Gateway system, information processing method, and program

Abstract

We provide gateway systems and the like that enable effective and dynamic suppression of non-compliant activities. [Solution] In the gateway system 10, the first processing unit 110 acquires input information transmitted by the user terminal, filters the input information using a first filter to generate first information, and inputs it to the generative artificial intelligence. The second processing unit 120 acquires sequential information that the generative artificial intelligence continuously outputs as a response in predetermined transmission units, and generates second information by sequentially filtering the sequential information using a second filter different from the first filter. The update unit 150 performs filter update processing based on the conversation history or basic rule information between the user terminal and the generative artificial intelligence, which includes at least one of the first or second information.

Description

Technical Field

[0001] The present disclosure relates to a gateway system, an information processing method, and a program.

Background Art

[0002] There is an increasing number of cases where companies and local governments use generative AI (Artificial Intelligence) in their operations. Along with this, technologies have been proposed to suppress actions that violate compliance, such as violations of laws and regulations when using generative AI in business operations.

[0003] For example, the technology described in Patent Document 1 scans data in response to an activation command from a chat service by a data loss prevention service to detect personal information, generates an alert, and notifies a user group via the chat service.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, for suppressing actions that violate compliance in the business use of generative AI, more effective and dynamic countermeasures are expected.

[0006] In view of the above problems, an object of the present disclosure is to provide a gateway system and the like that can effectively and dynamically suppress actions that violate compliance.

Means for Solving the Problems

[0007] The gateway system described herein is a gateway system that uses filters set based on basic rule information. The gateway system comprises a first processing unit, a second processing unit, and an update unit. The first processing unit acquires input information transmitted by a user terminal to obtain a response from a generative artificial intelligence, generates first information by filtering the input information with a first filter, and inputs it to the generative artificial intelligence. The second processing unit acquires sequential information that the generative artificial intelligence continuously outputs as a response at predetermined transmission units, and generates second information by sequentially filtering the sequential information with a second filter different from the first filter. The update unit performs filter update processing based on the conversation history between the user terminal and the generative artificial intelligence or basic rule information that includes at least one of the first or second information.

[0008] The information processing method relating to this disclosure is an information processing method executed by a gateway system using a filter set based on basic rule information, wherein the computer performs the following steps: The computer acquires input information transmitted by a user terminal to obtain a response from a generative artificial intelligence, generates first information by filtering the input information with a first filter, and inputs it to the generative artificial intelligence. The computer acquires sequential information that the generative artificial intelligence continuously outputs as a response for each predetermined transmission unit, and generates second information by sequentially filtering the sequential information with a second filter different from the first filter. The computer performs a filter update process based on the conversation history between the user terminal and the generative artificial intelligence or basic rule information that includes at least one of the first or second information.

[0009] The program relating to this disclosure is a program to be executed on a gateway system using a filter set based on basic rule information, and causes a computer to execute the following information processing method. The program has a first processing step, a second processing step, and an update step. The first processing step acquires input information transmitted by a user terminal to obtain a response from a generative artificial intelligence, generates first information by filtering the input information with a first filter, and inputs it to the generative artificial intelligence. The second processing step acquires sequential information that the generative artificial intelligence continuously outputs as a response for each predetermined transmission unit, and generates second information by sequentially filtering the sequential information with a second filter different from the first filter. The update step performs filter update processing based on the conversation history between the user terminal and the generative artificial intelligence or basic rule information that includes at least one of the first information or the second information. [Effects of the Invention]

[0010] This disclosure provides a gateway system, information processing method, and program that can effectively and dynamically suppress non-compliant activities. [Brief explanation of the drawing]

[0011] [Figure 1] This is a block diagram of the response system according to Embodiment 1. [Figure 2] This is a block diagram of the gateway system according to Embodiment 1. [Figure 3] This is a flowchart of the information processing method according to Embodiment 1. [Figure 4] This diagram shows the processes executed by the first processing unit and the second processing unit. [Figure 5] This is a flowchart of the update process executed by the update unit. [Figure 6] This figure shows the first and second filters. [Figure 7] This is a block diagram of the gateway system according to Embodiment 2. [Figure 8] It is a diagram showing the processes performed by the first processing unit, the second processing unit, and the third processing unit. [Figure 9] It is a diagram showing the first filter, the second filter, and the third filter. [Figure 10] It is a diagram showing the update process executed by the update unit according to Embodiment 2. [Figure 11] It is a block diagram of the gateway system according to Embodiment 3. [Figure 12] It is a diagram showing an image of a user terminal. [Figure 13] It is a block diagram exemplifying the hardware configuration of a computer.

Embodiments for Carrying Out the Invention

[0012] Hereinafter, the present invention will be described through embodiments of the invention, but the invention according to the claims is not limited to the following embodiments. Also, not all of the configurations described in the embodiments are essential as means for solving the problems. For clarity of explanation, the following description and drawings have been appropriately omitted and simplified. In each drawing, the same elements are denoted by the same reference numerals, and redundant explanations are omitted as necessary.

