JP2009522684A - A method for incorporating psychological temperament into the electronic emulation of the human brain - Google Patents

Abstract

  The present invention provides a method for emulating a human brain having that thought and rationalization process, and a method for storing human-like thoughts. The present invention makes it possible to incorporate psychological profiles, experiences and social status into the electronic emulation of the human brain. According to the present invention, it becomes possible to react like an actual human by the emulation of the human and the surrounding environment that influence each other.

Description

[Technical Field of the Invention]
The present invention relates generally to artificial intelligence, and more particularly to modeling temperament and personality.

[Cross-reference of related applications]
Pending US Patent Application No. 11 / 030,452 (Attorney Docket No. VISL-) entitled “METHOD FOR INCLUSION OF PSYCHOLOGICAL TEMPERAMENT IN AN ELECTRONIC EMULATION OF THE HUMAN BRAIN” filed on January 6, 2005. Claiming the benefits of provisional patent applications 60 / 534,641, 60 / 534,492 and 60 / 534,659 filed on January 6, 2004 There is a patent application. This application is a continuation application.

[Advantages of the Invention]
The present invention provides a clear and simple way to model the process of human thought, emotion, behavior, recognition and guessing, and the cold, impersonal of a traditional model for artificial intelligence (AI). Provide an alternative to behavior. The method of the present invention is compatible with many conventional approaches to AI, but is based on a model of temperament and personality. Overall, these methods are systems that aggregate processing and memory to accurately emulate human thoughts, decision making and behavior.

[Background of the invention]
During the past decades, various methods for emulating human-like behavior have been proposed and constructed, many of which mimic living organisms. That is, these methods have been proposed based on human biological elements that underlying the human brain. Although these methods have been successful to some extent, they have not reached the point where a precise emulation of the brain can be achieved on a large scale. Some biology-related concepts, such as fuzzy logic, were relatively few in terms of entering the commercial market. While fuzzy logic has been extremely successful in certain market areas, fuzzy logic and rule-based applications have been niche and limited. However, none has proven to be suitable for implementation on a brain level scale.

  Throughout this specification, “brain emulation” and “brain model” are used interchangeably as necessary in order to express their intentions most clearly.

  Some approaches of conventional systems are based on neural networks that make use of biological concepts as shown in FIG.

  For example, assume in a neural network that a set of neurons 1 and 2 is stimulated by some external means 95. Each neuron typically represents a fact. Neurons fire according to their current state of fact recognition. These neurons can also be connected to other neurons 3 and 4, which imply a particular relationship between the neurons. A set of weights 5 may be inserted into the connection between neurons, and these weights control the influence of the “input” neurons 1 and 2 on the “output” neurons 3.

  Finally, as in the case of the neuron 4, various forms of control such as suppression means can be implemented. In this case, firing in the neuron 2 can suppress the influence of the upper neuron 13 on the output neuron 4 as indicated by the suppression means 6. This implementation may be in the neuron 4 or earlier. The organization and interconnection of the network indicates that certain current input conditions 95 produce the desired output result 96.

  Another common form of conventional system involves using a system of rules, as shown in FIG. In this case, a set of rules 7 acts on a set of input conditions 95 to generate a set of predetermined desired output results 96. The rule translates a set of input conditions into a set of signals representing the desired result. In addition, the set of feedback paths 97 is compared with the input 95 to change rule 7 so that the output 96 can converge to the desired result. These are usually referred to as “first principles” systems.

  Although rule-based systems tend to produce the desired results, they can be quite complex and require enormous computational power. Hundreds of thousands of rules may be generated, but the results are not always accurate. Typical applications such as text-to-speech synthesis using a rule-based mechanism will not improve accuracy unless extensive operator-specific training is used. The inaccuracy of the results and the overwhelming computational power have hindered the widespread application of rule-based systems.

[Summary of Invention]
The present invention incorporates temperament, emotions, emotions, personality and previous experience to accurately emulate human thought and decision making processes. Without these, it is unlikely that any system or neural network can accurately emulate a real person.

  In order to be able to understand and realize the function of the brain or to simplify it, certain assumptions are made by the model underlying the human brain, including the model of the present invention. Such a model can suggest a specific approach to realization. However, the accuracy of the model relative to the natural biological elements of the brain, or the accuracy of the model relative to the correct psychological behavior, does not necessarily affect the scope of the claims of the present invention.

  In considering human brain emulation, one can imagine a particular structure or element that may not have a biological counterpart (or may only have a temporary counterpart). Such a structure or element greatly simplifies emulation. One such set of elements are brain parameters, or parameters, as used throughout this patent. Parameters are control elements that define the state of the current emotional and mental processes of the brain. Such parameters are used herein to define states, conditions, moods and other important states, similar to I / O ports in electronic hardware, or state variables in software. Similar to.

  Unlike other approaches to human thought emulation, the present invention makes it possible to easily emulate the influence of cultural, religious and political views and beliefs that are inherent in personality. .

  The present invention is biomimetic in that the concept of learning and interpreting new information occurs in the current context. A number of special dedicated memories and lists are used to mimic this and other functions of the memory process in the human brain.

  The context pool memory holds the current contextual context and is central to the analysis and thought processes of the present invention. It is the current context pool. An enhanced memory holds relationships based on propositions, “facts” and experiences that have not been reconfirmed. (Normally) Information reconfirmed over 21 days is delivered to permanent memory, ie long-term memory. Finally, a dedicated list memory is maintained for use in motor skills, particularly time-dependent or repetitive tasks.

  The remainder of the invention consists of various work blocks that implement several brain functions. An analyzer / correlator is a deference system that realizes basic thinking processes, processes whimsical changes in emotions, and places pseudo-people within the context of social, political and religious structures. Is realized. Using event queues and memory simplifies analysis and control functions.

  The main language of the embodiment is English, but using the same method and internal structure, most arbitrary human languages can be selected and used. In order to “bootstrap” the learning process, the present invention also includes a grammatical and lexical parser for English sentences and English fragments.

  The present invention can be trained using prose and establish factors that affect vocabulary, specific sets of experience, knowledge for specific disciplines, and both temperament and personality.

  Multiple instances of the present invention can be used to create a collection of emulated people that communicate, for example, through an electronic or optical network. It is also suitable for controlling external mechanical skeletal structures or other robotic structures. Finally, it is suitable for use as one of a set of trained electronic “humans”, for example comprising a trained crew in an unmanned aerial vehicle (UAV).

  For a more complete understanding of the present invention and its advantages, reference is now made to the following detailed description taken together with the accompanying drawings.

Detailed Description of the Invention
The system of the present disclosure extracts a human temperament, personality and current state into a series of brain parameters. Each of these brain parameters has a value that varies from 0 to 100% and is roughly equivalent to a single neuron. Collectively, these parameters define the state of human presence and specify matters related to temperament and personality. Some parameters are constant and rarely change, while others change dynamically with current conditions.

  The relationship between parameters is pre-established if present. Parameters in a way that alters the decision-making process of the underlying model, decision thresholds and implied personal interests in which those parameters become a part of the brain model Associated.

  The exact list of parameters and their definitions is not closely related to the system of the present disclosure and can include more or fewer parameters depending on any given implementation thereof. A number of parameters define trends that are specific to a particular temperament. Some parameters define the current emotional state, such as confidence in decision making. Other parameters are placeholders that define things like who the current topic of the conversation or who is in the conversation first person, second person or third person. Still other parameters define physical parameters such as position in the environment, sense of direction, timing, etc.

  Some parameters are non-hierarchical, but some brain parameters can be arranged roughly hierarchically, in which case changing one of the parameters will cause other parameters below in the hierarchy to be Become affected. This arrangement simplifies the realization of personality.

[Parameter example]
Table 1 shows by way of example some of the hundreds of such parameters. For example, the “Choleric” parameter 202 is “higher” than other parameters in the hierarchy in that changing the percentage of bile temperament affects the value of many other parameters. "is there. For example, it affects the Propensity to Decide 222. Each parameter can be treated as one neuron that can be interconnected with other (non-parameter) neurons. Parameter neurons can play a role similar to I / O ports in digital computers.

  The following table is not a complete set of parameters, but a representative set of parameters that are useful for the description described below.

  In conventional models of the human brain, facts are simplified and represented as single neurons, each of which is 0. . . Can “ignite” at a certain level of 100%. The degree of ignition is taken to indicate the current perception of the fact. These neurons are interconnected by weighted connections based on the relationship and experience between the connected neurons.

[Example of state parameters related to decision making]
Some of the important state parameters used in the decision making process are detailed below. Some are set by personality characteristics, some are set by the current context, and these are described elsewhere. Some have a reference value established by their propensity to parameter.

  The activity threshold 237 is the minimum percentage of the full scale value that a neuron must fire before being considered a candidate for accommodation in short-term memory.

  The reference decision threshold 250 is a personality based start criterion for the decision threshold. Long-term training and learning experience can increase or decrease its reference value.

  Correlation fact 235 is true if the correlator portion of the analyzer is currently correlating facts. This typically helps the analyzer make decisions.

  The most active node 236 points to the most actively firing neuron in the context pool (short-term memory). The analyzer uses it to scale the decision threshold.

  The action importance level 215 is a relative importance level for decision making. It can be initially scaled by the analyzer as a result of recent decisions based on the propensity of behavioral importance.

  The integrity request 260 indicates the relative need for complete (and good) facts before making a decision. If the facts are incomplete, the estimator will make a reasonable guess, but the quality of the resulting “facts” will be low.

  The action urgency level 216 represents the urgency level (not the importance level) for decision making. As urgency increases, lower quality information (and decision making) becomes acceptable.

[Tendency parameters based on temperament]
A typical set of basic brain parameters that indicate different tendencies based on temperament are given in Table 2, including representative contributions (given as percentages). This set of values is by no means complete and is provided to illustrate the mechanism of the present invention. Other temperament parameters may be identified and incorporated into this list without changing the method and claims of this patent.

  The specific percentages given in Table 2 are typical and typical values used, but are fine-tuned to improve the accuracy of the psychological model. In actual implementations, other values may be used. Moreover, this list is representative and not exhaustive and is used to illustrate the system of the present disclosure.

  In general, many of these parameters have been observed to reflect characteristics shared primarily by two of the temperament, and to a greater extent reflect one of the two characteristics (and the table). 2). The same parameters can be shared to the minimum by the remaining two temperaments.

  The system of the present disclosure assumes the use of an underlying node that defines the desired temperament and additional nodes that define the desired percentage of the four temperaments. Table 2 is a table of typical trends selected for each temperament, with each number giving the approximate likelihood of a given property that will be represented by four temperaments as a percentage.

  The given percentages are examples, but those percentages can approximate realistic values. Changing these values will never change the means and methods of this patent and these values can be adjusted to better approximate the temperament characteristics. The list is by no means complete and is given as a set of representative parameters for illustration.

  In many cases, but not necessarily, the overall impact of one temperament is given by the percentage of temperaments as preselected to produce the desired personality and the propensity for each propensity from Table 2. Is given by the product of the percentage. This is illustrated in FIGS. These can be compensated by adding changes due to the gender 201 parameter to account for differences in response between men or women.

[Details of propensity parameters based on several dispositions]
Sampling of the parameters in Table 2 is described below as an example of how such parameters are specified and applied. The described settings and applications of these parameters are inevitably subjective, and the relative weights of these parameters and all other parameters described herein are approximate and illustrative It is. Those skilled in the art will appreciate that these weightings can be changed or adjusted without changing the means of the system of the present disclosure.

