication experience. Each symbol represents a particular thing, process, and scene. When the machine receives external language information, multi-resolution feature extraction is performed on the input language (such as voice or text) through the S102 module. For example, features of the voice as a whole, such as features of rising and falling tone, accent, tone, speech speed, voice size and change, voice tone frequency and change, and the like, are obtained at a low resolution; obtaining specific words at medium resolution; specific syllable pronunciations are obtained at high resolution.

Through the module S103, these features can simultaneously activate the related memories in the memories through associative activation. Because languages are used very frequently, languages are often tied to other sensors that are typical of the images, dynamic images, sensations, sounds, etc. of the objects, processes, and scenes they represent. These closely related local networks are part of a relational network, and they are concepts.

When a language is entered, the associated memory represented by the language is activated. These memories may have both information about the language itself and memory related to the use of the language activated (e.g., speech emphasis or text emphasis to emphasize an emphasis such as an irreconcilable voice or a jeopardy tone, etc.). The activated information constitutes an activation information stream. To balance the context of the language and the current semantic recognition, the activation value of the activated information may decay over time. The parameters of the decline are related to the motivation and state of the machine (such as demand and demand state, mood and emotional state). Whereas chain activation enables context-dependent recognition of all input information. The input information here includes both the environmental information and the activated memory information. The mutual activation and assignment of the information embody the context association. This correlation is more extensive than statistically generated semantic library content. It does not relate to only speech, but also to all sensory inputs and associated memory. So through the multi-resolution information formed in S102, the common sense network formed in S103, the machine can achieve the connection of language to static and dynamic images, feelings, demands and emotions, as well as the connection of language to related language and memories. When such a connection is incorporated into the understanding of the language input by the machine and is responsive to the understanding of the language, and associated experience, it is apparent that the machine actually understands the true meaning of the input language.

The language input constitutes an input information stream and the corresponding activation memory also constitutes an activation information stream. The machine needs to reconstruct this active information stream when understanding the language. The reconstruction is to reconstruct the environment of the environment information, and to reconstruct the information related to self. The machine views the reconstructed activation information stream from the perspective of the third party. The reconstructed stream is entered as a new message, which the machine may restore to memory as part of the memory. So that the machine learns through the language and is a virtual experience. Is a virtual scene observed from the perspective of a third party. Wherein the reconstructed self-image is a representation of self-awareness and the reconstructed environment is virtual environment information. From these virtual experiences, the machine can learn knowledge, learning the connection relationships between various information and "interest" and "harm". This virtual experience will also activate relevant information in the memory, which will also be present in the memory, as well as other memories, as part of the relationship network, as well as being optimized by the memory and forgetting mechanism.

Language reconstruction is the basis for machine learning and understanding of language, and also the basis for machine learning of the experienced experience accumulated historically in humans. After the learning ability of language, the machine can learn all human knowledge, which is the basic ability leading to super intelligence.

The language is a way of transmitting the experience of others, the experience can be transmitted between machines through the language, and the cognition can be directly shared through a relational network, so the learning process of the machine can be very rapid, and the human must avoid the behavior which is unfavorable for the human and is generated after the machine combines various cognition through a preset rule. Because the machine intelligent technology related to the invention application can be seen by human, the activation process of the machine, the prediction result of the machine, the decision algorithm and the decision result of the machine can be directly read and understood by human. This is not the same as the current deep learning. At present, in deep learning, the number of parameters participating in decision making is too large, and the meaning is not clear, so that the decision making process is difficult to explain. In the invention, the machine is established by sharing local characteristics from perception to cognition through a memory and forgetting mechanism, the number of parameters participating in decision making is far smaller than that of the existing multilayer neural network, and each step of decision making can be understood, so the artificial intelligence provided by the invention is interpretable and controllable.

Prediction, decision and response process of the machine: the prediction, decision and response process of the machine is based on a relational network, a search range is limited through associative activation, the sizes and the occurrence probability of 'interest' and 'harm' are calculated through a statistical method through past experience (including causal probability between events, including 'interest' and 'harm' brought by the events), and then the probability of occurrence of a certain event is increased or decreased through a principle of driving away the harm under a preset algorithm. This is the overall process of prediction, decision making and response of the machine.