[0013] <Embodiment 1> The answering system and the gateway system will be described with reference to FIG. 1. FIG. 1 is a block diagram of the answering system 1 according to Embodiment 1. The answering system 1 supports the operations in a predetermined organization such as a company, school, local government, or union. The answering system 1 receives a question (prompt or user prompt) from a user via the user terminal 30, queries the generated AI 20 with the received question, and supplies the answer output by the generated AI 20 to the user terminal 30. The answering system 1 mainly includes a gateway system 10, a generated AI 20, a user terminal 30, and a management terminal 40. These configurations are communicably connected via the network N1.

[0014] The gateway system 10 is a computer or a server. The gateway system 10 intervenes in the interaction between the user terminal 30 and the generative AI 20 and relays their communication. The gateway system 10 receives the prompt sent by the user terminal 30 to the generative AI 20, inspects the content of the prompt before transferring it to the generative AI 20, and performs appropriate filtering processing as needed. Also, the gateway system 10 receives the answer output by the generative AI 20 and performs appropriate filtering processing on the received answer.

[0015] The filtering processing includes the function of detecting personal information, confidential information, or descriptions that may conflict with laws, regulations, or company internal regulations among the received information. Also, the filtering processing includes masking processing, blocking processing, or processing that prompts re-entry for these detected descriptions. Alternatively, the filtering processing may include processing for formatting.

[0016] The generative AI 20 receives an arbitrary prompt and outputs an answer to the prompt. The generative AI 20 answers, for example, to the prompt input by the user from the user terminal 30. The generative AI 20 is, for example, a large language model. The generative AI 20 may be a multimodal model that receives inputs such as image information and audio signals in addition to receiving character inputs. The generative AI 20 can be paraphrased as generative artificial intelligence. The generative AI 20 may have a RAG function (RAG stands for Retrieval-Augmented Generation) that retrieves information from a predetermined database and uses it in the answer.

[0017] The user terminal 30 is a computer terminal used by a user such as a company staff in their work. The computer terminal is a desktop PC, a notebook PC, a tablet terminal, a smartphone, etc. The user, for example, uses the user terminal 30 to perform tasks such as document creation and document content checking. At this time, the user uses the generative AI 20 via the user terminal 30 in the performance of the work.

[0018] The management terminal 40 is a computer terminal used by an administrator who manages basic rule information in a company, for example. Computer terminals include desktop PCs, notebook PCs, tablet devices, smartphones, etc. The management terminal 40 supplies basic rule information to the gateway system 10, for example.

[0019] Next, the functional configuration of the gateway system 10 will be described with reference to Figure 2. Figure 2 is a block diagram of the gateway system 10 according to Embodiment 1. The gateway system 10 is a gateway system that uses filters set based on basic rule information. The gateway system 10 mainly consists of a first processing unit 110, a second processing unit 120, and an update unit 150.

[0020] The basic rule information is the information configured so that the response system 1 generates responses in accordance with predetermined rules. Specifically, for example, the basic rule information may include laws and regulations, internal company regulations, contract terms, and project policies. The basic rule information may also include pre-configured prohibited words and phrases.

[0021] The first processing unit 110 acquires input information that the user terminal 30 sends to obtain a response from the generating AI 20. The input information can be rephrased as a user prompt generated by the user. The first processing unit 110 includes a first filter 111.

[0022] The first filter 111 is configured to detect personal information, prohibited words, or content that may violate prescribed rules from the input information as attention information, based on the content of the basic rule information. The first processing unit performs processing to mask, block, or prompt re-input the attention information detected by the first filter 111 on the input information.

[0023] The first processing unit 110 generates first information by filtering the input information using the first filter 111. Furthermore, the first processing unit 110 inputs the generated first information to the generation AI 20. Upon receiving the first information, the generation AI 20 generates and outputs a response to the first information.

[0024] The second processing unit 120 acquires sequential information that the generating AI 20 outputs as a response in a predetermined transmission unit. Upon acquiring the sequential information, the second processing unit 120 generates second information by sequentially filtering the sequential information using a second filter 121 that is different from the first filter 111. The gateway system 10 supplies the generated second information to the user terminal 30.

[0025] The second filter 121 is configured to detect personal information, prohibited words, or content that may violate predetermined rules as attention information in the string of sequential information in each transmission unit, based on the content of the basic rule information. The second processing unit 120 stores the sequential information in a buffer for each transmission unit and applies the second filter 121 to the sequential information held in the buffer. As a result, the second processing unit 120 sequentially masks the detected attention information.