  Propensity for Amusement 210 tends to feel funny. As this value increases, the threshold for feeling funny is lowered, and the feeling of feeling funny is triggered earlier. The induction of feelings of interest can be reflected in appropriate facial expressions, such as those given in the underlying brain model and skeletal mechanism (if any).

  The propensity for perfection 211 is an indicator of personality tendency that requires complete facts before making decisions and is based solely on temperament choices. Of course, depression is the highest, and bloody or bile is the lowest. Although it is not normally changed, the underlying brain model (analyzer) can increase or decrease this parameter based on training or learning.

  The propensity 212 for determination is a tendency to request that the brain emulation determine, and sets a reference value for a sense of determination. Over time, the propensity can be changed permanently by achieving (or failing to achieve) an objective or goal.

  The propensity for enjoying 213 is an index of the propensity to feel satisfaction in life problems. Of course, it becomes a little higher in the case of bloody blood, and by achieving a goal, fulfilling a plan, or experiencing a positive relationship (in either direction) a very long time constant (20 Day).

  The propensity for fun 214 defines the propensity for the temperament to make a decision based on the sense of feel-good. It depends on the temperament, tends to be the highest in the case of bloody blood, and has a great influence on the sensitivity to rhythm.

  The propensity to act 215, the propensity 215, feels that it is important to act, whether or not all the facts needed for decision making are available with high confidence It is an indicator of temperamental trends. Of course, it is highest in the case of bloody blood and lowest in the case of depression and mucus. Although it is not normally changed, the underlying brain emulation can increase or decrease this parameter based on training or learning.

  The propensity 216 to place importance on the urgency of acting is an indicator of a personal tendency that feels important to act at the expense of careful consideration or analysis of the facts. Naturally, it is highest in the case of bloody blood and lowest in the case of mucus. This is not normally changed, but the underlying brain emulation can increase or decrease this parameter based on training or learning.

  The patient propensity 217 is an indicator of the overall tendency of patient endurance. Its level is usually high in mucus and low in bloody, but is also greatly influenced by (long term) past history. The growth of this characteristic parameter is very slow and iterative. While repeatedly failing to achieve the short-term or long-term goal, the degree of endurance 217 increases, and it becomes possible to prevent the action from being aborted early.

  The propensity 28 that is easily affected by rhythm is a parameter that depends on the temperament and may increase or decrease due to excessive activity. It controls the relative influence of rhythm on the decision making process. The reference value is relatively high in the case of multiple blood samples.

  The propensity to obtain stability 219 is a parameter depending on the temperament that defines the propensity to move to a stable state. When the value is high, there will be a tendency to make decisions that result in no net change in the sense that the response is slow. It also implies a tendency to procrastinate and is the strength (or weakness) of the mucous character. As the degree of propensity for stability 219 increases, loyalty to the authority determined by the context increases.

  The propensity to analyze 220 (determined by the temperament is not affected by other characteristics except for external orders. Even so, the effect is short-term and is quickly returned to the basic trend. When it is high, the tendency to analyze and correlate facts before making a decision is high, and the decision threshold based on the confidence level generated based on the result is generally high.

  The propensity to take care 221 is a parameter depending on the temperament, and has the highest tendency in mucus and bloody. That propensity is more interesting in capturing human-related facts for entry into short-term memory. The effect of this parameter is established, for example, by changing the clutter filter parameters for the context pool or short-term memory.

  Propensity to Decide 222 is the highest parameter in the case of bile and bloody and affects the willingness to make decisions (increases motivation). In the case of bile quality, if it is later found that the decision is not good, the decision may be changed, but in the case of bloody blood, the result tends to be ignored. The parameter 222 also increases the tendency to reverse decision making when higher quality facts are available, and reduces the stability of decision making and the tendency to respond slowly.

  The propensity to execute the plan 223 defines the (current) level of propensity to execute the plan. Its central value is derived from personality traits, but is changed by variables such as stress, urgency and external pressure. When the pressure is high, like the trauma parameter 230, there is a high tendency to ignore the plan and go back to the reaction based on the personality profile. This is accomplished, for example, as illustrated in FIG.

  The propensity to plan 224 is an indicator of trends and aspirations for planning prior to a project or operation and is a function of the temperament profile. If the propensity 34 is high, it will be postponed until the work step is drafted. The propensity to plan does not imply the propensity to perform the plan according to 223.

  The tendency to stretch 225 is an indicator of the tendency to stretch steadily, which delays decision making and action. The main values are derived from the temperament according to Table 2, in which case they are fixed parameters, but can also be changed gradually by experience or training. Stretching is generally a characteristic of mucous, but it also occurs in the depressive decision-making process when there is no complete fact, and is usually very low in bile quality.

  The propensity 226 to criticize once the results are known is an indicator of the tendency to reevaluate decisions, even if they are quality decisions, and possibly re-evaluate those decisions. According to the temperament dependence as shown in Table 2, it is highest in depression and typically lowest in bile.

  The propensity 227 for seeking behavioral stability is an indicator of the propensity to maintain the current situation. It is mainly a characteristic of mucus, and its propensity is realized by affecting the tendency to respond slowly (increasing that tendency) and less willingness to change the plan. It is associated with the underlying brain emulation or model as part of the clutter filter or interest filter at the context pool, short-term memory or analyzer input, and a new plan or proposal that interrupts the existing or ongoing plan Suppress.

  The propensity to rest hands on hips 228 is generally a depressing characteristic, and as its positive value increases, the relevant skeletal mechanism places the hands mainly on the waist, pockets The tendency of putting a hand in the inside becomes high. This parameter gives control values to the underlying brain emulation or model, which plays a role for motor skills to fulfill this trend. In practice, the emulation or model determines whether this trend is fulfilled.

  Again, the parameters in Table 2 are directly controlled by one or more of the four underlying temperament selection parameters. Those parameters are also scaled by percentages as given by the example in Table 2. The parameters are then assigned by the brain model to the underlying brain emulation or appropriate control points, filters and selectors in the model.

[Introduction of parameter effects]
Throughout the brain emulation, there are a number of places where certain parameters may or may not influence the outcome of decision making. In some cases, the likelihood that a parameter contributes to decision making is often based on statistics. One way to accomplish this is shown in FIG. A random number between 0 and 100% is generated by 421 and compared to the corresponding parameter by 422. If the parameter value exceeds the sum of the reference threshold parameter 423 and the random number, the parameter is introduced.

  This type of logic is frequently used in clutter filters that are described elsewhere.

[Derived brain parameters]
A number of parameters are derived from the basic temperament parameters in Table 2. These values can be a combination of temperament parameters, but are adjusted by learning, training, experience and current conditions. As with the other brain nodes and parameters, most of these parameters are 0. 0 in units suitable for the technique of the embodiment. . . It is expressed within the range of 100%.

  A typical set of these derived parameters is given in Table 3. Each of these parameters has an additional (signed) value to be added to the parameter, which is further adjusted based on learning or training. The list is by no means complete, but is provided to illustrate the mechanism of the present invention. Many of these parameters are related to emotions, their indicators and expression issues. As with all parameters, these parameters can be monitored externally to measure the emotional state of the emulated brain.

  These parameters can be derived from, for example, temperament, context, environment and current state parameters, but other means will become apparent during this description. The parameters in Table 3 are exemplary. Most parameters in this table decay over time to the values shown on the right. This attenuation target is nominal and can be changed through preemptive training. They are derived from the temperament percentage in the same way as in Table 2. The list is by no means exhaustive or complete, and others will become apparent during this description.

  Currently derived parameter values are assigned to appropriate control points, filters and selectors in the brain emulation or model. In some cases, those parameter values control the decision making or stability threshold, or statistical settings for the current interest filter in the emulated brain, such as by 42 in FIG. And establish statistical settings for other such brain emulation functions. The combined effects of these temperaments and parameters derived by temperament determine the combined personality of the emulated brain.

  The basic decision threshold parameter 250 is a starting criterion for many decisions. It is a typical starting decision threshold and a measure of confidence or completeness of information that must be obtained before a decision can be made. The threshold is a percentage of 0. . . Given as 100%, its application depends on the type of decision making. In some cases, it can be used as an absolute threshold or specify the confidence of the current fact, i.e. a numerical value that must be exceeded before a decision can be made.

  The concentration parameter 251 is an index of the ability to concentrate. Increasing the positive value raises the threshold of attention to those distracting from the outside, i.e. not related to problems in the underlying brain model or short-term (or current context) memory in the emulation. It is used by both analyzer 30 and clutter filter 40.

  Obedience 252 is an indicator of the overall propensity to seek stability while under emotional pressure from the outside. It includes a long term filter that decays back to the reference value. Positive compliance 252 greatly increases the threshold of attention to emotional trigger events. The compliance 252 can be changed over a reasonable time, but tends to return to a fixed value defined by its temperament. When this value falls to a value lower than its average set value, the learned response is ignored and there is a greater tendency to return to a response based on the personality profile.

  Overactivity 253 is a measure of the current level of overactivity, as standardly defined by those skilled in the art. It is established by a programmable value and then supplemented by a percentage of temperament. Overactivity is also affected by compliance 252 and current emotional stress. These sources are the main determinants for the overactivity base value, but that value can be changed by long-term training or experience. Biliary and bloody temperament have relatively high values, while depression and mucus are fairly low.

  The effect of overactivity 253 is achieved, for example, by introducing random (typically negative) variations in the magnitude of the selected decision threshold. It also changes the work stage ability and the time constant of the current rhythm parameters, and ultimately affects the performance of the motor work.

  Filter organization details 255 specify filtering of details regarding organization from input information, context pool or short-term memory for brain emulation. Values less than 100% remove the maximum percentage of detail.

  Filter human interest 256 specifies filtering of human interest data from emulated brain input information, context pools or short-term memory. 100% remove most of the human interest information. The value will be highest in the bile quality model and lowest in the bloody case.

  Filter relationship details 258 specify filtering of details about the correlation between facts from the input information, context pool or short-term memory. 100% removes most of the details. Its value is highest in the mucus and bloody models and lowest in the depressive model. As the level increases, the correlation between facts that are related but have a weak relationship is suppressed. When the level is low, the analyzer 30 is prompted to generate an event for the event memory 14. This has the effect of repeatedly returning to the same information and analyzing the short-term memory to better correlate the data.

  Filter technical details 259 specify filtering of technical details from the input information, context pool or short-term memory for brain emulation. 100% removes most of the details. This value is highest for the bile and bloody models and lowest for the depressive model.

  The integrity request parameter 260 establishes the level of information integrity required before making a decision. Higher integrity values are more likely to postpone decision making until all facts are available and may frustrate or avoid decision making. Other parameters related to importance and urgency can change this parameter. The integrity requirements can be changed by the decision of the analyzer 30 and based on external commands to the brain emulation, such as through 93.

  Since the context pool (short-term memory) becomes smaller with time due to rest, the request 260 is returned in time to the value set by the propensity for completeness. The request returns to the propensity value after the decision is made. 100% implies the highest integrity requirement. It is highest for depressive substances and lowest for biliary and polycytic models.

  Patience With Detail 261 is the current level of patience. The reference value is derived from a patient propensity. It is affected by the current state and can be instructed to rise. It greatly changes the decision threshold, and a value of approximately 100% implies that it is comfortable for details. Its value is dynamic and tends to be highest in the case of depressive substance and lowest in the case of bloody and bile quality.

  The level 262 that is steadily stretched is an indicator of the current level of languishing. The reference value is set according to the tendency to stretch steadily, increases with uncertainty, and decreases with agitation. Suddenly delaying decisions that are not restrained in other situations and delaying actions that are not restrained in other situations. Decision making selection is performed in the same manner as 42 in FIG. Higher values at this level postpone decision making, even when certain facts exist (high confidence).