When information is entered, the machine does not need to exhaustively predict all possible outcomes, which is also an unfulfilled task. The machine need only evaluate those results that are activated that may occur in the experience associated with the input information. This is equivalent to using common sense to define the search scope.

In past experience, each possible outcome is linked to motivational states, such as demand states and emotional states, through a relationship network, which represent the "interest" and "harm" that such outcomes may cause to the machine. Thus, by combining the probability of occurrence with the associated demand and emotional state, the machine can infer the magnitude, type, and probability of occurrence of "interest" and "harm" that each outcome of the development of the thing may bring to itself. Therefore, the machine needs to synchronously take the probability of possible occurrence into consideration when predicting the 'benefit' and 'harm' brought to the machine by the possible occurrence. Therefore, the machine should evaluate both "interest" and "harm" and also evaluate their corresponding probabilities, and combine both to make a decision.

For example, within a limited range, the machine can adopt any current artificial intelligence prediction method, such as monte carlo search, decision tree, bayesian estimation, machine inference based on rules and the like, to infer the probability of the current event developing to the most probable result of the next step, and the "benefit" and "harm" brought to the machine by the result.

One possible way of evaluating is: a multi-dimensional space is created using various types of demands and emotional states, and various preset spaces such as an optimal pursuit space, an acceptable area, and an unacceptable area are preset for the machine in the space. This is equivalent to establishing various driving risk calculation rules for the machine. However, due to the diversity of requirements and emotion types, such rules are difficult to express explicitly, and all methods for calculating spatial distances can be adopted. The strategy employed by the machine may be to be as close to the best pursuit space as possible in spatial distance, away from the unacceptable space. Such that the rules are quantified in terms of spatial distance.

Another simpler method is to calculate Y = ∑ f_i (x1,x2,…)* p1_i (x1,x2,…)- ∑ G_j (y1,y2,…) *p2_j(y1, y2, …). Where Y is the final evaluation value, f_i(x1, x2, …) are various profit values (determined by demand, emotion type and status according to preset rules), p1_i(x1, x2, …) is the probability of likely occurrence of the corresponding benefit value. G_j(y1, y2, …) are various loss values, p2_j(y1, y2, …) is the probability that the corresponding loss value may occur. Σ is a summation sign, summing i and j, respectively. The aim of the machine is to maximize Y. The method is thatThe demand, emotion type and state are quantified as a profit value and a loss value, and then the probability sum is simply calculated.

The above described evaluation of the possibility of a single step is a technical evaluation. The evaluation can be realized by adopting the existing machine intelligent method and the existing statistical calculation method, such as the method of calculating the probability statistical average, or the minimum space distance, and the like.

Based on the evaluation of the single step possibility, the machine also needs to predict and plan the subsequent 'benefit' and 'harm' brought by the own response and the environmental response. This is a multi-step "benefit" and "harm" prediction and assessment process. This is a strategic plan, which is equivalent to finding the optimal path. The principle of selecting the path is just the trend and avoidance, and the selection method can comprise the existing decision-making algorithms such as a decision tree, a rule-based expert system, a Bayesian network, an evolutionary algorithm and the like. The basis of these algorithms is to build a relationship network with causal probabilities and "interest" and "harm" relationships.

When the machine evaluates multiple steps of "interest" and "harm," a simple evaluation method may be: the planned response of each step is taken as a virtual process to be taken as input information again, all information extraction, association activation, environment, self and language reconstruction processes are taken again, and then the method like single-step evaluation is adopted again to evaluate possible 'benefits' and 'harms'. This is a method of evaluating the possible influence of own response by using own experience. These effects include external feedback on the responses of the users (including feedback on the users from other people in the environment), and the influence of the feedback on the users 'benefits' and 'harms'.

By adopting the method, the machine can use an iterative method to fully evaluate the input information and plan the response according to the principle of tendency and avoidance of harm. The planning can be iterated for many times, and the iteration for many times is essentially a process for expanding a search range and searching for an optimal path. This is the planning nature of the machine.

The optimal path searching range of the machine during planning is determined by possible 'benefit' and 'harm'. Generally, if possible values of "interest" and "harm" are high, the machine expands the search range under a preset rule of a driving and avoiding algorithm, for example, the activation threshold is reduced or the number of iterations of evaluation is increased, so that a larger range of information is activated, a larger range of experience is included in a path prediction process, and an optimal path is selected from the larger range of experience.