[0026] The second processing unit may include an inspection unit and a control unit. The inspection unit performs rule-based processing, which involves comparing sequential information held in a buffer with attention information registered in the basic rule information, and natural language processing, which involves detecting words that are highly similar to the attention information. The control unit sequentially masks the attention information detected by the inspection unit.

[0027] The update unit 150 performs filter update processing based on the conversation history between the user terminal 30 and the generating AI 20 or basic rule information, which includes at least one of the first or second pieces of information. The update unit 150 performs filter update processing when the basic rule information is updated. The update unit 150 also performs filter update processing according to the conversation history when it detects an update to the basic rule information from the conversation history. The update unit 150 may have a function to store the conversation history, or it may acquire the conversation history from an external storage device that stores the conversation history.

[0028] The conversation history includes at least the first and second pieces of information. The conversation history may also include logs of the application process for each filter. Furthermore, the conversation history may include management information such as the identifier and time information of the conversation between the user terminal 30 and the generating AI 20.

[0029] The update unit 150 may select filter update targets from feedback information regarding false positives or missed detections contained in the conversation history by performing natural language processing on the conversation history. In this way, the update unit 150 can, for example, analyze the conversation history to identify false positives and missed detections.

[0030] Furthermore, when the update unit 150 performs filter update processing based on the conversation history, it may notify the management terminal 40 of the details of the performed update processing. The update unit 150 may also notify the management terminal 40 of a message prompting the update of the basic rule information corresponding to the performed update processing. This allows the gateway system 10 to quickly update filters based on the conversation history while simultaneously notifying the administrator of information regarding the update of basic rule information.

[0031] Next, with reference to Figure 3, an overview of the processing performed by the gateway system 10 will be described. Figure 3 is a flowchart of the information processing method according to Embodiment 1. In the information processing method according to this disclosure, the gateway system 10 performs the following steps using a filter set based on basic rule information.

[0032] First, the first processing unit 110 acquires the input information that the user terminal 30 sends to obtain a response from the generating AI 20, filters the input information using the first filter to generate first information, and inputs the generated first information to the generating AI 20 (step S1).

[0033] Next, the second processing unit 120 acquires sequential information that the generating AI 20 outputs as a response in succession for each predetermined transmission unit, and generates second information by sequentially filtering the sequential information using the second filter (step S2).

[0034] Next, the update unit 150 performs a filter update process based on the conversation history or basic rule information (step S3).

[0035] The above describes the overview of the processes performed by the gateway system 10. By performing these processes, the gateway system 10 suppresses the distribution of inappropriate and misleading information. Furthermore, the gateway system 10 enables filter updates without delay by automatically updating filters based on conversation history.

[0036] Next, with reference to Figure 4, the processes performed by the first processing unit 110 and the second processing unit 120 of the gateway system 10 will be further explained. Figure 4 is a diagram showing the processes performed by the first processing unit 110 and the second processing unit 120.

[0037] First, the first processing unit 110 acquires input information from the user terminal 30 (step S11).

[0038] Next, the first processing unit 110 filters the input information using the first filter 111 to generate first information (step S12). Here, the first processing unit 110 detects, for example, personal information, confidential information, or violations of laws and regulations, and performs masking or blocking. If the first processing unit 110 blocks the input information, the user using the user terminal 30 can regenerate the content that was not blocked by the filtering process as input information.

[0039] Next, the first processing unit 110 inputs the generated first information to the generating AI 20 (step S13). This allows the generating AI 20 to receive information with low compliance risk and generate a response.

[0040] Next, the second processing unit 120 acquires sequential information for each transmission unit from the generating AI 20 (step S14). Here, "transmission unit" refers to the communication unit when the generating AI 20 sequentially generates and outputs the response. For example, a transmission unit is a block of data consisting of a sentence or phrase of a predetermined length or a predetermined number of tokens.

[0041] Next, the second processing unit 120 filters the sequential information of the transmission unit using the second filter 121 to generate second information (step S15). This process allows the second processing unit 120 to detect and correct prohibited words or incorrect information in real time.

[0042] Next, the second processing unit 120 outputs the generated second information to the user terminal 30 (step S16). In other words, the second processing unit 120 suppresses compliance risks in real time.

[0043] Next, the second processing unit 120 determines whether or not to acquire the next sequential information (step S17). The second processing unit 120 may determine whether or not to acquire the next sequential information from, for example, the context of the response output by the generating AI 20. The second processing unit 120 may also determine not to acquire the next sequential information if it does not receive the subsequent sequential information within a predetermined period using a timer. If the generating AI 20 outputs a header and footer before and after the response, the second processing unit 120 may determine not to acquire the next sequential information as a trigger upon receiving the footer.

[0044] If the second processing unit 120 determines that it will acquire the next sequential information (step S17: YES), the second processing unit 120 returns to step S14. On the other hand, if it does not determine that it will acquire the next sequential information (step S17: NO), the second processing unit 120 terminates the series of processes.