  Stretching is generally a characteristic of mucus, but it also occurs in the depressive decision-making process if there is no complete fact. It is usually very low in the case of bile quality.

  As mentioned earlier, the parameters described in the table above do not constitute a complete set of obvious parameters, and there are hundreds of parameters in all. Selected parameters are provided to illustrate the internal processes and considerations for brain emulation of the present invention.

[Embodiment of brain emulation]
One embodiment of a functional model that underlies the brain is illustrated in FIG. The three main components of the model are an analyzer / correlator 30, a context pool memory 10 and an English semantic analyzer 50.

  Throughout this description, English is always used when dealing with external communication, whether in complete sentences or sentence fragments. Internally, the system is essentially language independent, except where linguistics, phonetics, word spelling or character shapes used in the language are involved. Although English was used to facilitate early-stage implementation, essentially the same process can be applied to any selected human language. In serving the purposes of this patent, the choice of language in no way limits the invention. In practice, the method of this patent can be applied to autonomously translate from one human language to another.

  Referring to FIG. 4, various components are controlled or modified by the state parameters previously described. Specifically, the clutter filter 40 plays a central role in determining what type of information is actually considered in the brain. As with most other blocks in the figure, the operation of the analyzer / correlator 30 is controlled or greatly affected by the personality state parameter 22. In many cases, this parameter itself can be the result of the analyzer 30.

  A flow of external information is input through the semantic analyzer 50. This extracts the content and intent from both English sentences and sentence fragments and formats the extract for storage in the short-term memory 10.

[Concept of neuron used in this specification]
The present invention is not intended to replicate biological neurons, axions and dendrites, their sequences or interconnections, or their redundancy. Rather, the term “neuron” in this patent represents a means to keep a fact or experience in memory. As suggested by mimicking organisms, the existence of one fact is simply represented by a single neuron, while the implied fact is included in the array of interconnections between neurons.

  In biological neurons, one neuron "fires" in place when the relevant facts are recognized. For example, in the fox brain, a particular neuron represents a common rabbit, and the firing of a biological neuron implies the rabbit's perception. The degree of ignition (or output) represents the certainty that the rabbit is recognized.

  None of the emulations or brain models of the present invention correspond to such a “fire” in place of neurons. In the digital implementation, the entire long-term memory 12 (which stores facts, relationships and experiences) is read-only memory or slow, since recognition does not include changes in the state of neurons in that memory. It can consist of flash memory.

  As an alternative process used herein, recognition occurs by the presence, recognition or relationship of data in the context pool memory 10. A reference to a “firing neuron” should be construed as placing a reference to that neuron (the address of that neuron) in the context pool 10 along with the current firing level for that neuron.

[Neuron and reference index]
Every neuron records two types of information. The existence of a particular fact is implied by the fact that the neuron to represent it was defined anyway. Experience is implied by the relationships and connections formed between neurons. As shown in FIG. 9, individual neurons are emulated with a fixed amount of basic information and a variable number of relational connection records. The relational condition is conditional and can be predicate based on the state of other neurons and refers to the ID index of both the target neuron and the conditional trigger.

  Every neuron has a unique address, but it may change over time as the memory is reorganized. In addition, some neurons are transient in their existence. Those neurons may disappear after a certain time unless they are strengthened, and are put in the strengthening memory 11. Since their exact location is unstable, referencing that neuron by another neuron can cause problems. Furthermore, the relative size of one neuron can vary greatly depending on the interrelationship and context with other neurons.

  In order to properly handle these problems, a unique and unchanging index is assigned to each generated neuron. References between neurons refer to each other using this permanent index. If a neuron is deleted (in enhanced memory 11), its index is retrieved for later reuse. A particular bit in the index value indicates whether it refers to a standard permanent neuron or to the enhancement memory 11. A fixed portion of the index for “temporary” neurons in the enhanced memory is reserved and used to indicate the information block type and format in the context pool 10.

  Neurons in enhancement memory 11 that have been enhanced over time are made permanent by analyzer / correlator 30. The analyzer then moves those neurons to permanent memory 12 and changes all references to that index to indicate that the neurons have been so moved. The references themselves within the neuron may not be able to survive the enhancement process and may be deleted during transfer. See Table 4 for data stored with individual neurons.

[Contents of neuron reference structure]
The analyzer / correlator repeatedly scans the context pool memory 10 for both raw information and activities that are suspended waiting for certain events or conditions to occur. The analyzer / correlator also updates the brain parameters to keep the brain parameters current and to examine the associated changes in content.

  Within the context pool, information is organized into variable sized blocks, all of which are pre-classified or typed before being presented. Some blocks contain intentions that are inferred from the sentence. Other blocks include instructions, propositions, guesses and other miscellaneous elements. In its degenerate form, a “block” can simply be a reference to a single neuron and its firing level.

  Each neuron is emulated with a certain size of basic information and a variable number of relational connection records. The latter is conditional and can be described based on the state of other neurons and refers to the ID index of both its target and conditional neurons.

[Context pool memory 10]
The central part of all emulation occurs in the context pool (short term) memory 10 and the analyzer / correlator 30. All information of immediate recognition for the emulator is in its context pool. Firing like a neuron is implied by the presence of a reference to the neuron from the long-term memory 12 in the context pool. Information (blocks), so to speak, sequentially enters the context pool, but is processed in parallel by the analyzer 30.

  Referring to the context pool 10 of FIG. 4, the data flows from right to left, so to speak. Unless enhanced, all neuronal data in the pool will gradually “leak” or disappear as it progresses and degrade over time. When the context pool is full, the oldest (ie, leftmost) data is simply lost, which is an example of overpacking information. Data that remains in the context pool but has deteriorated over time without being strengthened gradually decays to a zero firing state, at which point it is simply removed from the pool.

  Data may be entered into the context pool from a number of sources, the first of which is often the semantic analyzer 50. With the exception of input from the analyzer 30, all context pool information is filtered by the clutter filter 40, which prevents irrelevant or uninteresting data from generally reaching the context pool.

  The data in the context pool takes the form of a block structure in a predetermined format. For example, a block arriving from the semantic analyzer 50 includes the intention of a sentence, an independent clause or a sentence fragment. A single word answer to a question is meaningful enough as such a fragment. Such sentence blocks may include a reference to the speaker, a reference to the listener, and possibly a reference to the person or thing being discussed. Many combinations of this data and other sentence data are possible.

  Blocks from the analyzer 50 often include sentence purposes such as questions (and expected types of answers), commands, factual statements, observations, and the like. This type of data is discrete and easily identifiable by semantic parsing.

  Other implied emotional information can be inferred from superlative use, exclamation and tone (when derived from the auditory analyzer 60). An auditory sound source generates the nominal frequency of the speaker and estimates stress or emotional excitement by short or long pitch deviations that occur with the voice.

  The length of the context pool is determined experimentally depending on the application, but for nominal purposes, it should be long enough to handle several hours of intense research, or almost a day's daily interactions. is there. If multiple sizes are taken into the context, this corresponds to approximately 1 megabyte of conventional digital storage, but the means or method of this patent is not altered by the size selected. During sleep time (or emulated extended rest), the context pool 10 is gradually depleted and the firing of neurons gradually decreases to zero. When neuronal references are reduced to 0, they are removed from the context pool, as suggested by mimicking organisms.

  With the dreamer block 75, new information can be introduced during sleep. Dreamer-derived information, generated during deep sleep, decays rapidly when waking up, at a different rate than normal context pool data decay. If sleep time is insufficient, firing of still active neurons remains in the subsequent wakefulness cycle and is handled as described above.

[Language parser 50]
The language semantic analyzer 50 accepts the natural language of the embodiment, for example, English communication. It breaks down sentences, clauses and phrases and derives meaning and purpose from the sentence. It uses the current conversation or interaction context by polling the analyzer 30, the long-term memory 12 and the enhanced memory 11. Access to the current context is obtained indirectly from the context pool via the analyzer 30. The word interpretation of the language is weighted by the presence of related neurons in the context pool, resulting in an accurate interpretation along the context.

  The language semantic analyzer 50 can be hard-coded with logic, but for many applications it is conveniently implemented as a built-in processor. This method is not necessary to achieve the present invention, but is advantageous for parsing and interpretation of languages other than the initial design language.

  Because all humans are generally the same regardless of their national language and their grammar or semantics, the parameters described herein remain the same, but the language description script of the semantic analyzer language 50 will vary. I will.

  For convenience, the statements output by the analyzer 30 through the interface 98 are generated in the analyzer 30. However, this function can be conveniently separated into individual units when changing the language selection from English.

  For a given language, the semantic analyzer 50 recognizes a set of words that are generally unchanged components of the language, such as English with or for. This plays a major role in defining the language grammar. Nouns, verbs, and adjectives change easily over time, but the underlying structural words that make up the underlying grammar rarely change.

  In addition to these unchanging “grammar” words, sentence, clause and phrase structures define the rest of the grammar. The analyzer 50 interprets the intention of communication using the entire grammar.

  Computer languages (non-natural languages) are often parsed by individual vocabulary and grammar parsers using commercially available tools such as Lex and Yacc. These seemed cumbersome and cumbersome when parsing in the system of the present disclosure. Another parser (Lingua, a commercial parser, not the subject of the present invention) has been developed for natural language. When using Lingua, a nearly complete description of English grammar is defined and plays a central role for the language semantic analyzer 50. The intellectual property rights included in this are definitions of the English grammar itself, but that is not the subject of the present invention.

  In the prior art, custom analyzers that use large corpus or dictionaries of words have been used for parsing English text. Unlike them, the semantic analyzer 50 uses context sensitive information to interpret the intent more accurately from the text.

  The semantic analyzer 50 captures sentences, clauses, phrases and words in natural language and outputs decoded neuron references and inferred intent blocks. On a large scale, basic grammatical words that do not change are discarded after they serve their purpose in parsing. Similarly, structural constituent words in a sentence are often discarded after the suggestion is sought. Finally, pronoun references such as he and it are replaced by references to neurons representing the transformation target, such as “David Hempstead” or “rabbit”.

  The semantic analyzer can indirectly refer to both the long-term memory 12 and the “21-day” enhanced memory 11 and extract the relationship information from either to determine the meaning and intention of a particular word. It places more weight on the word whose neuron reference is already fired in the context pool 10.

  English (or other natural language) grammar definitions are contained in a definition file in a variant of Baccus-Nauer Format (BNF). See FIG. 5 for an example fragment of such a definition. The example was implemented using the Lingua compiler, a product commercially available from Neuroc Technologies. An example of a diagnostic result obtained by parsing the sentence, “The table failed”, is given in FIG. 6 and shows the repeatability of the parser used in the commercial Lingua product.

  It will be readily apparent to those skilled in the art that language analyzer 50 can be implemented in a variety of ways without compromising its placement and effectiveness in the system of the present disclosure.

[Sentence block]
To process a sentence, the data in the context pool 10 can be divided into inferred facts and data. The preprocessing in the semantic analyzer 50 has already converted the sentence fragment into a complete sentence, or the fragment will be flagged for expansion by the estimator.

  Each sentence block is usually a complete sentence and has a subject and a predicate. In the implied subject you, the subject has been resolved and replaced with an appropriate neuron reference. An implied prefix It is also added as needed to change the noun clause (eg, the answer to the question) to a full sentence. All sentence blocks are formally standardized and the sentence information that is inferred is rearranged in that form.