Therefore, the decision process of the machine is not mysterious, is relatively simple to implement with the help of common knowledge, and can adopt a preset program and the existing machine reasoning method to calculate the probability of main possible results and the probability of various gains and losses brought by the main possible results. The methods used by machines to evaluate "interest" and "harm" involve only probabilistic and statistical methods, which are well known. However, the method of machine evaluation of "interest" and "harm" is not limited to the two methods described above, and the machine may utilize any existing statistical decision algorithm.

It should be noted that the decision process of the machine is based on the relationship network. And the relationship network is closely related to the education suffered by the machine. Even the same learning material and different learning orders can generate different relationship networks, thereby forming different cognition and decision processes of the machine. The decision-making capabilities of the machine are also influenced by the machine's associative activation parameters, statistical algorithm parameters, underlying motivation and motivational states (including demand and demand states, emotional and emotional states, etc.). The decisions made by different experience, different set up machines, may be different.

The machine can also adjust the algorithm parameters for making decisions according to the feedback of own experience. For example, after the machine executes a series of responses according to its own decision, the actual result and the expected result have a large difference, and then these memories also enter the memory and become part of the relationship network. The next time the machine makes a decision, the memories become new experiences which affect the decision behavior of the machine in the next similar state.

The machine may also learn to summarize the experience under a preset algorithm. The specific implementation mode is as follows: the preset algorithm can select the process of obtaining larger 'profit' and 'harm' by the machine. The memory of these processes is entered virtually a number of times in the machine's processor, so that each time the associated memory is activated. Those common features that exist in such processes occur and activate the associated memory from time to time, enhancing their memory value. These enhanced memories are therefore more easily associatively activated and more easily referenced for use. Those related memories that are present only in a single specific process are gradually forgotten due to long-term difficult activation. This is the process by which the machine actively summarizes experience. Of course, without such a pre-set program, the machine itself would also summarize these experiences in the same way, but the samples and time required would be longer.

Since the machine's decision system will search for a multi-step optimal path, the machine may come up with a decision that accepts a small penalty but expects a large gain in the future.

For example, the trainer warns the robot that it will be "heavily penalized". And the machine can understand that the player can bring huge loss by further activating the relevant information through the explanation of the word of the heavy penalty. The cognition of 'hitting people' may cause loss, the machine can obtain direct experience through direct hitting people and feedback, related experience preset for the machine can be obtained, and mixed experience of various modes can be obtained. However, when the machine faces the situation that 'the owner of the machine is attacked and is very dangerous', the machine can cause 'loss value' when people hit, and the life of the owner of the machine can cause a balance between huge 'income values', and at the moment, the machine needs to carry out multi-step search to enlarge the search range so as to determine the probability of the huge 'income values', and other losses can not be caused. The decision is then made synthetically based on the probability of return and the probability of loss.

For example, if the owner's life is threatened based on experience (or pre-set experience), the "profit value" for saving the owner may be high. But if the attacker is the "police", then he cannot insert his hands, or else he suffers a great "loss" or the like with a probability of hundreds. This knowledge may be a postnatal announcement or a direct innate pre-set experience.

The machine execution system is similar to the machine decision system, with the difference that the machine execution system embodies the goals produced by the machine decision system; the embodied process is also a decision of the machine by subdividing specific objects layer by layer to the underlying drive command level of the machine until such time as the machine can directly execute the instructions. When the machine is subdivided layer by layer, a decision system of the machine is also adopted.

If the machine selects the voice output in the steps of selecting various possible responses, the machine is simple, and only the image feature diagram to be output is converted into voice, and then the dynamic feature diagram (including the concept representing the relationship) and the static concept are combined (the grammar knowledge existing in the relationship network) by utilizing the relationship network and the memory and adopting the concept replacement method to organize the language output sequence and call the pronunciation experience for implementation. It should be noted that machines may choose from experience (either self or others) to express the dynamics of an entire sentence (e.g., using different movement patterns of tone, or variation in stress to express questions, jeers, distrust, emphasis, etc., which are commonly used in humans. Because the machine learns these expressions from human life, any expression of a human can be learned by the machine theoretically.