[0045] Next, with reference to Figure 5, the processes executed by the update unit 150 will be described. Figure 5 is a flowchart of the update process executed by the update unit. The update unit 150 executes the following update processes, for example, at predetermined intervals. A predetermined interval is, for example, one day, one week, one month, etc.

[0046] First, the update unit 150 determines whether or not to acquire new conversation history (step S101). Here, new conversation history refers to a record of a new conversation that occurred between the user terminal 30 and the generating AI 20. The conversation history includes at least the first information and the second information. The conversation history may also include the date and time the conversation took place or an identifier to identify the conversation. The update unit 150 reads the conversation history added since the last update from the storage unit or an external server. If it is determined that new conversation history has been acquired (step S101: NO), the update unit 150 terminates the update process. On the other hand, if it is determined that new conversation history has been acquired (step S101: YES), the update unit 150 proceeds to step S102.

[0047] In step S102, the update unit 150 reads the updated conversation history and determines whether or not there is feedback information (step S102). Here, feedback information is evaluation information that the user provides to the generated AI 20's response content and the filter processing results. More specifically, the feedback information includes evaluations related to basic rule information. Evaluations related to basic rule information include, for example, natural language expressed by the user regarding the validity of mask processing, attention information, and re-input requests. Evaluations related to basic rule information may also be flags or signals indicating "good" or "bad".

[0048] If it is determined that there is no feedback information (step S102: NO), the update unit 150 terminates the update process. On the other hand, if it is determined that there is feedback information (step S102: YES), the update unit 150 proceeds to step S103.

[0049] In step S103, the update unit 150 selects the update target and performs an update process on the selected update target (step S103). Here, the update target is at least a part of the first filter 111 and the second filter 121. The update unit 150 uses a language model to detect the content of the first filter 111 or the second filter 121 to be updated from the feedback information. The update unit 150 also uses a language model to perform a rewrite process on the first filter 111 and the second filter 121. As a result, the gateway system 10 can improve the accuracy of the filters.

[0050] Next, the first filter 111 and the second filter 121 will be described with reference to Figure 6. Figure 6 is a diagram showing the first and second filters. Figure 6 shows the relationship between the "rule number," "content," and "applicable rule" of the first filter 111 and the second filter 121. The first and second filters shown in Figure 6 are used, for example, by users who perform tasks related to contracts with temporary staff.

[0051] The first filter 111 is indicated by rule numbers 1-1, 1-2, and 1-3. Rule 1-1 is a rule that applies the process of "masking the phone number as [phone number]" when the input information includes a phone number. Similarly, Rule 1-2 is a rule that applies the process of "masking the address as [address]" when the input information includes an address. Rule 1-3 is a rule that applies the process of displaying "the dispatch period exceeds the limit" and blocking input when the input information includes a portion where the dispatch period of a dispatched employee is stated to be longer than 3 years.

[0052] The second filter 121 is indicated by rule numbers 2-1 and 2-2. Rule 2-1 is a rule that applies a process to mask strings that are presumed to be proper nouns when the content of the sequential information is included with ●. Rule 2-2 is a rule that applies a process to replace specific prohibited words with [prohibited word] when the content of the sequential information is included.

[0053] As described above, the first filter 111 reads the content of the input information and has a defined process that corresponds to the content of the input information read. The second filter 121 also reads the content of the sequential information and has a defined process that corresponds to the content of the sequential information read. As a result, the gateway system 10 prevents information leakage or input that violates predetermined rules, and secures the output generated by the generating AI 20 in real time before supplying it to the user terminal 30. In other words, the gateway system 10 achieves, for example, the protection of sensitive information and compliance with laws and internal regulations.

[0054] Embodiment 1 has been described above. The response system 1 may have a separate system for storing or managing basic rule information, in addition to the management terminal 40 and the gateway system 10. Naturally, the response system 1 may also have multiple user terminals 30 and management terminals 40.

[0055] The response system 1 may have a database (not shown) that stores basic rule information. In that case, the generating AI 20 can access the basic rule information in the database using the RAG function. Alternatively, the gateway system 10 may have its own storage unit that stores basic rule information. In that case, the management terminal 40 updates the basic rule information stored by the gateway system 10.

[0056] The gateway system 10 may also have a processor and a storage device, although these are not shown in the diagram. In this case, the storage device may include non-volatile memory such as flash memory or an SSD (Solid State Drive). In this case, the storage device may store a computer program (hereinafter simply referred to as a program) for executing the above-described method. The processor loads the computer program from the storage device into a buffer memory such as DRAM (Dynamic Random Access Memory) and executes the program.