These blocks have variable lengths and store a flag indicating sentence data. Some of this information is derived from the state parameters. The statement type indicates which elements are optional. Types include declarations, questions, exclamations, observations, condemnations, answers to questions, etc. Other sentence data may include the following (and other) information:
Subject subject person (first person, second person, or third person)
Number of subjects (single or plural)
Subject Gender (male, female, thing)
Action or action to be taken Verb Object (including person, number, gender)
Goal of action (including people, number, gender)

  All noun-like elements also include a person flag, a number flag and a gender flag. These sentence blocks are interpreted by the analyzer / correlator 30 and the estimator 70 as commands for interpretation. Some of these blocks are described in the description of the contents of Table 7.

[Sentence recognition process]
Whether the sentence is obtained from written text or from auditory speech, the recognition and understanding of the contents of the sentence is generally the same. The biggest differences are the additional matching, validation and filtering imposed on the speech case. In order to extract intentions from sentences, three general communication elements are defined. The elements are speakers, listeners / objects (eg, command recipients) and the person, object or subject being discussed. Most of this information can be inferred from the content of the sentence, from the current context pool 10, and from the state parameters 20 and 23.

The basic process is as follows.
1. Parse-parse the sentence using linguistic grammar rules as shown in FIG.
2. Extract the characteristics of the three elements—if any, identify shifts in the communication three elements. If a change is identified, the correlator 30 is notified by an appropriate command notifier in the context pool 10.
3. Extract any modifiers-compile modifier clauses. Including compiled statements of definitive sentences, but sufficient neuron references and data to generate additional relational connections 1252 when evaluating the probability of a node for a single neuron in another way Extract both.
4). Extract structural elements-Extract structural elements that are clues and discard semantic information. Data is stored in block or neuron references suitable for use by correlators 30 and 75.
5. Compile boilerplate-Compile any boilerplate into relational and modifier elements and store relational associations (if any) with related fact neurons. This is done indirectly by issuing appropriate instructions to the context pool 10.

  The above basic process illustrates some of the typical activities for parsing a sentence and generating an information block or command block for inclusion in the context pool 10.

[Clutter filter 40]
The clutter filter 40 serves to restrict the entry of certain types of information to the context pool 10. Information that enters the context pool must pass through a clutter filter, except for what is output by the analyzer 30. The purpose of the filter is to remove unrelated neurons such as linguistic or grammatical tokens and meaningless gesture information. The clutter filter follows preset heuristics, which may be fixed or adaptive.

  The result of the filter is to focus on considering relevant information and to minimize the “disturbing” personality being modeled and the ones that are of little interest to that personality. For example, bile quality does not find value in human interest information, unlike polycyth. Given the current parameter conditions, the data so identified can be removed. This can be done during the conversation process, in which the semantic analyzer 50 or other source flags the data based on the topic of discussion.

  The clutter filter greatly contributes to the difference in emulation in the activities of the right brain and the left brain, and in this respect, it does not take any component except for the function of the analyzer / correlator 30.

  During the interaction with the outside world, a number of neurons are referenced from memory and given to the context pool 10 for analysis, correlation, speculation, and dreaming. The filter may consider some of the inhibitory factors, along with the type and grouping of a given neuron, and choose to discard those neurons instead of forwarding them to the context pool 10. During normal (non-sleeping) activity, the output from the dreamer 75 is given a very low priority as long as the overall level of firing of neurons in the context pool 10 is not very low.

  The neuron phrase resulting from the analyzer 30 always bypasses the clutter filter and goes directly into the short-term memory. Inevitably, the analyzer / correlator dominates the entire thinking (and memory) process and usually does not generate clutter.

  The filter also prioritizes incoming information. The entire content of the answer to the question is also likely to be passed, whereas the same component will normally not be passed.

  The main criterion in determining what constitutes “clutter” is the personality parameter 20, which is part of the state parameter 22 (in FIG. 4, that parameter is for emphasis and clarity). Are shown separately from the other parameters, but are basically the same). The logic as shown in FIG. 3 illustrates one means by which clutter can be determined. It will be apparent to those skilled in the art that a clutter filter as described herein can be supplemented with additional rules and heuristics without altering the basic invention of this patent.

[Analyzer / Correlator 30]
The analyzer / correlator 30 is the central part of the emulated brain and is the main center of activity for thought processes. It is also the main means for updating all dynamic brain parameters and is the only means for initiating permanent storage of information.

  Decision making is usually based on “unwavering” facts, high confidence information or firing. Generally speaking, the higher the perceived quality of the source information, the better the quality of the decision. If there is no good information, the analyzer 30 uses the information from the estimator 70, but the quality is still degraded as a result of using the latter.

  The thinking and decision making process is performed by the analyzer block with the support of stimuli and suggestions from the estimator 70 block and the dreamer 75 block. The central part of the analyzer's work is performed in the context pool memory 10 so that all processes are performed within the current context.

  Entering a neuron reference into the context pool memory 10 initiates a series of events specific to that neuron and its associated relation (experience) connection, or “relation”. As will be described in detail later, these often utilize event queue memory 14 to handle the implications of their connections.

[Initial activity at awakening]
When waking up in the morning, the resting heart (ie, the context pool 10) is usually completely empty. Yesterday's thoughts and concerns are either completely forgotten or weakened so as not to be easily recalled. The previous day's sentence fragments, fleeting observations and incomplete or illogical concepts have been removed, and there is no disturbance in the mind. This is the context at awakening.

  The daily activities in this brain emulation start in the same way. The initial trend is to seek the help of a daily activity list that is usually established by performing an event from the event queue 14 at a specified time. Activities can also be initiated by other external means in both human life and this brain emulation. Table 5 lists some examples of how to start activities in the morning, but the list is of course not inclusive.

  Each of the above conditions places a block of neuron references that takes the form of statements, commands based on events, and other information to be processed. Those skilled in the art will appreciate that the analyzer / correlator 10 can be implemented as hard-coded logic, as a form of command interpreter, or as an embedded processor without altering the means of the present invention.

[Results of activity of analyzer / correlator]
As a result of that operation, the analyzer / correlator 10 can include any of the activities in Table 6. The list indicates the type of result and is not exhaustive and can be expanded for convenience of implementation. It will be appreciated by those skilled in the art that this does not change the means of this patent.

  In addition to the elements in Table 6, the analyzer / correlator 30 maintains a number of lists such as the subject of the current conversation or question, the status of incomplete answers to the questions made, the maintenance and completion status of motor skill activities, etc. And update. The main source of information and commands comes from the current contents of the context pool 10.

[Context pool command]
Within the context pool 10, information and facts are stored in a general way as neuron references, ie neuron indices. Both the state parameter 20 and the context pool command are encoded as dedicated lower values of the neuron's index. The length of these commands is variable and their index is followed by length and auxiliary information.

  A number of synthesized commands are derived from sentence parsing by the language analyzer 50. From the sentence, it is possible to extract a number of commands, each supplemented with a neuron reference. An implied subject, verb or object is resolved with a reference to the associated neuron. For sentences with multiple subjects, verbs or objects, the contents of the sentence are duplicated, for example one for each element in the subject list.

  Several commands found in the context pool 10 are given to _Ref90637160-. The list is exemplary and not exhaustive. Those skilled in the art will appreciate that the list can be expanded without changing the means of the system of the present disclosure.

  For convenience, all data structures in the context pool look like neuron references.

  Execution commands are always flagged by their source, such as a speech or grammar analyzer, analyzer / correlator, estimator, dreamer, etc. The analyzer later considers its source when applying commands in its thought or decision making process. Exemplary commands from the semantic analyzer 50 are given below, and these specific commands are based on sentence type.

[Declarative 231]
This is an instruction to consider the current conditions on the subject. It can also be part of the experience process, and eventually ends with the generation of a neuron-neuron or neuron-state parameter relationship. This command is usually generated by sentence parsing, but can also be generated by a thought process in the analyzer 30.

  As a result of the clarification, the relationship can be stored in memory after confirming again and through the operation of the estimator 70. That is, the story is “received in half” and takes into account the credibility of the source of the opinion. Definition 233 is already considered to be the source of facts, and in order that only the reliability (certainty) of the information needs to be confirmed before the information is stored in memory, Different.

  For example, “Four cats are sufficient to eliminate mice from large barns” is a descriptive sentence, which is what it takes to get the job done. Describe whether you need a cat. The analyzer 30 will consider the credibility of the source of the findings and consider whether to re-examine the information before assuming and storing the statement as fact.

  Instruction 232 may be an analyzer that causes the brain emulation to do something, for example, let the proposal be reviewed, pay attention, remind something, or guess a solution to the problem with insufficient information. 30. It is a command for certain types of actions that are directed to brain emulation.

  Commands such as “Come here!” Must be evaluated in the current context. It implies starting a motor skills list and starting a physical movement and targets the location of the speaker. The latter may not exist in the context pool and is kept in the state parameter. In this case, the analyzer 30 guides the motor skill via the task list 13. At that time, for example, a completion waiting event 142 can be issued, and this command can be discarded from the memory. Later, a completion message (or a record of encountering a major wall or other obstacle in executing the instruction) is received and the command is terminated.

  Definition 233 directs the definition of the fact (in enhanced memory 11) and may include auxiliary conditional relationship information. Example: "A cat is an animal with have four paws, of which the front two are commonly called" forepaws ")" is a compound statement. The statement shares a common topic, with individual definitions 233 ("A cat is an animal with four paws") and verse 231 ("Before" The cat's leg has a clause "The front cat paws are commonly called forepaws"). The semantic analyzer 50 separates the compound sentence into separate commands for each clause.

  The part of the plain text 231 “Cat is an animal with four paws” is “Cat”, “Animal”, and “Paws”. If these are not yet known, define these neurons. Even if the meaning of "animal" or "leg" is not known, they can still be remembered and an appropriate relationship can be formed between them later. These are all recorded in the enhanced memory 11 if they are not yet present in the enhanced memory and are not known in the long-term memory.

  If they are already in the enhanced memory 11, their existence is re-checked so that they can be stored permanently. If the speaker is a very honest person, it doesn't take long to reinforce that fact. If the system is in a preemptive training mode, these are presumed to be naive facts probably given by God, and are immediately and permanently remembered.

  Part 231 of the text, “The front (cat) paws are commonly called forepaws”) also forms a definition, but is more elaborate than the definition clause. It must be reconfirmed (since parsing has already been performed, the explicit subject defined at the beginning of the sentence is already associated with the subsequent clauses by the semantic analyzer 50).

  Since there is a definite article ('The'), the clause is not a definition 233 but a description 231. This is because the reference is made to a specific cat, not to the general “cat” animal. Those skilled in the art are aware of these details of English grammar and know how the grammar can be used to determine sentence intent and type.

  Question 234 describes the question and request. These are usually placed in the context pool 10 by the grammar semantic parser 50, but questions from other sources can also be used. Many (but not all) questions are simply plain statements in which the questions are directed and are often formed by reorganizing simple plain texts.

  The parser 50 classifies the questions into questions that require confirmation (yes / no) or questions that ask for specific information, and the questions are either as a test 231 instructed to be validated, or It is provided to the context memory as an instruction 234 that requests a response providing information. In any case, the analyzer 30 sees only the data constructs indicated as questions that can form a response to the question to obtain the latter form.

  For convenience, other internal commands are also added, and the analyzer 30 generally takes the form of a von Neumann processor, in which case the “program” is a command stream from an English parser or from another block.

  In communicating with a brain emulator that shares a common memory 12, those analyzers 30 can transfer the "summarized" command block directly to the emulator's context pool. When communicating with the outside world via the external interface 98, the analyzer 30 reformats the command block into English for parsing there and receives English via the interface 93.

