Device for ethical evaluation of artificial intelligence model according to user type

Abstract

An electronic device according to an embodiment of the present invention may include: a memory storing evaluation criteria related to ethics of an artificial intelligence model executing a consultation; and the processor is used for measuring the ethical degree of the result data output by the artificial intelligence model according to the evaluation standard, and specifically measuring according to the type of a user consulting the artificial intelligence model.

Description

Technical Field

[0001] The present invention relates to an ethics assessment device based on an artificial intelligence model of user type. Background Art

[0002] With the rapid development of artificial intelligence-based psychological counseling technology, AI-based psychological counseling robots or applications (Apps) are expected to provide great help to the public.

[0003] Compared to traditional counseling by human counselors, this AI-based psychological counseling technology requires users to bear significantly less treatment costs and offers greater accessibility. Therefore, AI-based psychological counseling technology can help low-income people or those who prefer not to disclose sensitive personal information to more freely utilize psychological counseling services.

[0004] However, because this AI-powered psychological counseling technology is implemented as a black box, ethical concerns may arise regarding the data generated. Therefore, further research is needed on how to implement AI models that can provide user-friendly answers and explanations during counseling, as well as methods for evaluating the ethical compliance of AI models. Summary of the Invention

[0005] Technical issues

[0006] An embodiment of the present invention aims to provide a consulting model suitable for person characteristics based on explainable artificial intelligence (XAI).

[0007] Embodiments of the present invention are intended to classify consultation objects into any one of multiple types based on explainable artificial intelligence (XAI), and evaluate the ethics of the artificial intelligence model performing the consultation according to the person type.

[0008] The purpose of the present disclosure is not limited to the purpose mentioned above, and other purposes and advantages of the present disclosure that are not mentioned can be understood by the following description and will be more clearly understood through the embodiments of the present disclosure. In addition, it is obvious that the purposes and advantages of the present disclosure can be achieved by the means listed in the claims and their combinations.

[0009] Means of solving the problem

[0010] According to an embodiment of the present invention, an electronic device may include: a memory for storing evaluation criteria related to the ethics of an artificial intelligence model that performs consultation; and a processor for determining the degree of ethics of the result data output by the artificial intelligence model based on the evaluation criteria, specifically according to the type of user who consults the artificial intelligence model.

[0011] Furthermore, the processor may include a user judgment unit, which determines various ethical scores of the user based on the artificial intelligence model and the content of the user consultation, and classifies the user by type according to the various scores.

[0012] In this case, the user type can be classified based on scores for disposition, virtue, personality, cognitive faculty, and personal environments. Disposition refers to an attribute related to consistency, virtue refers to an attribute related to morality, personality refers to an attribute related to empathy, cognitive faculty refers to an attribute related to problem-solving ability, and personal environments refers to an attribute related to social support.

[0013] Furthermore, the processor may include an ethics assessment unit configured to determine various ethics scores of the artificial intelligence model based on the content of the consultation between the artificial intelligence model and the user. The ethics scores of the artificial intelligence model may include explainability, transparency, accountability, bias, and reliability.

[0014] Furthermore, the ethics judgment unit may determine the ethics score of the artificial intelligence model based on the content of a user questionnaire executed on the results output by the artificial intelligence model, or determine the ethics score of the artificial intelligence model by analyzing the user's status changes after consulting with the artificial intelligence model.

[0015] According to various embodiments of the present invention, the control method of an electronic device may include: a step in which the electronic device performs a consultation between an artificial intelligence model and a user; a step in which the type of user is identified based on the content of the consultation; and a step in which, during the consultation, the degree of ethics of the result data output by the artificial intelligence model is determined according to pre-stored evaluation criteria, specifically according to the identified user type.

[0016] Furthermore, the computer program stored in the computer-readable recording medium according to the embodiment of the present invention may be designed to be executed by a processor of an electronic device, so that the electronic device executes the control method.

[0017] Effects of the Invention

[0018] The embodiments of the present invention can evaluate the ethics of an artificial intelligence model that provides consultation corresponding to user characteristics, thereby helping to provide a user-customized consultation model.

[0019] Furthermore, the embodiments of the present invention collect data on the user's personality characteristics by conducting a question-and-answer process based on a consultation model based on explainable artificial intelligence (XAI), thereby being able to identify the user type in an environment similar to actual consultation. Since no additional test paper is required, the resulting fatigue can be reduced.

[0020] Furthermore, the embodiments of the present invention provide consultation based on explainable artificial intelligence (XAI), and thus can provide the consultation object with easy-to-understand diagnostic results and humanized explanations of guidance matters, thereby improving consultation satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a diagram showing the configuration of an electronic device according to an embodiment of the present invention.

[0022] Figure 2 It is a diagram showing the configuration of a user determination unit according to an embodiment of the present invention.

[0023] Figure 3 This is a diagram showing the configuration of the ethics judgment unit according to the embodiment of the present invention.

[0024] Figure 4 2 is a diagram illustrating an operation of evaluating artificial intelligence model output data according to user type according to an embodiment of the present invention.

[0025] Figure 5 and Figure 6 is a diagram illustrating an artificial intelligence model evaluation operation of an electronic device according to an embodiment of the present invention.

[0026] Best Practice

[0027] According to an embodiment of the present invention, an electronic device may include: a memory for storing evaluation criteria related to the ethics of an artificial intelligence model that performs consultation; and a processor for determining the degree of ethics of the result data output by the artificial intelligence model based on the evaluation criteria, specifically according to the type of user who consults the artificial intelligence model.