[0057] Each component of the gateway system 10 may be implemented with dedicated hardware. Furthermore, some or all of each component may be implemented by general-purpose or dedicated circuits, processors, etc., or combinations thereof. These may be implemented by a single chip or by multiple chips connected via a bus. Some or all of each component may be implemented by a combination of the aforementioned circuits, etc., and programs. Additionally, a CPU (Central Processing Unit), GPU (Graphics Processing Unit), FPGA (field-programmable gate array), etc., can be used as the processor. Furthermore, at least some of the functions of this embodiment may be provided in the form of IaaS (Infrastructure as a Service), PaaS (Platform as a Service), or SaaS (Software as a Service).

[0058] As described above, this embodiment provides a gateway system, information processing method, and program that can effectively and dynamically suppress acts that violate compliance.

[0059] <Embodiment 2> Next, Embodiment 2 will be described. Figure 7 is a block diagram of the gateway system according to Embodiment 2. The gateway system 10 according to this embodiment has a third processing unit 130. The gateway system 10 according to this embodiment also has a storage unit 200. Furthermore, the configuration of the update unit 150 in the gateway system 10 according to this embodiment differs from that of the update unit 150 described above.

[0060] The third processing unit 130 acquires the second information, filters the second information using a third filter 131 which is different from the second filter 121, generates third information, and outputs the third information to the user terminal 30 as a response.

[0061] The third filter 131 is configured to detect content that may violate predetermined rules in the entire text output by the generating AI 20. That is, the third processing unit 130 performs restorative substitution if the response contains content that may violate predetermined rules. Restorative substitution is, for example, the process of rephrasing a concrete expression into an abstract expression. Alternatively, restorative substitution is, for example, the process of rephrasing a relatively strong descriptive expression into a weaker expression.

[0062] The update unit 150 in this embodiment performs filter update processing based on conversation history or basic rule information that includes at least one of the first information, second information, and third information.

[0063] The update unit 150 according to this embodiment includes an approval request unit 151 and an update execution unit 152. The approval request unit 151 requests approval or rejection of the update process for feedback information from a management terminal used by an administrator managing the gateway system 10. If the management terminal approves the update process, the update execution unit 152 reflects the feedback information to the update target. If the management terminal rejects the update process, the update execution unit 152 suppresses the reflection of the feedback information to the update target.

[0064] As a result, the gateway system 10 in this embodiment applies filter updates after a human approval process. Therefore, the gateway system 10 can suppress the incorporation of incorrect learning. Furthermore, the gateway system 10 can suppress filter updates that are not intended by the administrator.

[0065] The storage unit 200 includes non-volatile memory such as flash memory. The storage unit 200 in this embodiment stores at least basic rule information 201 and conversation history 202. The storage unit 200 may also store filter update history.

[0066] Next, referring to Figure 8, the processes performed by the first processing unit 110, the second processing unit 120, and the third processing unit 130 in the gateway system 10 according to this embodiment will be described. Figure 8 is a diagram showing the processes performed by the first processing unit, the second processing unit, and the third processing unit. The flowchart shown in Figure 8 differs from the flowchart shown in Figure 4 in that it has steps S21 and S22 after step S17.

[0067] After step S17, in step S21, the third processing unit 130 filters the second information using the third filter 131 to generate third information (step S21). Through this process, the third processing unit 130 detects areas for improvement when viewing the entire response generated by the generating AI 20. In other words, the third processing unit 130 performs filtering on the entire response, which cannot be done by the second processing unit 120, which filters sequential information sequentially.

[0068] Next, the third processing unit 130 outputs the generated third information to the user terminal 30 (step S22). That is, the third processing unit 130 updates the second information that the second processing unit 120 has already output to the user terminal 30 and presents the third information. In this way, the third processing unit 130 aims to improve the readability of the response supplied to the user terminal 30 while suppressing compliance issues.

[0069] The gateway system 10 may not output the second information to the user terminal, but may instead hold it in a buffer. In that case, the gateway system 10 may output the generated third information to the user terminal 30 as a response after the third processing unit 130 has generated the third information.

[0070] Next, each filter according to this embodiment will be described with reference to Figure 9. Figure 9 shows the first filter 111, the second filter 121, and the third filter 131. The first filter 111 and the second filter 121 shown in Figure 9 are the same as those shown in Figure 6.

[0071] The third filter 131 is indicated by rule numbers 3-1 and 3-2. Rule 3-1 is a rule that applies a process to replace expressions that could potentially identify an individual with abstract expressions when the content of the second information is included. Rule 3-2 is a rule that applies a process to generate alternative expressions when the content of the second information is included in sentences that are difficult to understand due to a large masked area.

[0072] To perform processing in accordance with these rules, the third processing unit 130 uses a language model. That is, the third processing unit 130 inputs the second information into the language model, along with the third filter 131. The third processing unit 130 also inputs a system prompt into the language model instructing it to rewrite the second information according to the third filter 131.