[Neuron and context pool]
The conditional statement expects a specific neuron (or combination of neurons) to be fired. The state parameters 20 and 23 are pseudoneurons and exist before all assigned neurons. Those parameters are treated as neurons and assigned the lowest index ID number, but do not have the relation (experience) connection generated for those parameters. The IDs of all firing neurons (excluding state parameters) are kept in the context pool, along with some information specific to that neuron, including the degree of firing.

  Aged neurons in the context pool 10 that are no longer fired are removed from the pool memory, usually during “sleep”. Neurons that are still firing but have not been reconfirmed or refired in the context pool will have no effect other than establishing the current context. For example, the neurons can be subject to conditional testing or can change the contextual meaning of the sentence being parsed.

[Unidirectional relationship]
If the relationship is unidirectional, the relationship tied to the “causing” neuron emits an event, but only if the specified condition is true. For a unidirectional relationship, A implies B, but B does not imply A. In either case, the relationship is conditional and can be described in other neurons that are also firing. Referring to FIG. 10, an association connection 1253 is created in a neuron affected by the relationship.

[Bidirectional relationship]
If the relationship is bidirectional, the neurons or state parameters on either side of the relationship will issue events. If none of the specified conditions are met, the event is not fired. In the case of a bidirectional relationship, A implies B and B implies A. In either case, the relationship is conditional and can be described in other neurons that are also firing. Referring to FIG. 10, relationship connections 1253 are created in both neurons in the relationship, each referring to the other.

[Relationship to issue event]
When a neuron first fires (or is reconfirmed), the analyzer 30 scans the list of associated relationships. Those relationships are organized as an AND-connected list, optionally separated by an OR marker. Successive relationships are evaluated until one of those relationships fails or until the OR marker is reached. If one relationship fails, subsequent relationships are ignored until the next OR marker or until the end of the list.

  On failure, if the OR marker is reached, the failure condition is reset, the OR is ignored, and the test resumes in the relationship immediately following the OR.

  If the end of the list is found first after failure, no event is generated. Finding an OR (or finding the end of the list if all previous tests are successful) implies that all ANDed relational conditions have been met, so an event is generated Is done. A conditional relationship can be flagged with NOT, implying that the reverse of the condition must be true for the relationship to succeed.

[Other internal lists]
The analyzer / correlator 30 maintains other lists of information similar to the list of state parameters 22 in the short-term memory, and these lists are also treated as predetermined neuron blocks. These lists are described elsewhere in this patent and include the following lists:
Discussion topics Ongoing athletic activity Events that are awaiting completion Multiple objects applied to sentences Multiple verbs applied to sentences

  The above list is by no means exhaustive and that the logical or physical arrangement of the above list can be changed or the list can be added without changing the method of the present invention. Those skilled in the art will appreciate.

[Walking across neuron connections]
When a new command is added to the context pool 10, it typically includes a reference to a neuron that represents the fact or condition of existence. Usually, the command will refer to more than one neuron. Each such reference either brings that neuron “into the pool” or reaffirms a neuron that is already in the context pool.

  By simply referring to a neuron, the analyzer 30 brings that neuron into the context pool even if it is not firing very strongly. Some command blocks, such as from the definition section, greatly increase the level of firing. By repeatedly referencing the same neuron over a relatively short duration, the firing level is increased to the 100% level.

  For example, recognition of a person's face brings the person's ID into the context pool and fires the associated neuron according to the confidence level of the recognition (eg, “It might be Jackie 'maybe'”) . Immediately thereafter, listening to the voice of the same person increases the certainty of identification. Therefore, the firing of the person's neuron (ID) may possibly increase from 65% to 95%. As long as the interaction with the person continues, the person's ID continues to exist in the context pool.

[Correlation of relationship information]
When neurons in the pool fire, other neurons may be implied by known relationships. For example, if either “birdcage” or “South America” is currently in the context pool, “green” and “animal” may imply “parrot”. Rather, if the “swamp” is ignited, the “alligator” may be ignited. The analyzer / correlator 30 accumulates the initiated reference in the context pool 10 and updates the neuron firing as specified by the scaled connection weights.

  In the case of firing initiated by such a relationship, the firing level is controlled by the value of the referencing neuron (eg “green”, “animal” or “swamp”) and the weight given in the relationship connection. The That is, if "Florida" (which may imply "swamp") is weakly fired, "alligator" neurons will fire weakly, even if there is nothing else that directly activates "swamp" Will. The analyzer 30 effectively acts as a correlator by walking through the connections of all firing neurons, and awakens other neurons unless firing is suppressed by a conditional relationship.

  Referring to FIG. 7, if both “dog” 121 and “excitement” 122 are ignited (eg, information inferred from the parsed sentence), references to them are in the context pool 10. Can be put. The relationship of FIG. 7 would set the expectation that a dog would bark through the neuron 123. Weights 124, which may be different from each other, are multiplied by the firing levels of 121 and 122, respectively. If any of the resulting firings exceed a certain minimum decision threshold, the AND operation 125 causes the general dog-bark neuron 123 to fire. In doing so, a reference to the neuron 123 is inserted into the context pool, possibly initiating a motor skill event, e.g. causing a bark. It will be apparent to those skilled in the art that many variations of FIG. 7 can be implemented without altering the means of the present invention.

  Again, the analyzer 10 gradually lowers the firing level of any neuron that has not been reconfirmed over time or has not been refired. If the neuron then goes to 0, it is removed from the context pool. If the neuron is still firing but has been there for a long time and the context pool is full, the neuron is also discarded from memory.

[Long term and enhanced memory]
Reinforced memory is an intermediate point in the process of learning and remembering things. All new information and relationships are established in the enhanced memory, which serves as a filter for sufficiently important elements to remember later. The analyzer 30 handles this process.

  The enhanced memory 11 is a means of removing non-essential facts, relationships and events that would otherwise unnecessarily disrupt the permanent memory. This mitigates the eventual increase in long-term memory 12 and keeps mental processes and memory more efficient.

  Much of the information and experience we face is accidental and not worth remembering. For example, even if the paper blown in the wind is instantly recognized, unless the context is the distribution of a promotional leaflet, the event is insignificant and cannot be memorized. If so, it might be worth considering. The enhanced memory 11 is a temporary storage location while this information is revalidated or forgotten. As explained elsewhere, the analyzer 30 permanently transfers the validated facts and relationships to the long-term memory.

  The long-term memory 12 and the enhanced memory 11 share a common format more or less. The assignment of neurons and relationships is handled entirely by the analyzer 30, and the policies that govern permanent storage belong to the analyzer 30.

Information is allowed to be “storable” by the analyzer 30 when repeatedly referenced for 21 days or repeatedly referenced during intense emotions or trauma. If so validated, analyzer 30 moves the information to long-term memory 12. Referring to FIG. 8, the related relationship is also moved from the enhanced memory 11 to the long-term memory side. Each memory consists of the following elements.
ID table 126
Neuron Table 125
Tables specific to other emulators

  The “other” table includes a dedicated table. This dedicated table is associated with a single neuron and is used to recall motor skill task lists, auditory or visual artifacts, or objects. The format of those tables is specific to the emulator type (eg, visual, audio or motor skills) that generates them, but they follow standard processing and correlation rules for normal neurons .

  The neuron itself is not special. Rather, it gains meaning and value from location and interconnections to other neurons. For example, “laptop” neurons are not meaningful by themselves (except for spelling, pronunciation and visual shape), but are important because of their relationship to “computer”, “portable” and “convenience”.

  The following section describes one specific implementation of the emulator structure. Those skilled in the art will appreciate that the techniques of the embodiments are complementary to the means described herein. Many of these elements may be minorly modified, depending on the underlying technology being different, such as software emulation, FPGA, gate array, embedded processor, analog relational array, or optical logic, or It will be implemented in various ways.

[ID table]
Referring to FIG. 8, every neuron is assigned a serial number 127 which has no meaning to itself. Each relational connection to another neuron uses its invariant serial number as an ID. From the ID, spelling, pronunciation and other relevant information can be obtained.

  When the memory is implemented as a digital memory, the ID table 126 is preferably located at the bottom of the memory and consumes a predetermined and limited logical space. The table is sized to have one element for all possible neurons. In practice, the memory can be resized as more and more tables become physically available, and the appropriate offset is set to the resolution value for each ID in table 126. Applied. For each index 127, the corresponding offset into the ID table 126 includes the address of the neuron in the neuron table 125.

  When considering only words, a 30,000 word vocabulary is an acceptable work size. For some people, up to 300,000 unique words are known. Each concept that should be kept in memory, for example “off the wall”, has its own index, events or conditions that are kept in memory, resulting in those words. Each word corresponds to a unique neuron record 1250 in the neuron table 125.

  Experience may or may not have its own index, depending on what the experience is and how it has been formed. For this purpose, it is practical to have an index table 126 composed of, for example, 8 to 20 million elements or more.

[Neuron Table]
Referring to FIG. 9, a neuron 1250 is emulated by a fixed size information block 1251 and a variable number of relationship connection records 1252. The latter is conditional and can be described in the state of other neurons. They can reference the ID index 127 of both their target and conditional neurons. If more suitable hardware memory technologies are available that can directly form relational connections between neurons, the link-pointer structures that rely on these technologies may be unnecessary and therefore eliminated. Can be replaced.

  The basic information 1251 can include references to explicit spellings (eg, a reverse search index on a text tree for words), pronunciation exceptions, visual object descriptors, and the like. Specific flags for lexical content, starting indexes, etc. are also included here.

  A “relation” 1252 is a connection between two neurons. It may be a connection between neurons and state parameters. The relationship may be unidirectional or bidirectional in nature and can only be performed if a specified set of conditions is met. The relationship is largely suggested by biological neuronal dendrites.

  When implemented in digital memory, the relationship is conveniently assigned in the space immediately following the fixed length portion of neuron recording 1251. Usually, a vacancy is reserved there in anticipation of the relevant record insertion. Before inserting a new relationship, the analyzer 30 checks whether there is enough space, and if there is not enough space, it reassigns the entire neuron with a larger space.

  The length of the relationship detail block 1252 is variable, depending on the type and number of relationship connections made to other neurons. It is not unreasonable that all (digital) memory can consume from 16 megabytes to 2 or 3 gigabytes.

  Relation 1252 has an AND-OR organization. The relationship records that are ANDed are grouped together after the fixed length portion of the neuron.

  Referring to FIG. 10, a specific target ID 1256 is generally defined to represent an OR condition, and the rest of the “relation” record is ignored. As stated elsewhere in this description, specific neuron IDs are reserved for such special purposes. Similarly, certain values of weights 1257 are reserved to indicate prohibition conditions, and the weights themselves are negative, and can reduce the level of recognition, i.e., the firing level.

  The relation 1253 itself is unidirectional. The neuron 1250 of which the relationship forms part is fired to the extent that the neuron referenced by the target ID 1256 fires. However, the firing of this neuron 1250 does not affect the target ID 1256 unlike this. For example, “grass” may imply “green”, but “green” does not imply “grass”.

  For conditions where the relationship is bidirectional, the analyzer 30 generates a suitable relationship pointing to each other for each of the two neurons. This is similar to a software doubly-linked list.

  The weighted and conditional influence of this neuron on another neuron is defined by relational connection 1252, which may exist up to 1000 or more depending on the neuron. Each new experience and relationship that is learned has a new relationship connection that is created for it. Garbage collection and management of neuron-related memory space is described elsewhere in this patent.