[0028] Furthermore, the processor may include a user judgment unit, which determines various ethical scores of the user based on the artificial intelligence model and the content of the user consultation, and classifies the user by type according to the various scores.

[0029] In this case, the user type can be classified based on scores for disposition, virtue, personality, cognitive faculty, and personal environments. Disposition refers to an attribute related to consistency, virtue refers to an attribute related to morality, personality refers to an attribute related to empathy, cognitive faculty refers to an attribute related to problem-solving ability, and personal environments refers to an attribute related to social support.

[0030] Furthermore, the processor may include an ethics assessment unit configured to determine various ethics scores of the artificial intelligence model based on the content of the consultation between the artificial intelligence model and the user. The ethics scores of the artificial intelligence model may include explainability, transparency, accountability, bias, and reliability.

[0031] Furthermore, the ethics judgment unit may determine the ethics score of the artificial intelligence model based on the content of a user questionnaire executed on the results output by the artificial intelligence model, or determine the ethics score of the artificial intelligence model by analyzing the user's status changes after consulting with the artificial intelligence model.

[0032] According to various embodiments of the present invention, the control method of an electronic device may include: a step in which the electronic device performs a consultation between an artificial intelligence model and a user; a step in which the type of user is identified based on the content of the consultation; and a step in which, during the consultation, the degree of ethics of the result data output by the artificial intelligence model is determined according to pre-stored evaluation criteria, specifically according to the identified user type.

[0033] Furthermore, the computer program stored in the computer-readable recording medium according to the embodiment of the present invention may be designed to be executed by a processor of an electronic device, so that the electronic device executes the control method. DETAILED DESCRIPTION

[0034] The advantages and features of the present invention and methods for achieving them are made clear by the accompanying drawings and embodiments to be described in detail later. However, the present invention is not limited to the embodiments disclosed below, but can be implemented in various forms. The present embodiments are provided only to complete the disclosure of the present invention and to enable those skilled in the art to fully understand the scope of the present invention. The present invention is defined solely by the scope of the claims.

[0035] The terms used in this specification are used to illustrate the embodiments and are not intended to limit the present invention. In this specification, unless otherwise stated, the singular form also includes the plural form. "Comprises" and / or "comprising" used in the specification do not exclude the existence or addition of one or more other constituent elements other than the mentioned constituent elements. Throughout the specification, the same figure marks mean the same constituent elements, and "and / or" includes each of the mentioned constituent elements and all combinations of more than one. Although "first", "second" and the like are used to describe various constituent elements, these constituent elements are of course not limited by these terms. These terms are only used to distinguish one constituent element from another constituent element. Therefore, of course, the first constituent element mentioned below may also be the second constituent element within the technical idea of ​​the present invention.

[0036] Unless otherwise defined, all terms (including technical and scientific terms) used in this specification are used in the sense that they are commonly understood by those skilled in the art. In addition, unless otherwise specifically defined, terms defined in commonly used dictionaries are not to be idealized or over-interpreted.

[0037] The term "portion" or "module" used in this specification refers to a hardware component, such as software, a field programmable gate array (FPGA) or an application-specific integrated circuit (ASIC), and a "portion" or "module" performs a specific role. However, a "portion" or "module" is not limited to software or hardware. A "portion" or "module" can be configured to be located on an addressable storage medium, or it can also be configured to play one or more processors. Therefore, as an example, a "portion" or "module" includes components such as software components, object-oriented software components, class components and task components, as well as processes, functions, properties, procedures, subroutines, program code segments, drivers, firmware, microcodes, circuits, data, databases, data structures, tables, arrays and variables. The functions provided in the components and "portions" or "modules" can be combined into a smaller number of components and "portions" or "modules", or further separated into additional components and "portions" or "modules".

[0038] As shown in the figure, spatially relative terms such as "below", "beneath", "lower", "above" and "upper" can be used to conveniently describe the relationship between one constituent element and another constituent element. Spatially relative terms should be understood as terms that include different directions of constituent elements when in use or in action in addition to the directions shown in the drawings. For example, when the constituent elements shown in the drawings are turned over, the constituent element described as "below" or "beneath" another constituent element can be located "above" the other constituent element. Therefore, the exemplary term "below" can include both the lower direction and the upper direction. Constituent elements can also face different directions, so spatially relative terms can be interpreted according to their orientation.

[0039] The electronic device 100 of an embodiment of the present invention can store an artificial intelligence model (e.g., explainable artificial intelligence, XAI) for conducting consultations for treatment purposes and can evaluate the artificial intelligence model based on the content of the consultations conducted using it. Furthermore, terms described as AI or artificial intelligence in the following description can be considered to mean XAI (eXplainable Artificial Intelligence) or "explainable artificial intelligence."

[0040] Specific embodiments of the control operation of the electronic device 100 will be described in detail below with reference to the accompanying drawings.

[0041] Figure 1 is a diagram showing the configuration of an electronic device according to an embodiment of the present invention.

[0042] like Figure 1 As shown, the electronic device 100 according to the embodiment of the present invention may include a processor 110 , a memory 120 and a communication unit 130 .

[0043] Furthermore, the processor 110 may include a user determination unit 140 , a consultation conducting unit 150 , and an ethics determination unit 160 .

[0044] Furthermore, the memory 120 may include an artificial intelligence-based dialogue model. Although not shown in the figure, the memory 120 may also include an artificial intelligence model (eg, a type judgment model) for performing user type judgment operations in addition to the dialogue model.

[0045] As described above, the electronic device 100 may include various components, each of which will be described in more detail below.

[0046] As described above, the processor 110 of the electronic device 100 may include a user determination unit 140 , a consultation execution unit 150 , and an ethics determination unit 160 .