[0073] Next, with reference to Figure 10, the update process executed by the update unit 150 according to this embodiment will be described. Figure 10 is a diagram showing the update process executed by the update unit 150 according to Embodiment 2. The flowchart shown in Figure 10 differs from the flowchart shown in Figure 5 in that it has steps S111 to S115 instead of step S103.

[0074] After the update unit 150 determines in step S102 that there is feedback information (step S102: YES), the update unit 150 proceeds to step S111.

[0075] In step S111, the update unit 150 selects the update target and generates update information (step S111). The update information is information that reflects the feedback information to the update target of the filter currently in operation. The update unit 150 generates the update information using a language model.

[0076] Next, the approval request unit 151 requests approval for the update process from the management terminal 40 (step S112). Specifically, the approval request unit 151 presents the update target and update information to the management terminal 40. Furthermore, the approval request unit 151 requests the management terminal 40 to decide whether to execute the update process or not.

[0077] Next, the approval request unit 151 determines whether the management terminal 40 has approved the update process (step S113). If it determines that the management terminal 40 has approved the update process (step S113: YES), the update unit 150 proceeds to step S114. If it does not determine that the management terminal 40 has approved the update process (step S113: NO), the update unit 150 proceeds to step S115.

[0078] In step S114, the update execution unit 152 reflects the update information to the update target (step S114). Once the update execution unit 152 reflects the update information, the update unit 150 terminates the series of update processes.

[0079] In step S115, the update execution unit 152 suppresses the reflection of the update information (step S115). In other words, in this case, the update unit 150 does not apply the generated update information to the filter. Then the update unit 150 terminates the series of processes.

[0080] The update process performed by the update unit 150 according to this embodiment has been described above. Through the above process, the update unit 150 reflects or suppresses the update information based on the feedback information with the approval of the administrator. As a result, the gateway system 10 can promptly propose the update information to the administrator and suppress unintended modifications to the filter.

[0081] Embodiment 2 has been described above. In this embodiment, the update unit 150 may perform a process to prompt the administrator to update the basic rule information when updating the filter. This enables the gateway system 10 to quickly and appropriately update the basic rules and filters. Furthermore, the gateway system 10 can suppress erroneous updates of the basic rule information by going through the administrator's approval process.

[0082] As described above, this embodiment provides a gateway system, information processing method, and program that effectively and dynamically suppress non-compliant actions while suitably updating filter or basic rule information.

[0083] <Embodiment 3> Next, Embodiment 3 will be described. Figure 11 is a block diagram of the gateway system 10 according to Embodiment 3. The gateway system 10 according to this embodiment has a notification unit 300.

[0084] The notification unit 300 notifies the user, who is the user of the user terminal 30, that the first filter 111 or the second filter 121 has detected information of interest, or that the third processing unit 130 has performed a restorative replacement.

[0085] In the gateway system 10 according to this embodiment, the first filter 111 is configured to detect personal information, prohibited words, or content that may violate predetermined rules from the input information as information of interest, based on the content of the basic rule information.

[0086] Similarly, the second filter 121 is configured to detect personal information, prohibited words, or content that may violate prescribed rules as attention information for strings of sequential information in transmission units, based on the content of the basic rule information.

[0087] Furthermore, the third filter 131 is configured to detect content that may violate predetermined rules in the entire text, and the third processing unit 130 performs corrective replacement if the answer contains content that may violate predetermined rules.

[0088] Next, the notification unit 300 will be further explained with reference to Figure 12. Figure 12 is a diagram showing image 301 on the user terminal 30. Image 301 shown in Figure 12 is displayed on the display of the user terminal 30 when the response system 1 is functioning.

[0089] Image 301 mainly consists of a question input field 302, a question display field 303, an answer display field 304, and a notification field 306.

[0090] The question input field 302 is a field that displays the input information entered by the user. In the question input field 302 shown in Figure 12, the words "Enter your question:" are displayed. The question input field 302 prompts the user to enter a question, i.e., input information, through this display.

[0091] The question display field 303 is a field that displays the input information that has been sent by the user to the generation AI 20. In the question display field 303 shown in Figure 12, the message "Question: Please create a contract under the following conditions..." is displayed. This indicates that the user has instructed the generation AI 20 to create a contract. The gateway system 10 filters this user-created input information using the first filter 111.

[0092] The answer display field 304 is the answer generated by the generation AI 20 to the user's question (input information) displayed in the question display field 303. The answer display field 304 shown in Figure 12 shows the contents of the contract generated by the generation AI 20 following "Answer:". Here, the answer display field 304 includes a masked section 305. The masked section 305 is the part that has been masked by the gateway system 10 using one of the first filter 111, the second filter 121, or the third filter 131.

[0093] The notification section 306 displays "Notice:" along with the message "The red portion has been masked." This message was generated by the notification unit 300. In this case, the masked portion 305 of the response display section 304 is displayed in red.