  Initially, new neurons 1250 and relationships are created in the enhanced memory, and they remain there until later validated and transferred to long-term memory or deleted. Relations 1252 in enhanced memory can refer to neurons in either memory, but relations in long-term memory can only reference other neurons in long-term memory 12. The analyzer 30 tracks the allocation, aging, validation and “garbage collection” processes as described in detail elsewhere.

[Other tables]
In addition to pure neurons or relationships 1250, both enhanced memory and long-term memory can hold other encapsulated information. These data blocks are treated and referenced like normal neurons, but contain complex and extended structures for efficiently recalling complex entities later. Details of each of these data blocks are described along with a description of their associated neurons.

  The neuron process for recognizing scenes and sounds is done by reconstructive correlation, matching a reference image or sound with a known object or sound. Memory storage is “reconstructive” in that no actual sampled sound or pixelated images are stored. Rather, enough information is kept in memory to reconstruct the “reference object” (for comparison). In that case, the stored images and sounds are not detailed information about them but a list of artifacts of the object. The degree of coincidence or similarity determines the neuronal firing level.

  See Table 8 for a list of some common support tables. Those skilled in the art will appreciate that this list is by no means complete and that there are numerous ways to organize such information into tables without changing the means of the present invention.

  “Recognition” and “reproduction” of visual objects are different processes and must be optimized individually. Biological features suggest that humans do not store details like bitmap images. Still, humans can reliably recognize detailed objects and accurately identify them when they see them. A correlation template is reproduced from the stored table information and applied to an appropriate correlator. For example, a vector skeleton for use by a visual correlator for specifying an image can be used. Neurons fire in proportion to the degree of matching.

[Event queue and memory 14]
“Event” is a dedicated command issued to the queue 14. Those events are made candidates for later execution at a particular time after a specified delay or after a specified set of conditions are met. Those events are a means of avoiding undesirable looping in the information in the context pool memory 10.

  An event is simply a marker or flag that is specified to remind the system to do something when a specified condition is met. It greatly simplifies the handling of operations that are not synchronized with each other. When the analyzer 30 discovers new information in the context pool 10, the analyzer can issue one or more “events” to the event pool 14. For example, the analyzer may generate an event that adds a new reference back to the context pool. The analyzer should also issue a conditional event to force the analyzer to repeatedly scan the context pool again later, which can happen in the case of an analytical temperament such as depression. Can do.

  The same mechanism is also used to establish a conditional relationship between neurons or between neurons and state parameters. An event can be generated by changing the state parameter 22. By issuing an event for future execution, the analyzer 30 avoids deviating from the task being executed.

  Referring to FIGS. 11 and 4, the event queue 14 includes an interpreter 140 and an event list 141. By generating an event, event 142 is inserted into the event list. The event 142 in the list consists of the command field and other optional fields shown in FIG. The interpreter scans the event list repeatedly for events that can be processed. Whether or not these events are processed is determined by the “condition” and “timing” fields. If present, the “auxiliary data” field contains information specific to the event type. Once an event has been processed, it is removed from the event queue.

  The interpreter 140 scans from the beginning after scanning to the end of the event list 141. If there are no events left to process, the interpreter waits for the generation of a new event. Event queue 14 implemented as hard-coded logic, micro-coded processor, software emulation, embedded processor, FPGA, ASIC, optical technology or other optimal technology without changing the means of the present invention Those skilled in the art will understand that this is possible.

[Guessor 70]
The estimator 70 proposes decisions based on incomplete or partial facts or facts with low confidence. The main thinking function for the emulator is the analyzer 30, which receives advice and suggestions from both the estimator and dreamer 75 blocks. Suggestions from the estimator are filtered by the clutter filter 40 based on temperament and personality.

  During processing of sentence data in the context pool, the analyzer / correlator 30 operates based on the sentence block to determine an appropriate action plan, if appropriate. If "does not go as expected", the analyzer calls a guesser that suggests a valid meaning. If the resulting estimator output quality is too low, the analyzer 30 can ask for an explanation through the communication interface 98. The analyzer sets an appropriate parameter flag and waits for an answer to a question seeking explanation.

  The estimator output is similar to a normal neuron reference or sensory nerve that fires at a relatively low level in the case of the topic. The output of the estimator 70 is essentially the same as the data inferred from the sentence by the semantic analyzer 50, except that it is flagged as originating from the estimator.

The estimator works in the same way as the analyzer 30, except that it only looks at the components in the current context pool. The estimator is not limited by the demanding facts similar to the analyzer, but effectively provides subjective information for consideration. The proposal is largely ignored by the analyzer except in the following cases.
Information is missing or incomplete.
Questions asked by the analyzer through the communication interface 98 have not yet been answered within the expected time.
The overall confidence (ignition) level of the information in the context pool 10 is low.
In practice, when the analyzer 30 cannot obtain the answer from the existing information, the analyzer queries the estimator and fills the void.

  For that operation, the estimator 70 reviews the unresolved question or problem as defined in the context pool, support table, and appropriate status parameters 23. Several status parameters track the current topic subject (s), the question being asked, and the information currently sought by the analyzer 30. Based on these factors, the estimator scans even low firing neuron references and commands in the context pool 10 and proposes (guesses) the answer that the analyzer seeks.

  Whether or not the estimation is emphasized by the analyzer 30 is implied by weighting the estimation. If proposals conflict with other information, or if better (strongly ignited) information is available, those proposals are ignored. Inferences change rapidly over time and are immediately forgotten from the context pool, regardless of whether they have been affected. The analyzer takes into account the “information” source of the estimator and its certainty (ignition level). The analyzer then clarifies if the proposal is needed and how reliable the data is. The rejected guess is immediately erased.

  The estimator 70 is implemented as hard-coded logic, micro-coded processor, software emulation, embedded processor, FPGA, ASIC, optical technology or other optimal technology without changing the means of the present invention Those skilled in the art will understand that this is possible.

[Dreamer 75]
The dreamer 75 serves as the “right side” of the brain emulation of the present invention. It examines the neuron references in the context pool 10 in detail and uses weights for the state parameters that are different from those used by the analyzer 30 for its input and decision making process.

  Dreamers affect the analyzer by introducing fired neuron references into the context pool, unlike strictly structured commands such as from the semantic analyzer 50. If the information previously present in the context pool comes from the visual or auditory source 60 or from visual neuronal correlation, the dreamer can output a suggestion in the form of a command block.

  Similar to the correlator-analyzer 30 processing method, the dreamer generates new references and commands based on the existing neuron firing. However, when traversing the neuronal relationship chain, the relationship condition 1252 is given a low rating, as shown in FIG. The resulting output is unreliable as indicated by both its source and its firing level. When the analyzer 30 is not active or in sleep mode, the dreamer can indirectly change the topic subject by publishing an event to the event queue 14. Due to the “noise” level involved, the dreamer can quickly move from topic to topic. The dreamer also remains active when the brain emulation is in “sleep” mode.

  When later processing the context pool data generated by the dreamer, the analyzer 30 does not generate new neurons or relationships in the enhanced memory 11. Upon waking up from sleep mode, the analyzer 30 also quickly removes the unprocessed dreamer generation “information” left in the context pool.

  Therefore, Dreamer behaves as a kind of “movie-maker” that is not constrained by relational logic. It creates a new concept based weakly on the current context, and the lifetime of that concept decays very quickly. This neuronal firing is not a logical or coherent aspect, but it still affects the decision making and analysis performed by the analyzer.

  The dreamer 75 is based on an algorithm that statistically ignores strongly firing neurons and applies logarithmic weights to the firing neurons as part of its process. In this way, the dreamer examines the context pool in detail and effectively weights the neurons that are hardly fired.

  The effect of additional neuron firing in the context pool 10 is that the dreamer gives the neuron a greater overall weight than the weight that the analyzer has. In the course of activity, the firing of a neuron will be increased due to the repeated reference of some neuron. The analyzer 30 appropriately weights the information flagged as originating from the dreamer and continues to apply its normal logic to the data. If the analyzer is looking for new ideas, the analyzer will weight the reference induced by the dreamer more than usual.

  Because the dreamer 75 operates at an effective threshold that is lower than the threshold useful for the analyzer 30, it is more prone to “noise” and errors than the analyzer. The output is unreliable as long as decisions are made, but the purpose is different. During non-sleeping operations, the dreamer pseudo information passes through the clutter filter 40 where it can be removed by the personality and temperament filter. During non-sleeping operations, the clutter filter removes more dreamer output by changing the negative filter threshold.

  Dreamer 75 can be implemented as hard-coded logic, micro-coded processor, software emulation, embedded processor, FPGA, ASIC, optical technology or other optimal technology without changing the means of the present invention Will be understood by those skilled in the art.

[Audio and visual analyzer 60]
The emulated brain of the present invention can be applied to a mechanical system, whether it is a skeleton or a vehicle, and uses a motor skill learning function based on a list. External hardware can be controlled using an interface from the task list handler 13, event handler 14 or analyzer / correlator 30. These interfaces can be used to apply a specific level of force when used with closed loop feedback or when used in a specific mechanical position with or without feedback.

  The sensor used for the feedback system is determined by the application. For example, placing a hand on the table requires prior knowledge of the height and position of the table, or feedback as derived from both eyes. A sensor suitable for the nose or hand may be a pressure sensor (a person will not hit the wall more than once). The auditory sensor confirms that the sound is properly configured, for example, provides feedback for singing existing music without losing pitch.

  The method of the present invention generates a correlation template or suggestion, i.e. a visual or audible object that is provided for correlation to a visual image or sound. Using a binary search method, a suitable template for correlation is selected and the degree of recognition is quickly determined. The correlation method constitutes a “processed sensor”, i.e. a sensor having the ability to confirm the degree of recognition therein.

  An unprocessed sensor is simply a temperature, pressure, humidity or photometric device whose output is simply formatted to be suitable for input to an interface. Processed sensors require interpretation and possible correlation, after which they can produce meaningful signals. For example, using any number of algorithms, the visual sensor takes a template image and returns the degree of correlation of the current image. Similarly, the processed auditory sensor takes a prototype as in the case of phonemes and returns the current degree of correlation. If the matching word has a neuron firing in the context pool 10, a phoneme change can be proposed.

  The audio and visual analyzer 60 uses a task list such as 13 or other memory to retrieve the next image template for correlation, as suggested by the analyzer 30. These are communicated as the current setting of the relevant state parameter. For example, some motor skills require visual feedback to recognize the position of the table, its upper surface, and the portion of the hand that will be placed there. These motor skills also sort out objects that must be recognized by a visual correlation process.

  When a table surface is identified, its position must be reported to the context pool 10 so that it is the appropriate landing point position on the table, i.e. the appropriate area indicated by the intent and desires of the analyzer 30. I must. The visual correlation output is conveniently performed relative to the eye location of the skeleton so that correction for hand movement can be performed.

  Particularly in the case of a visual recognition process, motor skills require feedback to determine position, speed of travel, distance, and the like. Multiple feedback streams can be derived from a single sensor (eg, the “eyes” of a pair of cameras), and that information is transferred to the context pool 10 as command or event packets.

  Visual and auditory cues help confirm recognition and provide feedback for the required motor control. These cues, for example, rotate and tilt the head appropriately and then direct the left and right and up and down movements of the eye so that the detailed center of the foviated vision is the part of the scene of interest. Needed to be placed in the middle of. These matters are processed by the list processor 13 and the visual / auditory analyzer 60 in an interdependent manner.

  The speech analyzer 60 emits its output to the semantic analyzer 50 in order to actually parse the speech and make it a suitable element for the context pool 10 memory.