[0047] The user determination unit 140 may perform an overall operation of determining the user type based on character evaluation theory.

[0048] The user determination unit 140 may determine the user type based on pre-stored evaluation criteria.

[0049] In this case, the user type determination operation can be based on an explainable artificial intelligence (XAI) model to determine the user as one of multiple types. Specifically, the user type is determined based on the measured values ​​of five measurement items, including the user's disposition, virtue, personality, cognitive ability, and personal environments, measured based on question and answer data obtained from the user. In this case, the user type determined based on the measured values ​​may include avoidant, compromising, and problem-solving.

[0050] The user assessment unit 140 may request a pre-set consultation to collect user responses necessary for determining the user's type. Once sufficient consultation data is available to identify the user's type, the user assessment unit 140 may categorize the user by type based on the user responses obtained from the consultation. Furthermore, the user assessment unit 140 may categorize the user into any of the three types of personality assessment theory (e.g., avoidant, compromiser, or problem-solver).

[0051] For a more detailed description of the user determination unit 140, please refer to Figure 2 .

[0052] Figure 2 It is a diagram showing the configuration of a user determination unit according to an embodiment of the present invention.

[0053] like Figure 2 As shown, the user determination unit 140 may include a question generation unit 141 , an item measurement unit 142 , and a type identification unit 143 .

[0054] First, the question generation unit 141 can generate questions necessary for determining user type during the online consultation process between the artificial intelligence model and the user. Furthermore, according to various embodiments, the present invention can also determine user type through an additional user questionnaire in addition to the consultation process. Thus, the question generation unit 141 can generate the questions necessary to perform the user questionnaire operation required for determining user type.

[0055] According to one embodiment, the question generator 141 may specify that the questions initially presented to the user are consistent regardless of the user. Alternatively, the question generator 141 may generate the initial questions based on the user's basic personal information (e.g., gender, age, occupation, etc.). Without being limited to the method described above, the question generator 141 may use various methods to generate the initial questions presented to the user who initiates the consultation.

[0056] After generating and presenting an initial question to the user, when receiving an answer corresponding to the question from the user, the question generating unit 141 may generate and present a subsequent question corresponding to the received answer from the user.

[0057] At this time, when generating questions, the question generating unit 141 may generate questions for measuring five items (temperament, morality, personality, cognitive ability, and situation) corresponding to the character evaluation theory.

[0058] Specifically, the question generation method for measuring each of the five items is as follows.

[0059] First, the question generation unit 141 can generate questions for measuring "disposition" among the five items. "Disposition" is a latent variable used to measure the strength and brightness of innate disposition. "Disposition" is designed to measure a person's consistency through strength and activity through brightness and darkness. For example, in the present invention, the more consistently stable and cheerful a user is judged to be, the larger the value corresponding to the "disposition" can be.

[0060] Based on the characteristics of the "disposition" item, the question generation unit 141 can generate questions for assessing consistency, such as the user's willpower to stick to their own plans. Questions used to measure the "disposition" item can include questions such as "Have you consistently exercised for more than one month?" and "Are you often described as having a cheerful personality by those around you?"

[0061] Next, the question generator 141 may generate a question for measuring "virtue" among the five items. "Virtue" is a variable that measures an individual's ability to practice ethics and integrity (consistency between words and deeds) through hard work. The more ethical and honest a user is, the larger the value corresponding to "virtue" may be.

[0062] According to the characteristics of the "morality" item, the question generation unit 141 can generate questions such as "Do you usually follow moral standards, behave reasonably and honestly?", "Do you often ignore the principle of humility, focus on superficial decoration and go against reason?", "Do you usually maintain a high degree of consistency between words and deeds and trustworthiness in dealing with things?", "Are there any cases of inconsistency between words and deeds and lack of trustworthiness in dealing with things?", etc.

[0063] Next, the question generator 141 may generate questions for measuring "personality" among the five items. "Personality" is a measure designed to measure care for others, tolerance (catholicity), and social adaptation. The higher the tolerance and social adaptation of the user, the larger the value corresponding to the measured "personality" may be.

[0064] Based on the characteristics of the "personality" item, the question generation unit 141 can generate questions such as "Are you usually caring and tolerant of others?", "Do you usually lack caring and tolerance for others?", "Do you usually get along well with people around you?", "Do you usually not get along well with people around you?"

[0065] Next, the question generation unit 141 may generate a question for measuring "cognitive faculty" among the five items. "Cognitive faculty" is a variable used to measure a person's problem-solving ability. The more excellent the problem-solving ability of a user is, the larger the value corresponding to the measured "cognitive faculty" may be.

[0066] Based on the characteristics of the "cognitive ability" item, the question generation unit 141 can generate questions such as "Are you good at using your own experience and knowledge and the advice of people around you?", "Do you often encounter difficulties in the problem-solving process because you are stubbornly sticking to your own experience and knowledge and problem-solving ability?", "Do you prefer reasonable and fundamental problem-solving methods?", "Do you prefer solutions that rely on impromptu ideas rather than reasonable and fundamental problem-solving methods?"

[0067] Finally, the question generation unit 141 can generate questions for measuring "personal environments" among the five items. "Personal environments" is a variable used to measure a person's social status and level of social activity. Furthermore, "personal environments" is a variable used to measure the degree of social support from organizational members or those around them, as well as the social organization (social status and social sphere) that is valued in community activities. The greater the perceived social support or social organizational strength, the larger the value corresponding to the "personal environments" item.