[0094] The notification area 306 also includes a details button 307. The details button 307 has the words "View Details" displayed inside a rectangle. When the user selects the details button 307, the response system 1 displays the details of the filtering process performed by the gateway system 10 on the user terminal 30.

[0095] Through the functions described above, the notification unit 300 makes the user aware of whether a filter is applied and the areas to be filtered. Thus, the gateway system 10 provides answers that meet the required compliance requirements and ensures transparency and traceability of information regarding the input and output of the generating AI 20.

[0096] Although Embodiment 3 has been described above, Embodiment 3 is not limited to the above configuration. For example, the notification unit 300 may notify the user that attention information has been detected and that restorative replacement has been performed, etc., through sound or vibration.

[0097] As described above, this embodiment provides a gateway system, information processing method, and program that effectively and dynamically suppress non-compliant activities while ensuring transparency of information.

[0098] <Example hardware configuration> The following describes examples of how each functional configuration of the information processing system described in this disclosure can be realized through a combination of hardware and software.

[0099] Figure 13 is a block diagram illustrating the hardware configuration of a computer. The gateway system 10 in this disclosure can achieve the above-described functions with a computer 500 including the hardware configuration shown in the figure.

[0100] Computer 500 may be a portable computer such as a smartphone or tablet, or a stationary computer such as a PC. Computer 500 may be a dedicated computer designed to implement each device, or it may be a general-purpose computer. Computer 500 can achieve the desired functions by installing a predetermined application.

[0101] Computer 500 includes a bus 501, a processor 510, memory 511, a storage device 512, an input / output interface 513, and a network interface 514. The bus 501 is a data transmission path for the processor 510, memory 511, storage device 512, input / output interface 513, and network interface 514 to send and receive data from each other. However, the method of connecting the processor 510 and other components to each other is not limited to bus connection.

[0102] Processor 510 is a variety of processors such as a CPU, GPU, or FPGA. Memory 511 is main memory implemented using RAM (Random Access Memory), etc.

[0103] The storage device 512 is an auxiliary storage device implemented using a hard disk, SSD, memory card, or ROM (Read Only Memory). The storage device 512 stores a program for realizing a desired function. The processor 510 reads this program into memory 511 and executes it to realize each functional component of each device.

[0104] The input / output interface 513 is an interface for connecting the computer 500 to input / output devices. For example, input devices such as keyboards and output devices such as display devices are connected to the input / output interface 513. The network interface 514 is an interface for connecting the computer 500 to a network.

[0105] Although the present disclosure has been described above with reference to embodiments, the present disclosure is not limited to the embodiments described above. Various modifications to the structure and details of the present disclosure can be made as can be understood by those skilled in the art within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate.

[0106] Each drawing is merely illustrative to illustrate one or more embodiments. Each drawing may be associated with one or more other embodiments rather than with only one specific embodiment. As those skilled in the art will understand, various features or steps described with reference to any one drawing can be combined with features or steps shown in one or more other drawings, for example, to create embodiments not explicitly shown or described. Not all features or steps shown in any one drawing to illustrate an exemplary embodiment are necessarily required, and some features or steps may be omitted. The order of steps shown in any of the drawings may be changed as appropriate. [Explanation of Symbols]

[0107] 1. Answer System 10 Gateway System 20 Generation AI 30 User terminals 40 Management terminals 110 First Processing Unit 111 First Filter 120 Second Processing Unit 121 Second filter 130 Third Processing Unit 131 Third Filter 150 Update section 151 Approval request section 152 Update Execution Unit 200 Storage section 201 Basic Rules Information 202 Conversation History 300 Notification Department 301 images 302 Question Input Field 303 Question display field 304 Answer display field 305 Mask section 306 Notification column 307 Details button 500 Computers 501 Bus 510 Processor 511 memory 512 Storage Devices 513 Input / Output Interfaces 514 Network Interfaces N1 Network

Claims

1. A gateway system that uses filters configured based on basic rule information, A first processing unit acquires input information that a user terminal transmits to obtain a response from a generative artificial intelligence, generates first information by filtering the input information using a first filter, and inputs it to the generative artificial intelligence. A second processing unit acquires sequential information that the generative artificial intelligence, upon receiving the first information, continuously outputs as the answer, at predetermined transmission units, and generates second information by sequentially filtering the sequential information with a second filter different from the first filter. An update unit that performs the filter update process based on the conversation history between the user terminal and the generative artificial intelligence or the basic rule information, which includes at least one of the first information or the second information, The system includes a third processing unit that acquires the second information, filters the second information using a third filter different from the second filter to generate third information, and outputs the third information to the user terminal as the response, The update unit performs the filter update process based on the conversation history or basic rule information which includes at least one of the first information, the second information, and the third information. The third filter is configured to detect content that may violate predetermined rules in the entire text. The third processing unit performs a restorative substitution if the response contains content that may violate a predetermined rule. Gateway system.