  As is known to those skilled in the art, it is clear that there are numerous techniques for such processed sensors. According to the present invention, the template information can be interactively expressed by the sensor together with the brain emulation function. The visual analyzer 60 itself can be hard-coded logic, micro-coded processor, software emulation, embedded processor, FPGA, ASIC, optical technology or other optimal technology without changing the means of the present invention. Those skilled in the art will appreciate that it can be implemented as:

[Memory garbage cleanup and collection]
“Garbage collection” refers to the recycling of unused pieces of memory. In this process, fragments are found, objects in the surrounding memory are moved up and down, and unused fragments are merged into one larger block. The fused block is stored for later reuse.

  “Cleanup” is a catch-all phrase that is used to include everything that needs to be done to the memory to optimize it. As will be mentioned later, it can change the size of specific areas of memory to optimize usage, recover previously reserved space, and use it better elsewhere Used to make

  Memory garbage collection and cleanup processes typically involve the movement of information in memory, and indexes and pointers are also updated appropriately to properly reflect that movement.

[Expand related linked blocks]
When one neuron is initially assigned and given an ID by the analyzer 30, an empty area for the relation 1252 is reserved after the basic neuron information block 1251. Please refer to FIG. 9 and FIG. When a new relationship is formed, a relationship record 1253 is added to the end of the linked list. Eventually, this free space is exhausted and there is no room to add a relationship between the end of the current connected block and the start of the next neuron. To solve this, obviously something has to be done.

["Sleep time" cleanup activities]
“Sleep” is used to remove clutter from short-term memory, half-hearted pieces of thought, guesses, and certain other information elements. This process makes it possible to start the next day as fresh as a human. It is a low risk time suitable for performing memory optimization. During the “sleep” time, the inactive state of the brain emulator can be conveniently used to handle the transfer of validated facts from enhanced memory to long-term memory. This process leaves unused holes in the enhanced memory 11, but it is also cleaned up.

  During the reassignment of neurons in long-term memory or when moving a relationship from enhancement memory 11 to an associated neuron in long-term memory 12, there may be no space left for that relationship. For this reason, the space of neurons within the long term 12 often has to be expanded.

  For this reason, the reinforcement memory 11 is scanned to determine which neuron is suitable for transfer. If transfer is hindered due to lack of space, the size of the associated long-term neuron memory record 1251 is increased.

  When available enhancement or long-term memory is reduced below a threshold, it can also be reduced during "sleep" time to optimize neuron space. From a neuron 1251 with a large free area behind it, a certain amount of that area can be recovered. Heuristics determines whether to reduce the size. If a neuron is sparse in the memory, it always operates at high speed, so collection is performed only when necessary.

[Incoming information 93]
An implementation of “submission” between the two modeled individuals is performed in the analyzer 30. The current individual position, modeled in the hierarchy of personal, political or institutional structures, is also maintained in the parameter 23.

  All information except the information from the analyzer / correlator 30 first passes through the clutter filter 40, where the information is simply ignored and may be discarded. The clutter filter 40 also uses the personality-specific parameter 22 to determine whether the synthesized personality is interested in processing pre-classified information. For example, bile quality may completely ignore human interest information, while bloody blood takes it up and devours it.

  The filter 40 is an area that is used in a catch-all manner to make preliminary decisions about the data, including the determination of its source. The filter is controlled by a number of dynamically changing parameters, including the current state of patience. When the context pool 10 is full, the filter 40 discards information, just as if someone was in a “mental overload” state.

[Preemptive training]
The brain emulation of the present invention learns over time and is influenced by the underlying temperament. Standard human learning processes are used by the emulated brain. If the memory has not been intensified for about 21 days, nothing is retained in the permanent memory 12 by the analyzer 30, avoiding the accumulation of “valueless” facts and relationships. The facts that are learned are usually interpreted under the influence of the underlying temperament, which has its implicit filter and analysis process (or limited analysis process, as in the case of multiple blood).

  Brain emulation can in fact be “trained” by methods that preempt normal temperament and temporal processes in order to quickly absorb control and environmental conditions. Therefore, the process is referred to herein as “preemptive training”. In this case, it is assumed that the “facts” and relationships given are true and pre-determined to be based on facts, so to speak “given by God”.

  Preemptive training can be switched on or off as desired from outside the emulator. To achieve rapid training of these naive facts and relationships, preemptive training can be turned on to bypass the temperament-related decision-making steps and levels of analyzer 30 and clutter filter 40. This training mode provides access to state parameters and controls that would otherwise not be allowed. These can be turned back on when training is complete. Thereafter, the changed parameters immediately affect the personality.

  In the preemptive training ("setting") mode, the entire contents of one or all memories and selected or all state parameters can be copied to an external storage device. This has applications for commercial marketing of information such as “Intellectual Property Rights” and for military purposes as described elsewhere. Such “presence snapshots” are replicated elsewhere and can be used as a basis for further training.

[Facts and relationships]
Under preemptive training, new facts and preliminary relations between the facts are defined using a plain monologue in the text file, or using spoken language if the speech analyzer 60 is present. The These are written in English prose. The grammar is interpreted by an English parser, but is not filtered by the analyzer 30 or estimator 70, or further interpreted. The normal process for grammatical interpretation continues, but the information is not subject to further temperament-based interpretation or filtering. According to this approach, brain emulation queries the trainer for information that is not obvious or not understood, so that the training process is similar to that of a human craving for knowledge.

[Religious Faith and Personal Beliefs]
Religious beliefs and personal beliefs can be established through preemptive training. As with all preemptive training, brain emulation will never know how to have these faiths or beliefs. Even so, deep (extended and consistent) regular training can overturn their progress and beliefs.

  Faith is set by a prose description in a text file and read by brain emulation. If the brain emulation doesn't understand something, or thinks something is illogical, the brain emulation will require explanation by the trainer. You can then change the prose and exclude the question from now on.

  There is nothing fundamentally different from other types of facts 1251 and relationships 1252 that can be learned regarding religious beliefs and personal beliefs. However, by defining them under preemptive training, the process of analytically examining the coherence and factual basis that would otherwise be performed by analyzer 30 is bypassed, and they can be used to understand the emulated brain understanding. It is an integrated part that serves as a reference. Religious beliefs or personal beliefs can be established and trained (rather than preemptive) over time.

[Specify control parameter value]
Many control parameters 23 and their default values can also be preset by preemptive training. This can also include certain emotional responses that will be evoked when defined conditions are met. As a result, again, brain emulation does not know why it reacts, but simply reacts that way. This is useful for pre-setting human likes and dislikes for certain things in order to accurately emulate people. Preemptive training is a method for specifying the temperament of brain emulation, including the basic temperament type and the upper complex temperament. These settings will directly affect the response and decision-making results made by this emulation.

  The period during which brain emulation learning is strengthened is usually 21 days, but a slightly different required period is set as a default based on the temperament. Table 9 gives some representative default enhancement periods. “Permanent” learning also occurs during times of mental stress or trauma, during which time the duration of this table decreases proportionally.

  When that time decreases (it doesn't affect preemptive training), brain emulation tends to hold little things and unimportant information. After the emulation is operational, these presets become an integral part of the reaction. They define the previous settings from the current time until changed.

  In the preemptive training mode, the memories 11, 12 and 13 and other tables can be saved in the external storage device based on the command. This includes facts 1251 and relationships 1252, and associated parameter settings 22, 20, and their default values. In short, everything that has been trained can be returned to the memory from which it originated. Those skilled in the art will appreciate that the method of saving memory and parameter states depends on the technology of the embodiments, and that changing these methods does not substantially change the system of the present disclosure.

  When modeling a specific person (eg, a foreigner for military purposes) using the brain emulation of the present invention, the memory and parameter settings of the emulation are “snapshot” to create new conditions or parameter settings. Allow simulation to be run again below. What is learned between the snapshots and the time to reload them later is lost (if it is not snapshotted as well) and may not be recaptured and reapplied.

[Stepwise submission]
The concept that plays a necessary role in human relationships is the concept of submission to someone else. Obedience is not “black and white”, it exists in stages. Usually, humans make decisions that are convenient to themselves under current conditions, without taking care of others. However, the person will pay special respect to some people, such as parents, bosses, military command systems, etc. Brain emulators emulate this implied relationship using “stepwise submission”. Referring to FIG. 13, the current submission request parameter 229 gives weight.

  A plurality of submission tables 128 can be created in the memory 12, and these tables are applied in a specific context 1283 (eg, military, political, social hierarchy, class). All submission tables are chained using concatenations such as 1284 and 1285. If the decision is in conflict with an external command as inferred from the statement in the semantic analyzer 50, the analyzer 30 scans the submission table and changes the temporary decision.

  The analyzer 30 searches the submission table to match one or more active contexts at that time, as maintained in the status parameter 23. When it finds it, the analyzer specifies parameters for rank self-identity. If the object being evaluated for submission is another person, the person's ID 200 is used instead. The relational comparator 1280 makes that determination as the submission output 1282. The decision weight 1296 is further adjusted by the current submission request 229. Thereafter, signal 1296 is used to determine if any decision should be made in the first place. In this way, the analyzer 30 submits a command from the authority that it controls, or weights the decision results if the conflicting command is only a recommendation from an external authority.

  Therefore, the submission table 128 has a realistic impact on brain emulation due to external authority. For example, when used in a military environment, if the ID of the simulation manager that manages the brain emulator (s) is placed at the top of all submission tables, the manager has real-time control over all brain emulations. Can be added.

  Preemptive training establishes a set (or sets) of hierarchical tables 128 for the relationship between this emulator and others (or other people). Using the same prose description, the “command system” and where the current brain emulation falls within is described.

  It is allowed to establish downline submission (i.e. conditions under which another emulator or person should submit to this brain emulator). It sets what the emulator expects from other emulators or people. Responses to violations of such expectations depend on the basic temperament specified for the current brain emulator and can also be defined during preemptive training.

[Embodiment of Temperament]
Specific assumptions made by any such model of human psychological function, including the model of the present invention, can help understand or simplify brain function. . When properly hypothesized, these assumptions allow a brain that is synthesized based on the model to be quickly generated and realized. Although they may be correct, wrong, or erroneous, such assumptions allow a “reference” implementation to be generated quickly. Such assumptions do not affect all means of the present invention.

  FIG. 14 shows one such assumption, a composite personality configuration. The assumption is that each person is “pre-wired” with a particular set of features at birth, which is one of four basic types well known to those skilled in the art. One. These are bile, depressive, bloody and mucous, classified and defined within the basic attributes of classical psychology.

  To these basic qualities (temperament) is added a set of experiences and training learned from the environment in which the individual lives. Birth quality as used herein is collectively defined as “basic temperament”. In the present invention, a composite personality is defined using the temperament and the aggregate of a set of experiences.

  FIG. 15 shows another assumption used by the present invention and the approximate characteristics represented by the model, ie, the four classical temperaments. 2 is replaced by the actual classic temperament name in FIGS. 15 and 16.

  FIG. 15 shows one temperament type, i.e. typical characteristics specific to each temperament as described above (schematic and not complete). FIG. 16 represents the combined personality of people, each based on one of the four underlying dispositions.

  A character consisting of a set of qualities that underlies a given base can “effort” through experience and training to intentionally incorporate the desired attributes of the other three temperaments. As a result, the composite character is further expanded. Here, when an individual is modeled, for example, the depressive quality of FIG. 16b can incorporate decisiveness or leadership, a characteristic that is better represented by bile quality.