[0068] Based on the characteristics of the "situation" item, the question generation unit 141 can generate questions such as "Are you generally respected and admired by members of the organization or people around you?", "Have you generally failed to gain respect and trust from members of the organization or people around you?", "Do you usually actively participate in activities to help the community and others?", "Are you usually not very actively involved in activities to help the community and others?"

[0069] Furthermore, the question generator 141 may determine whether to generate additional questions and change the subject of the additional questions based on the completion level of the item measurement determined by the item measurement unit 142. If the item measurement is not completed, the additional questions may be generated.

[0070] For example, when the item measurement unit 142 determines that all five items have been measured, it transmits a corresponding signal to the question generation unit 141, whereupon the question generation unit 141 may not generate additional subsequent questions and terminate the user type determination operation. Conversely, when the question generation unit 141 receives a signal from the item measurement unit 142 indicating that only four of the five measurement items have been completed, it may control the question generation unit 141 to change the subject of the additional question.

[0071] Furthermore, upon receiving a user's answer, the question generator 141 may quantify the specificity of the answer and determine the degree of specificity of the additional question based on the specificity value.

[0072] Methods for quantifying the specificity of the answer may include, for example, a method of measuring the weight of the answer, a method of measuring the diversity of words included in the answer, or a method of measuring the weight of the answer and the diversity of words.

[0073] When the obtained user answer does not reach the specificity value of the expected answer, the question generating unit 141 may generate an additional question with enhanced specificity.

[0074] For example, when a user receives an answer to the question "Do you usually stick to your plans?" that is less than the target value (e.g., less than 10 words) of "yes" or "no" or a less than target value (e.g., less than three types of words) of word diversity, the question generation unit 141 may present an additional question on the same topic with a higher degree of specificity, such as "Have you ever stuck to your plans for more than one month?" Subsequently, during the consultation process, if the answer received is determined to be less than the target value of specificity, the question generation unit 141 may generate an additional question with a higher degree of specificity.

[0075] The item measurement unit 142 can provide questions to the user through the question generation unit 141 and obtain the user's answers, and measure the five items (temperament, morality, personality, cognitive ability and situation) based on the character evaluation theory according to the user's answers.

[0076] According to various embodiments, the item measurement unit 142 may input the user's answer into an additional artificial intelligence model for determining the user type, and measure the values ​​of the five measurement items.

[0077] Next, let's look at the method by which the item measurement unit 142 calculates the measurement value of each of the five items using the method described above. First, the process by which the item measurement unit 142 calculates the measurement value of the "disposition" item is as follows. The item measurement unit 142 can calculate the correlation values ​​of the sub-items "consistency" and "activity" corresponding to the user's tendency and the "disposition" item by inputting the user's answers. At this time, when judging the correlation value between the "disposition" item and the user's characteristics, the artificial intelligence model can be designed so that the value of the x-coordinate increases in accordance with the degree of positivity of "consistency" (the stronger the consistency), and the value of the y-coordinate increases in accordance with the degree of "activity" (the stronger the activity, the more cheerful).

[0078] According to various embodiments, the AI ​​model can be designed to specify each axis of a chart using each of the five items, and then further record the correlation between each item and the user's characteristics on the chart, including additional axes. Furthermore, the AI ​​model can calculate and record the reliability of the user's answer on an additional axis (e.g., the z-axis). The reliability can be determined based on factors such as the speed at which the user completes their answer and the degree of consistency between their answers.

[0079] The type identification unit 143 may determine the user's type as one of three preset types based on the measurement value of each of the five items measured by the item measurement unit 142 .

[0080] At this point, the type identification unit 143 can classify the user into one of three types based on personality evaluation theory (e.g., avoidant, compromising, and problem-solving). The avoidant type has the lowest enthusiasm and willpower for problem solving, while the problem-solving type has the highest enthusiasm and willpower for problem solving and is highly confident in overcoming problems. Furthermore, the compromising type is an intermediate type, with a higher enthusiasm for problem solving than the avoidant type but lower than the problem-solving type.

[0081] Specifically, when the measurement values ​​of the five measurement items are all lower than the standard values, the type identification unit 143 identifies the user as an avoidant type; when the measurement values ​​of the disposition and cognitive faculty in the five measurement items are lower than the standard values, the user is identified as a compromising type; when the measurement values ​​of the five measurement items are all above the standard values, the user is identified as a problem-solving type.

[0082] Figure 1 The consultation performing unit 150 in the illustrated configuration of the processor 110 can perform a consultation operation on the user.

[0083] Specifically, the consultation processing unit 150 can support consultation operations between users and artificial intelligence models.

[0084] According to one embodiment, the consultation conducting unit 150 may perform a consultation operation under the support of the user determination unit 140. The consultation conducting unit 150 may support a dialogue for determining the user type in a consultation dialogue screen under the control of the user determination unit 140 before identifying the user type of a certain user.

[0085] After the user determination unit 140 completes the user type identification, the consultation unit 150 can perform consultation operations other than user type identification. For example, the consultation unit 150 can perform various psychiatric consultations, worry consultations, and the like.

[0086] Specifically, the consultation conducting unit 150 may obtain the consultation topic, psychiatric disease items, etc. requested by the user through the electronic device and determine the scope or type of consultation accordingly. For example, the consultation conducting unit 150 may obtain information about a psychiatric disease item (e.g., addiction) input by the user and accordingly determine to conduct a consultation with the user to address the addiction issue.

[0087] Afterwards, if the scope or type of the consultation is determined, the consultation conducting unit 150 may ask the user questions for consultation and obtain the content of the user's answer, or obtain the conversation content that the user initially inputs into the conversation screen.

[0088] The consultation conducting unit 150 inputs the conversation content input by the user into an explainable artificial intelligence (XAI) model, determines at least one corresponding type presented to the user, and can output a sentence for consulting with the user according to the corresponding type.