The machine needs to divide the targets of the sequence (including intermediate targets and final targets) according to the targets, which relate to different time and space, so as to coordinate the execution efficiency of the machine. This is also from experience and can be performed using a preset algorithm. The approach taken is by selecting as a group closely related targets in time and closely related targets in space. Because the dynamic characteristic diagram and the static characteristic diagram are combined to form an information combination, and the environment space of the related memory is provided with time and space information, the classification method can be adopted in the step. This step corresponds to a change from the general scenario to the sub scenario. It should also be noted that this total scenario may also be only one of the phase goals of completing the machine planning, such as increasing the probability that a certain condition may result in a good profit margin.

Therefore, the machine needs to be divided according to time and space, and links needing to be executed in the current time and space are taken as the current targets. And temporarily putting other time and space targets to one side as inheritance targets. After the machine targets the middle link, the machine still needs to further subdivide the time and space (write the lower level script again). This is a process of increasing temporal and spatial resolution. The process of converting one target into a plurality of intermediate link targets by the machine is still a process of analyzing various possible results and possible occurrence probabilities by using decision-making capability and selecting own response according to the principle of 'benefiting and avoiding'. The above process is iterative, and the process of dividing each target into a plurality of intermediate targets is a completely similar processing flow. Until the underlying experience of the machine is resolved. The underlying experience is that for language it is the muscles that are mobilized to make syllables. For an action, it is decomposed into the issuing of drive commands to the relevant "muscles". This is a tower-like decomposition structure. The machine starts from the top level target and decomposes one target into a plurality of intermediate link targets. This process is to create virtual intermediate process targets that are retained if they are "on demand". If "not compliant," it is recreated. Whether the policy meets the requirement or not means that after the machine analyzes the income and loss conditions, whether the policy meets the preset acceptable standard or not is confirmed.

The above processes are iteratively expanded layer by layer, and finally colorful responses of the machine can be established.

The prediction, decision and response processes of the above machine do not need new algorithms. The method is an existing algorithm, and can realize the prediction, decision and response processes of a machine through reasonable organization on the basis of a relational network. The invention focuses on disclosing how to establish methods, processes and steps for realizing the intelligence (strong artificial intelligence) of a general machine by organizing the existing algorithms and data. The conventional statistical methods and artificial intelligence algorithms are not in the scope of the present invention, and are not described herein.

Claims

1. A method for establishing a connection relationship between information, comprising:

it is considered that information adjacent in input time has a connection relationship with each other.

2. The method for establishing the connection relationship between information according to claim 1, comprising:

this connection is optimized by a memory and forgetting mechanism.

3. The method for establishing the connection relationship between information according to claim 1, comprising:

the information is represented by multi-resolution features, including features representing the feature combination mode; the connection relation is based on the multi-resolution characteristics of the established information, and the connection relation between the information can be different in different resolutions.

4. The method for establishing the connection relationship between information according to claim 1, comprising:

the input information includes external input information, internal monitoring information, motivation of the machine and motivation activation status information.

5. The method for establishing the connection relationship between information according to claim 1, comprising:

the machine adopts an information storage method, and is characterized in that the machine can express that the connection relation exists between the input time adjacent information in an information storage mode.

6. A method of machine prediction, decision making and execution, comprising:

after the machine inputs external or internal information, the machine activates the related information by adopting a method of association activation in the relationship network; the machine predicts the influence of similar events on own motivation by counting the probability of each activated motivation and the corresponding memory value; the machine selects and pushes the paths of the event development according to the influence of the event obtained by statistics on the motivation of the machine and the manner of trending toward profit and avoiding harm, and the sequence targets on the paths are the behavior decision of the machine.

7. The method of claim 6, comprising:

the machine takes the response of the planned output as the virtual information input of the machine, and continues to adopt the modes of associative activation, information reconstruction and prediction, decision and execution to evaluate the feedback which can be obtained by the response of the machine.

8. The method of claim 6, comprising:

the machine takes the expected external feedback as the virtual information input of the machine, and continues to adopt the modes of associative activation, information reconstruction and prediction, decision and execution to evaluate the 'benefit' and 'harm' possibly brought to the machine by the external feedback to the machine.

9. The method of claim 6, comprising:

the machine may expand the search range by lowering the activation threshold or increasing the number of iterations of prediction and evaluation, finding the optimal response path from a larger search range.

10. The method of claim 6, comprising:

the method for the machine to push the path of the event development is a dynamic process, and in the execution process, the machine continuously incorporates new information into the prediction, decision and execution processes according to the same prediction method under the condition of the new information, and updates the behavior decision.