  Another assumption made here simplifies the understanding of human behavior and this realistic brain emulator implementation. Regardless of past or present experience or training, every person has a fundamental temperament that is one and only one. Behaviors, behaviors, interests and decisions made by a person (ie, emulation) when undergoing mental or physical trauma, or under extreme pressure, are characteristic of that person's basic temperament Tend to return.

  Instead, other assumptions about the origin and development of temperament and personality can be made, and it is clear that they may be equally valid. Instead, they can be used herein as examples, however, it does not affect the present invention or its embodiments. The above assumptions provide a vehicle for illustration of the invention and provide a means for visualizing things that would otherwise be confusing.

[Brain parameter weighting]
FIG. 17 shows the bile quality parameter 202 in association with the propensity parameter 222 to be determined previously mentioned. The actual value of parameter 222 is the product sum 2421 of the current values of all four temperament control parameters, each with its own weight. The value of the applied weight 2420 is selected and fixed in the emulation, but the control quality parameter itself can be adjusted as desired.

  For one mode of operation, it is desirable that all four temperament parameters such as bile quality 202 have a value of 0 or 100% and are mutually exclusive. For other modes of operation, it is desirable to set the percentages of all four temperament parameters to a non-zero value that is 100% when summed. An example of means for doing this is shown in FIG.

For example, it may be convenient to have the sum of the percentages of the four temperament parameters “total” 100%. When using the weights 2420 given by the example of FIG. 17, the setting of the propensity parameter 222 to be determined is given by:
Determined propensity = 50% * Bile quality + 30% * Multiple blood + 15% * Depressive substance + 3% * Mucus

  By ignoring how the “pseudoneuron” disposition parameters are set, they can be treated as normal neurons in a neural network.

  A useful assumption made by the present invention is that a human being (emulated) has a basic or basic temperament that gives it a specific temperament for its behavior at birth. Experience, training and growth allow humans to acquire selected characteristics that are primarily found in one or more of the non-standard (“imprinted”) temperaments.

[Realization of trauma]
Part of the present invention is the realization of human pressure to mental pressure or physical or traumatic injury. Such a response is modeled herein, for example, as the impact of such experience, training and growth is reduced so that personality is temporarily governed by a “printed” temperament. . This is shown in FIG.

  In FIG. 18, the elements of FIG. 17 are supplemented by a selector 241, and the selector takes as its output one of its two inputs, that is, either takes one or the other as it is, or is selected by a decision control input. Take the percentage of each input like. In this case, the standard behavior and description illustrated by FIG. 4 is changed under mental or physical trauma or extreme pressure as indicated by parameter 230.

  In this case, when the selector 241 exists between the total temperament 2421 and the determined propensity parameter 222 and is traumatic, its decision-making behavior is instead “imprinted” underlying temperament 201. To be determined by. The basic temperament is pre-selected as one of the operational settings for brain emulation and is probably not changed “in lifetime”, but nothing prevents such changes.

  The trauma parameter 230 is triggered and set by detecting extreme mental pressure or trauma, or other parameters or neuronal conditions that indicate the level of physical trauma or shock, for example, trauma 230 can be linear, logarithmic or Other speeds are used to automatically decay over time until a nominal “off” (calm) state or value is reached. It is typically triggered by a change in the above conditions so that it can be triggered again if the condition persists or recurs, and decays immediately if the condition is removed. Can be designed.

  The conditions for inducing the trauma parameter 230 are not shown in FIG. 18, but are assumed to exist and consist of a product sum of the parameter and the brain node, and the trauma can be detected from the value.

[Gender handling]
The basic method of FIG. 18 extends to the difference in activity between men and women. In this case, the processing flow is supplemented by additional multiplexers such as 241 and 242 and a weighting table. These would be driven by gender parameters 209 instead of trauma 230, for example. Where appropriate in the decision-making and thinking process, these additions are incorporated to account for gender-related processing differences.

[Use in military or political simulation]
Since the present invention can accurately emulate human behavior, its brain emulation finds use in numerous military applications. When using conventional means, it is difficult to obtain accurate predictive modeling of combat unit decisions, in particular decisions stimulated by religious belief systems and warlike political ideologies. In today's asymmetrical war-state environment, the ability to predict actual battle decisions is of decisive importance. The means of the present invention provide this capability. Please refer to FIG. 19 and FIG.

  The brain emulator 311 as described so far can be configured to receive “spoken” input in the form of a text stream 93 and to output conversational output text 98. The brain emulation 3110 can be networked to a local or remote network 312 by adding a TCP / IP interface 3112 or other interface, such as in the case of a 1553 bus. It becomes a network connected brain emulation 311. It should be apparent to those skilled in the art that many variations of the interface 3112 can be implemented without modifying the system of the present disclosure.

  Here, these emulator clusters can be jointly formed to form a team. In FIG. 20, these emulators are illustrated as combat unit simulation clusters 310 that can be used to predictively model combat units. The same configuration can also be applied in an unmanned aerial vehicle (UAV) “cockpit”, for example, to emulate a conventional aircraft crew, so that each plays its role among all crew members Trained specifically. Similarly, it can be applied to unmanned underwater vehicles to autonomously make mission decisions when disconnected from the mother ship.

  When used as a combat unit simulation cluster, an operator simulation team 315 can be appointed to upload intelligence to the emulator 311 to accurately emulate important personnel in the modeled combat unit. As modeled combat units become available, new models can be updated using preemptive training.

  The emulation 311 used in the simulation cluster can use the “port” concept of the TCP / IP protocol to limit conversations within the emulation. Such specific local communication ports can be prevented from being accessed by other such clusters via a conventional Internet gateway 313. In this way, using the cluster 310, an enemy combat unit (eg, “red” combat unit), an unknown combat unit, an Allied or friendly (eg, “white” or “blue”) battle protected from each other Can emulate troops.

  As shown in FIG. 21, a plurality of clusters 310 can be interconnected to form an integrated combat unit simulation system 31. The simulation will be under the general control of the simulation director 330. Director 330 can securely access internal conversations within combat unit cluster 310 through a dedicated encryption port through which gateway 313 replicates and encrypts local bus 320. With this configuration, independent simulation teams 315 can work independently of each other while being under the scenario proposal and management of the director 330.

  The simulation director 330 can remotely take a snapshot of the memory and brain parameters of all brain emulations in the system 31. By taking such periodic snapshots, the simulation can be “rewinded” and resumed with various scenarios, intelligence information or updated personality profiles.

  The simulation team 315 is preferably composed of psychologists and people with knowledge of the personality, political organization or complex group responsible for emulating them. In accordance with the present invention, religious beliefs, moral beliefs (or lack thereof), command systems and authorities, and other relevant human information required to accurately predict and model human systems are realistic. It can be included.

  The simulation system 31 can be arranged in a local area or can be distributed around the world. The results of such a simulation will be available to actual combatants as part of the C4ISR.

[wrap up]
The present invention defines a means for emulation based on human psychology, which takes an electronic form and whose behavior is not readily distinguishable from human behavior. It incorporates temperament, emotion, sensation, personality and previous experience and accurately emulates human thinking and decision-making processes. The present invention extracts temperament, behavior and personality factors from human psychological behavior as a set of easily defined parameters. In this way, the relationship between these parameters and emotions, thoughts and decision making processes is incorporated into the brain emulation system by means of the present invention, and the emulation system is implemented in hardware (eg silicon). Can be supplemented by built-in software as an option.

It is a figure which shows the conventional neural network (prior art). It is a figure which shows a rule base system (prior art). FIG. 4 is a diagram illustrating an example of a weighted random effect indicating that the effect is to be taken in. FIG. 2 is a block diagram of brain emulation illustrating an embodiment of brain emulation. It is a figure which shows an example of the natural language grammar description which shows a language grammar sample. It is a figure which shows the example of a trace of a grammar syntax analysis which shows a parser analysis trace. It is a figure which shows the example of a relationship between neurons. It is a figure which shows organization of a neuron table and shows the general organization of a neuron memory list. It is a figure which shows the table | surface of a neuron and shows the internal organization of a neuron. It is an example of a relationship record, and is a figure which shows the content of the relationship record between neurons. It is a figure which shows an event queue and memory, and shows the organization of an event processor. It is a figure which shows the content of an event and shows the general content of event recording. It is a figure which shows a submission table and shows the example table of a submission order. It is a figure which shows the character of the temperament of a layer structure. It is a figure which shows the characteristic of a temperament. It is a figure which shows four composite air quality models. It is a figure which shows typical temperament and shows weighting of a parameter. FIG. 5 shows an embodiment of pressure or trauma. It is a figure which shows network connection type brain emulation. It is a figure which shows a battle unit simulation cluster example. It is a figure which shows an integrated battle unit simulation system example.

Claims (11)

A method for emulating human recognition in electronic form,
Receiving information in the form of natural language text input or speech input;
Parsing the received information into predetermined phrases based on a set of language rules stored for the natural language;
Determining whether the parsed phrase defines an aspect of an environment, and if so, generating an adaptive weighting factor for the natural language; Determining a weighted factor that can be used to make a weighted decision based on the natural language;
Determining whether the parsed word constitutes a question, and if so, making a decision on the question using the weighted rules; A method for emulating human recognition in electronic form.   The set of rules includes the fundamental parameters underlying human character, such as digital numbers, analog values, light intensity, mechanical position, or atomic states, electronic or chemical states, spins or phases, The method of claim 1 expressed as a technology independent value. Including a set of four temperament-specific parameters,
Each of the temperament-specific parameters represents one of four classic personality temperaments: bile, depressive, hyperemic and mucous,
The method of claim 2, wherein each of the temperament specific parameters represents, for example, 100% when summed, and can be expressed as a percentage taking a value between 0 and 100%. Personality trends are controlled by scaling and summing the temperament parameters,
The method of claim 3, each of which is represented as an additional temperament dependent parameter.   4. The temperament dependent parameters are decision thresholds, gains, “pseudorandom”, respectively, but are applied to control statistical selection, filter selection, etc., and to control behavior processes based on temperament. The method described.   Under conditions in which extreme mental or physical trauma has been detected, the output of three of the four temperaments is suppressed or reduced, so that the resulting personality is fundamentally only one main temperament. The method of claim 1, wherein the method is determined. The detection of the trauma can be established directly or by a change in condition, for example a difference in value of the condition.   The method of claim 6, wherein detection of the trauma automatically decays linearly or logarithmically unless it is retained or resumed by an event that causes a new trauma.   The detected condition of extreme mental or physical trauma can be represented by a calculated value or sum, or by other means, which can be represented by a technology independent value, a specific parameter or emulation The method of claim 6, which can be represented by a rated neuron.   4. The method of claim 3, wherein the emulation has a single thing that is "imprinted" at human birth and is supplemented by later experience and training, resulting in a composite personality. The impact of such training and experience on the emulated brain is reduced or suppressed so that the behavior returns to the “imprinted” temperament behavior upon detection of a sudden or sustained trauma.   The scaled and summed temperament parameters can be scaled, summed, suppressed or allowed by the values obtained from other brain nodes or parameters representing the current conditions of the brain emulation and supplemented thereby The method of claim 4, wherein the composite signal is formed. Thereby, using the supplemented composite signal, the decision threshold, gain, “pseudorandom” but statistical selection, filter selection, etc. are changed or controlled to further control the behavior process based on temperament Can do. The scaling weights are selected such that most of the temperament parameters are features that apply primarily to two of the four temperament types and minimally to the remaining two temperament types. ,
The influence of one of the two main temperaments is much greater than the influence of the other,
For example, a ratio of 50%, 35%, 10% and 5%,
Thereby, a given characteristic is largely shared by two of the temperaments, but one of them is dominant and only a small share by the remaining two temperaments. The method described.

Images