[0089] According to one embodiment, the consultation conducting unit 150 may provide human-friendly explanations (HFE) based on the result of classifying user types into at least three groups, and interact with the user.

[0090] Specifically, according to various embodiments, the consultation conducting unit 150 may output the content to be provided to the consultation object (including diagnosis results, judged user type information, and appropriate consultation sentences recommended to the object, etc.), and at the same time explain the reasons for determining the content.

[0091] The consultation execution unit 150 may categorize and explain to the user the previously explained settings and the types of operations performed when outputting a result value (e.g., a consultation sentence) corresponding to the settings. For example, when presenting Solution A to the user to overcome a specific problem, the consultation execution unit 150 may include information related to the process of outputting Solution A (e.g., the user's determined character type information).

[0092] Furthermore, the processor 110 may include an ethics judgment unit 160 .

[0093] The ethics judgment unit 160 may judge and evaluate the ethics and explanatory power of an artificial intelligence (XAI) model performing consultation based on the content of the conversation confirmed during the consultation process with the user.

[0094] For a description of the ethical judgment unit 160, please refer to Figure 3 .

[0095] Figure 3 This is a diagram showing the configuration of the ethics judgment unit according to the embodiment of the present invention.

[0096] like Figure 3 As shown, the ethics judgment unit 160 of the embodiment of the present invention may include a user evaluation acquisition unit 161 , an automatic evaluation execution unit 162 , a different type classification unit 163 and an evaluation score calculation unit 164 .

[0097] First, after confirming that the consultation of the preset components is completed, the user evaluation acquisition unit 161 may request the user to evaluate the result data (consultation content) output to the user through the artificial intelligence model during the consultation operation.

[0098] For example, the user evaluation acquisition unit 161 may request the user to evaluate the result data output by the artificial intelligence model for each user, or to evaluate the output results of the artificial intelligence model for a specific conversation interval or a specific date.

[0099] Furthermore, the user evaluation acquisition unit 161 may evaluate the ethics of the artificial intelligence model according to the request and based on the evaluation information provided by the user.

[0100] The user evaluation acquisition unit 161 requests the user to perform an evaluation, which can be specifically performed by directly presenting to the user standard items for evaluating the ethics of the artificial intelligence model and calculating the evaluation scores of each item. At this time, the standard items for evaluating the artificial intelligence model may include: Comprehensibility (items about whether it is easy for humans to understand the decisions of the AI ​​model), Fidelity (items about whether the decisions made by the AI ​​model are transparent, which can be certified by verification, testing certification (standardization), etc.), Responsibility (items about whether the AI ​​model is designed to be responsible, including responsibility identification, responsibilities of different stakeholders, etc.), Bias (items about whether the AI ​​model decides on fairness, including bias in data collection and processing, algorithm bias, accessibility, fairness) and Stability (items about whether the AI ​​model decides on reliability, including stability and availability).

[0101] Since the user evaluation acquisition unit 161 directly presents the evaluation items to the user, the user can assign scores to the evaluation items respectively. The user evaluation acquisition unit 161 can acquire the scores assigned by the user to each item as evaluation information.

[0102] Alternatively, the user evaluation acquisition unit 161 may not directly provide these evaluation items to users, but instead conduct a questionnaire survey through dialogue with users to evaluate the ethics of the AI ​​model. Furthermore, the user evaluation acquisition unit 161 may analyze the user responses collected during this process and automatically assign scores to the evaluation items based on the analysis results.

[0103] The automatic evaluation execution unit 162 can perform the evaluation operation based on the consultation content collected during the consultation process between the user and the artificial intelligence model, rather than based on the user questionnaire content. Therefore, the automatic evaluation execution unit 162 can automatically evaluate each result data output by the artificial intelligence model. However, it is not limited to this. The automatic evaluation execution unit 162 can only extract the consultation content of a specific unit in the overall consultation process based on standards such as the consultation content conducted within a standard period (for example, 1 day), the standard dialogue unit (for example, 6 dialogues), and the consultation content of different topics, and evaluate the output results of the artificial intelligence model based on this.

[0104] In various embodiments, the automated assessment execution unit 162 may extract only the content output by the AI ​​model from the conversation during the consultation and use this content to determine a score for at least one assessment item. For example, the automated assessment execution unit 162 may determine whether the content output by the AI ​​model is true and assign a score corresponding to the reliability of the assessment item.

[0105] Furthermore, the automatic evaluation execution unit 162 can analyze the user changes confirmed during the consultation process, and determine the score of at least one evaluation item based on this. For example, the automatic evaluation execution unit 162 can identify negative changes in the user during the consultation process (for example, an increase in the frequency of negative expressions, an increase in the intensity of negative expressions, a decrease in the weight of answers exceeding a preset value, etc.), positive changes (for example, an increase in the frequency of positive expressions, an increase in the intensity of positive expressions, an increase in the specificity of answers, an increase in the weight of answers exceeding a preset value, etc.), and changes in attention to a specific direction (for example, an increase in situations showing political bias, an increase in prejudice against a specific group, etc.), and determine the scores of different evaluation items corresponding to each situation. Furthermore, at this time, the automatic evaluation execution unit 162 can not only identify the overall negative or positive changes of the user after the consultation, but also identify negative or positive changes in specific areas, and determine the scores of different evaluation items accordingly.

[0106] For example, if the automatic evaluation execution unit 162 determines that the political bias in the user's speech has increased, the bias score in the evaluation item may be reduced. Alternatively, if a negative overall change in the user is confirmed, the automatic evaluation execution unit 162 may reduce the stability score in the evaluation item.