11. The method of claim 6, comprising:

in the process of executing the sequence target, the machine decomposes a single target in the sequence target into more bottom layer targets according to the same machine prediction, decision and execution algorithm, and decomposes the sequence target layer by layer in the execution until a bottom layer driving command which can be directly executed by the machine is decomposed; the behavior of the machine under the series of underlying drive commands constitutes the response of the machine to the input information.

12. The method of claim 6, comprising:

when the machine predicts, not only the motivation state influence which the behavior of the machine may bring to the machine is predicted, but also the possible behaviors which are not self and the motivation state influence which the behaviors may bring to the machine are predicted; the machine in predicting non-self likely behavior is based on the assumptions of (1) past memory about non-self behavior in a similar state and (2) the assumption that non-self is also making decisions in a manner that leads to a profit-and-harm.

13. An apparatus that can learn, decide and execute autonomously, comprising:

the sensor system comprises a general sensor group simulating human senses, a sensor group used for a specific purpose of the machine, and a sensor group used for monitoring the self running state of the machine;

the system comprises a preset motivation system, a system and a method, wherein symbols are used for representing a type of motivation, the state of the motivation is represented in a certain mode, and the starting point of the machine in decision making is to drive some motivations to be in certain states;

a relational network system including multiresolution reduction of input information, and recognizing that a relationship exists between adjacent input information, and expressing the relationship in a memory in a specific manner; the relation between the multi-resolution information is optimized by adopting a memory and forgetting mechanism; the relationship network contains the information itself, and also contains the motivation and motivational state motivated by the information;

the association activation system comprises systems for giving initial values to input information once or for multiple times, the systems are realized by adopting a preset program mode, and parameters in the preset programs are adjusted according to learning results and in a mode of benefiting tendency and avoiding harm; the method also comprises a joint activation mode according to a 'near activation' principle, a 'similar activation' principle and a 'strong memory activation' principle; when the association activation system stores the input information, the memory value given to the input information is a positive correlation statistical function of the initial activation value given for one time or multiple times, and is also a positive correlation function of the activation value of the motivation state activated by the information in the memory;

the machine decision system comprises a machine decision system, a memory module and a memory module, wherein the machine decision system uses input information and activates related memory in a way of associative activation; the machine limits the search range of memory and the evaluation range of decision by the machine through a associative activation mode, and counts the size and probability of 'benefit' and 'harm' possibly brought to the machine according to the probability of possible occurrence of motivation and motivation states; the method comprises the steps of distinguishing self from non-self, predicting possible responses of the self and the non-self based on self experience and principles of trending interest and avoiding harm, and further iteratively evaluating the magnitude and probability of 'interest' and 'harm' possibly brought to the self by the responses; the machine determines the possible development direction of the event and the influence of the development direction of the event on the machine through the iterative evaluation; the machine adopts response to improve the probability of making the development direction of the event tend to the self-favorable direction development according to the mode of benefiting tendency and avoiding harm, so as to reduce the probability of making the development direction of the event tend to the self-harmful direction development; the machine needs to update the internal and external information of the machine at any time in response execution, and the same decision method is adopted to correct the response of the machine, so that the aim is to improve the probability that the development direction of the event tends to the direction beneficial to the machine and reduce the probability that the development direction of the event tends to the direction harmful to the machine;

the machine execution system is similar to the machine decision system, with the difference that the machine execution system embodies the goals produced by the machine decision system; the embodied process is also a decision of the machine, and the decision is to subdivide a specific target into the bottom driving command level of the machine layer by layer until the machine can directly execute the instruction; when the machine is subdivided layer by layer, a decision system of the machine is also adopted;

the preset response system of the machine comprises all or part of the following functional systems: the system comprises a correlation system between the emotion and the motivation state of the machine, a correlation system between the emotion and the emotion expression, a preset pronunciation and action system of the machine, a preset conditioned reflex system and the like; these systems are mainly composed of pre-set programs and parameters that can be adjusted by learning the experience obtained.

14. A method of suggesting generic artificial intelligence, comprising:

the machine maps the input data to the local common features; the machine maps the local common features to concepts; the machine establishes the relationship between the motivational demand of the machine and the external information through the relationship network, and uses the relationship to make prediction, decision and response.