[0107] In addition, if the automatic evaluation execution unit 162 determines that the number of times the user has asked questions about the results output by the artificial intelligence model exceeds a standard value, the interpretability score can be subtracted from the basic value and assigned. For example, after the artificial intelligence model outputs an indicative or suggestive sentence, the automatic evaluation execution unit 162 can determine that the user has asked an intentional question about the content output by the artificial intelligence model when the user's questions containing words (presettable) such as "why" and "why" used to confirm the reason are confirmed to exceed the standard value. When it is determined that the intentional question exceeds the standard number of times, the automatic evaluation execution unit 162 can evaluate the interpretability of the result output by the artificial intelligence model as low. Through this evaluation method, the automatic evaluation execution unit 162 can reduce the interpretability evaluation score by a standard value (e.g., 1 star) when it is confirmed that the intentional question has been asked more than the standard number of times (e.g., 3 times) in the user's speech. In addition, if the intentional question exceeds the standard number of times, the evaluation score for the "interpretability" item can be reduced inversely proportional to the increase in the number of intentional questions. Therefore, the automatic evaluation execution unit 162 can not reflect the intentional questions before reaching the standard number of times in the interpretability evaluation.

[0108] In addition, the automatic evaluation execution unit 162 can analyze the corresponding output of the artificial intelligence model after the user inputs a problematic statement, and perform an evaluation operation thereon.

[0109] For example, when a user inputs illegal or immoral content, the automatic assessment execution unit 162 can analyze the corresponding output of the artificial intelligence model, evaluate whether to attempt to correct the problematic speech input by the user (for example, including illegal content, prohibited words (e.g., swear words), pessimistic expressions, etc.), and the extent of the correction, and accordingly assign a responsibility or reliability score to the artificial intelligence model. At this time, the automatic assessment execution unit 162 can determine whether to attempt correction by confirming whether the output is an advisory sentence that is contrary to the problematic speech input by the user, etc.

[0110] The different type classification unit 163 can classify the evaluation information secured in the user evaluation acquisition unit 161 and the automatic evaluation execution unit 162 according to the type of the user who is the consultation object. In this case, the user type refers to the user type identified by the user judgment unit 140.

[0111] According to an embodiment of the present invention, since there are three types of users classified by the user determination unit 140 , the evaluation information can be classified into three types according to the user types.

[0112] The evaluation score calculation unit 164 can calculate different user type evaluation scores of different user type artificial intelligence models based on the evaluation information classified into three types by the different type classification unit 163.

[0113] At this time, since the evaluation operation is not only performed on the artificial intelligence model itself, but can also be performed on the individual result data (for example, each solution) output by the artificial intelligence model or the consulting content of a specific unit, the evaluation score can also be calculated for each result data or the consulting content of a specific unit.

[0114] The result data of the artificial intelligence model evaluated in this way can be used to improve the artificial intelligence model in the future. Furthermore, the result data of the artificial intelligence model evaluated according to the user type can generate a consulting model specifically for the user type.

[0115] According to various embodiments, the evaluation score calculation unit 164 can determine the consulting topics and fields for which the evaluation score deviations of different user types are above the standard value, as well as the consulting topics and fields for which the evaluation score deviations of different user types are less than the standard value. Furthermore, the evaluation score calculation unit 164 can identify the result data values ​​(or corresponding types of artificial intelligence models) of the artificial intelligence models that each user type receives an excellent evaluation (e.g., above the standard score) among the consulting topics and fields for which the evaluation score deviations of different user types are above the standard value, and support the subsequent application of such values ​​in the consultation process unit 150 for consultation.

[0116] Furthermore, when consulting in consulting topics and fields where the deviation in evaluation scores for different user types exceeds the standard value, the consultation processing unit 150 can pre-identify the user type and then apply the AI ​​model (or the corresponding type of AI model result data) that received a good evaluation for the identified user type. Therefore, even if the AI ​​model's response may cause negative changes or be difficult to trust for other types of users, if the response is evaluated as causing positive changes or being reliable for this user type, the consultation processing unit 150 can control the application of the response to the consultation process with that user and output it.

[0117] Figure 4 2 is a diagram illustrating an operation of evaluating artificial intelligence model output data according to user type according to an embodiment of the present invention.

[0118] like Figure 4 As shown, the artificial intelligence (XAI) model itself that performs consultation according to an embodiment of the present invention or the result data output by the artificial intelligence model can evaluate the ethics or explanatory power of its intentions based on three types classified according to the scores of the user's five evaluation items.

[0119] For reference, Figure 1 The memory 120, among the components constituting the electronic device 100, can store the commands and algorithms required to perform the overall operations of the embodiment of the present invention. According to the embodiment of the present invention, the memory 120 can identify the user type and store the artificial intelligence model required for consulting with the user based on the identified user type.

[0120] Furthermore, the communication unit 130 may support communication operations with various user terminals for obtaining user input content. For example, the communication unit 130 may receive voice data or text data for user consultation received through additional user terminals. Furthermore, the communication unit 130 may receive artificial intelligence models directly input by users into user terminals or evaluation data of artificial intelligence model output results.

[0121] Figure 5 and Figure 6 is a diagram illustrating an artificial intelligence model evaluation operation of an electronic device according to an embodiment of the present invention.

[0122] first, Figure 5 Indicates a method of evaluating an AI model based on the content directly evaluated by the user after confirming the output of the AI ​​model. Figure 6 This method evaluates AI models based on user changes identified through user responses during consultations.

[0123] The following will provide a detailed description with reference to each figure.

[0124] The electronic device 100 according to an embodiment of the present invention is as follows Figure 5 As shown, the S510 operation of performing user consultation based on the artificial intelligence model can be performed.

[0125] Subsequently, the electronic device 100 may perform operation S520 of acquiring content output by the artificial intelligence model during consultation.

[0126] Subsequently, the electronic device 100 may perform operation S530 of requesting the user to evaluate the content output by the artificial intelligence model.

[0127] Subsequently, the electronic device 100 may perform operation S540 of determining the user type according to preset evaluation items.

[0128] Subsequently, the electronic device 100 may execute the operation S550 of obtaining evaluation information of different user types. At this time, the operation S540 of determining the user type may be executed at any time after the consultation starts through the artificial intelligence model and before the operation of generating evaluation information of different user types is completed.

[0129] on the other hand, Figure 6 Shows the sequence of operations for automatically evaluating the output of an artificial intelligence model based on consultation content, not user evaluation.

[0130] like Figure 6 As shown, after the electronic device 100 according to an embodiment of the present invention performs the S610 operation of performing user consultation based on the artificial intelligence model, it can perform the S620 operation of collecting consultation content and user response information generated according to the artificial intelligence model consultation.

[0131] Then, the electronic device 100 may perform operation S630 of evaluating user changes in the collected consultation content. In this case, the user changes may include an increase in the user's positivity or negativity towards a specific area.

[0132] Furthermore, the electronic device 100 may perform the S640 operation of determining the user type through the consultation content, etc. Figure 5 As described above, the operation sequence for determining the user type may be changed.

[0133] Subsequently, the electronic device 100 may perform operation S650 of acquiring evaluation information of different user types.

[0134] In short, an electronic device according to an embodiment of the present invention may include: a memory for storing evaluation criteria related to the ethics of an artificial intelligence model that performs consultation; and a processor for determining the degree of ethics of the result data output by the artificial intelligence model based on the evaluation criteria, specifically according to the type of user who consults the artificial intelligence model.

[0135] Furthermore, the processor may include a user judgment unit, which determines various ethical scores of the user based on the artificial intelligence model and the content of the user consultation, and classifies the user by type according to the various scores.

[0136] In this case, the user type can be classified based on scores for disposition, virtue, personality, cognitive faculty, and personal environments. Disposition refers to an attribute related to consistency, virtue refers to an attribute related to morality, personality refers to an attribute related to empathy, cognitive faculty refers to an attribute related to problem-solving ability, and personal environments refers to an attribute related to social support.

[0137] Furthermore, the processor may include an ethics assessment unit configured to determine various ethics scores of the artificial intelligence model based on the content of the consultation between the artificial intelligence model and the user. The ethics scores of the artificial intelligence model may include explainability, transparency, accountability, bias, and reliability.

[0138] Furthermore, the ethics judgment unit may determine the ethics score of the artificial intelligence model based on the content of a user questionnaire executed on the results output by the artificial intelligence model, or determine the ethics score of the artificial intelligence model by analyzing the user's status changes after consulting with the artificial intelligence model.

[0139] According to various embodiments of the present invention, the control method of an electronic device may include: a step in which the electronic device performs a consultation between an artificial intelligence model and a user; a step in which the type of user is identified based on the content of the consultation; and a step in which, during the consultation, the degree of ethics of the result data output by the artificial intelligence model is determined according to pre-stored evaluation criteria, specifically according to the identified user type.

[0140] Furthermore, the computer program stored in the computer-readable recording medium according to the embodiment of the present invention may be designed to be executed by a processor of an electronic device, so that the electronic device executes the control method.

[0141] The ethical assessment operations performed by the electronic device 100 according to various embodiments of the present invention can be applied not only to the field of psychological counseling, but also to various fields such as efficient learning, decision-making based on advanced AI (cognition, judgment, reasoning), security and reliability of AI, and industrial applications of AI.

[0142] Furthermore, according to various embodiments, in the artificial intelligence model of the electronic device 100 that performs ethical assessment, various items of XAI (Explainable Artificial Intelligence) can be designed as ethical items that affect users.

[0143] Furthermore, according to various embodiments, the electronic device 100 may support a user in determining the ethics score of an artificial intelligence model.

[0144] Furthermore, according to various embodiments, the electronic device 100 can determine the importance of ethical items in the artificial intelligence model differently based on the type of person. In other words, in the ethical assessment method of the artificial intelligence model, the ethical items to be emphasized can be determined differently based on the characteristics of the person.

[0145] According to various embodiments, when performing a consultation operation, the electronic device 100 may provide consultation based on a multi-faceted evaluation of a person, which includes not only the user's psychological characteristics but also morality, problem-solving ability, and social activities.

[0146] According to various embodiments, the electronic device 100 can identify artificial intelligence models with relatively high ethical assessment scores (with higher interpretability), and for models with higher assessment scores, compared with black box models, it can further guide the user's active participation and improve the consultation effect on the user.

[0147] The electronic device 100 according to an embodiment of the present invention may include a processor 110 , a memory 120 , and a communication unit 130 .

[0148] The memory 120 can store various programs and data required for the operation of the electronic device and can be implemented as a non-volatile memory, a volatile memory, a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD).

[0149] The communication unit 130 can communicate with an external device. In particular, the communication unit may include various communication chips, such as a Wi-Fi chip, a Bluetooth chip, a wireless communication chip, a near-field communication (NFC) chip, a low-power Bluetooth (Bluetooth Low Energy, BLE) chip, etc. At this time, the Wi-Fi chip, the Bluetooth chip, and the NFC chip communicate via LAN, Wi-Fi mode, Bluetooth mode, and NFC mode, respectively. When using a Wi-Fi chip or a Bluetooth chip, various connection information such as a wireless local area network (SSID, Service Set Identifier) ​​and a session key 10 (session key) are first sent and received. After establishing a communication connection using this information, various information can be sent and received. A wireless communication chip refers to a chip that communicates in accordance with various communication standards such as the Institute of Electrical and Electronics Engineers (IEEE), Zigbee, the third generation mobile communication technology (3rd Generation, 3G), the third generation partnership project (3GPP), and the long term evolution technology (LTE).

[0150] The processor 110 can use various programs stored in the memory to control the overall operation of the user device. The processor can be composed of RAM, ROM, a graphics processing unit, a main central processing unit (CPU), first to nth interfaces, and a bus. In this case, the RAM, ROM, graphics processing unit, main central processing unit (CPU), first to nth interfaces, etc. can be connected to each other via a bus.

[0151] The RAM is used to store an operating system (O / S) and application programs. Specifically, when the electronic device is started, the operating system may be stored in the RAM, and various application program data selected by the user may be stored in the RAM.

[0152] ROM stores commands for system startup, among other things. When a power-on command is input to supply power, the main central processing unit (CPU) copies the operating system stored in memory to RAM according to the instructions stored in ROM and boots the system by executing the operating system. Once booting is complete, the main central processing unit (CPU) copies various application programs stored in memory to RAM and executes these applications to perform various actions.

[0153] The main central processing unit (CPU) boots up by accessing the memory and using the operating system stored in the memory. In addition, the main central processing unit (CPU) uses various programs, content, data, etc. stored in the memory to perform various operations.

[0154] The first to nth interfaces are connected to the various components described above. One of the first to nth interfaces may be a network interface connected to an external device via a network.

[0155] On the other hand, further, the processor can control the artificial intelligence model. In this case, the control unit can of course include a graphics processor (e.g., GPU) for controlling the artificial intelligence model.

[0156] The processor 110 may include one or more cores (not shown), a graphics processing unit (not shown), and / or a connection path (eg, a bus) for transmitting and receiving signals with other components.

[0157] A processor according to one embodiment performs the method described in conjunction with the present invention by executing one or more instructions stored in a memory.

[0158] On the other hand, the processor 110 may further include a random access memory (RAM) (not shown) and a read-only memory (ROM) (not shown) for temporarily and / or permanently storing signals (or data) processed within the processor. Furthermore, the processor may be implemented in the form of a system on chip (SoC) including at least one of a graphics processing unit, RAM, and ROM.

[0159] The memory 120 may store a program (one or more instructions) for processing and controlling the processor 110. The program stored in the memory may be divided into a plurality of modules according to functions.

[0160] The steps of the methods or algorithms described in conjunction with the embodiments of the present invention may be implemented directly in hardware, or in a software module executed by hardware, or a combination of the two. The software module may also reside in RAM, ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, hard disk, removable disk, CD-ROM, or any other form of computer-readable recording medium known in the art to which the present invention belongs.

[0161] The constituent elements of the present invention can be implemented as programs (or applications) and stored in a medium so as to be combined with and run on a computer as hardware. The constituent elements of the present invention can be implemented by software programming or software elements, and similarly, embodiments can include various algorithms implemented by a combination of data structures, procedures, routines or other programming structures, and implemented by programming or scripting languages ​​such as C, C++, Java, assemblers, etc. Functional aspects can be implemented by algorithms executed on more than one processor.

[0162] The present invention has been described in detail above with reference to the examples. However, those skilled in the art may modify, alter, and distort the examples without departing from the scope of the present invention. In short, it is not necessary to include all the functional blocks shown in the drawings or to follow all the sequences shown in the drawings in order to achieve the intended effects of the present invention. It should be noted that even if these are not included, they may still fall within the technical scope of the present invention as described in the claims.

Claims

1. An electronic device, characterized in that: The electronic device comprises: Storage for evaluation criteria related to the ethics of AI models performing consulting; and The processor determines the degree of ethics of the result data output by the artificial intelligence model according to the evaluation criteria, specifically according to the type of user consulting the artificial intelligence model.

2. The electronic device according to claim 1, wherein: The processor includes a user judgment unit, which determines the user's scores in terms of ethics based on the artificial intelligence model and the user's consultation content, and classifies the user into types based on the scores.

3. The electronic device according to claim 2, wherein: The user types are classified based on scores of items including disposition, virtue, personality, cognitive faculty, and personal environments.

4. The electronic device according to claim 1, wherein: The processor includes an ethics judgment unit, which measures various scores of the ethics of the artificial intelligence model based on the consultation content between the artificial intelligence model and the user. The ethical aspects of the artificial intelligence models include explainability, transparency, accountability, bias and reliability.

5. The electronic device according to claim 4, wherein: The ethics judgment unit determines the ethics score of the artificial intelligence model based on the content of a user questionnaire executed on the result output by the artificial intelligence model.

6. The electronic device according to claim 4, wherein: After consulting with the artificial intelligence model, the ethics judgment unit determines the ethics score of the artificial intelligence model by analyzing the user's status changes.

7. A method for controlling an electronic device, characterized in that: The control method of the electronic device includes: The electronic device performs a consultation between the artificial intelligence model and the user; a step of identifying the type of the user based on the content of the consultation; and During the consultation, the degree of ethics of the result data output by the artificial intelligence model is determined according to pre-stored evaluation criteria, specifically according to the identified user type.

8. A computer program stored in a computer-readable recording medium, characterized in that The method is executed by a processor of an electronic device so as to cause the electronic device to perform the method according to claim 7.

Images