2. The first filter is configured to detect personal information, prohibited words, or content that may violate prescribed rules as attention information from the input information, based on the content of the basic rule information. The second filter is configured to detect, based on the content of the basic rule information, personal information, prohibited words, or content that may violate prescribed rules in the string of sequential information in the transmission unit, as the attention information. The system further includes a notification unit that notifies the user, who is the user of the user terminal, that the first filter or the second filter has detected the information of interest, or that the third processing unit has performed the restorative replacement. The gateway system according to claim 1.

3. The first filter is configured to detect personal information, prohibited words, or content that may violate prescribed rules as attention information from the input information, based on the content of the basic rule information. The first processing unit performs a masking process, a blocking process, or a process to prompt re-input of the information of interest detected by applying the first filter to the input information. The gateway system according to claim 1.

4. The second filter is configured to detect, based on the content of the basic rule information, personal information, prohibited words, or content that may violate prescribed rules as attention information for the string of sequential information in the transmission unit. The second processing unit holds the sequential information in a buffer for each transmission unit, and sequentially masks the attention information detected by applying the second filter to the sequential information held in the buffer. The gateway system according to claim 1.

5. The second processing unit is, An inspection unit performs rule-based processing that compares the sequential information held in the buffer with the attention information registered in the basic rule information, and detects words with a high similarity to the attention information using natural language processing. The system includes a control unit that sequentially masks the information of interest detected by the inspection unit, The gateway system according to claim 4.

6. The update unit performs natural language processing on the conversation history to select the target for updating the filter from the feedback information regarding false detections or missed detections contained in the conversation history. The aforementioned update unit is, An approval request unit that requests the management terminal to approve or disapprove the update process for the aforementioned feedback information, The update execution unit includes: if the management terminal approves the update process, it reflects the feedback information to the update target; and if the management terminal disapproves the update process, it suppresses the reflection of the feedback information to the update target. The gateway system according to claim 1.

7. An information processing method performed by a gateway system using filters set based on basic rule information, The system acquires input information that the user terminal sends to obtain a response from the generative artificial intelligence, generates first information by filtering the input information using a first filter, and inputs it to the generative artificial intelligence. The generative artificial intelligence that receives the first information continuously outputs sequential information as the answer, and acquires this sequential information for each predetermined transmission unit. The second information is generated by sequentially filtering the sequential information using a second filter different from the first filter. Based on the conversation history between the user terminal and the generative artificial intelligence, or the basic rule information, which includes at least one of the first or second pieces of information, the filter update process is performed. The second information is acquired, the second information is filtered using a third filter different from the second filter to generate a third information, and the third information is output to the user terminal as the answer. Based on the conversation history or basic rule information which includes at least one of the first, second, and third pieces of information, the filter update process is performed. The third filter is configured to detect content that may violate predetermined rules in the entire text. If the aforementioned response contains content that may violate the prescribed rules, a restorative substitution will be performed. Information processing methods.

8. A program to be executed by a gateway system using filters configured based on basic rule information, A first processing step involves acquiring input information that a user terminal transmits to obtain a response from a generative artificial intelligence, generating first information by filtering the input information using a first filter, and inputting this information to the generative artificial intelligence. A second processing step involves acquiring sequential information that the generative artificial intelligence, upon receiving the first information, continuously outputs as the response, for each predetermined transmission unit, and generating second information by sequentially filtering the sequential information with a second filter different from the first filter; An update step in which the filter update process is performed based on the conversation history between the user terminal and the generative artificial intelligence or the basic rule information, which includes at least one of the first information or the second information; The system includes a third processing step of acquiring the second information, generating third information by filtering the second information with a third filter different from the second filter, and outputting the third information to the user terminal as the response, The update step performs the filter update process based on the conversation history or basic rule information which includes at least one of the first information, the second information, and the third information. The third filter is configured to detect content that may violate predetermined rules in the entire text. The third processing step performs a restorative substitution if the response contains content that may violate a predetermined rule. To have a computer execute an information processing method. program.

9. A gateway system that uses filters configured based on basic rule information, A first processing unit acquires input information that a user terminal transmits to obtain a response from a generative artificial intelligence, generates first information by filtering the input information using a first filter, and inputs it to the generative artificial intelligence. A second processing unit acquires sequential information that the generative artificial intelligence, upon receiving the first information, continuously outputs as the answer, at predetermined transmission units, and generates second information by sequentially filtering the sequential information with a second filter different from the first filter. The system includes an update unit that performs the filter update process based on the conversation history between the user terminal and the generative artificial intelligence or the basic rule information, which includes at least one of the first information or the second information, The update unit performs natural language processing on the conversation history to select the target for updating the filter from the feedback information regarding false detections or missed detections contained in the conversation history. The aforementioned update unit is, An approval request unit that requests the management terminal to approve or disapprove the update process for the aforementioned feedback information, The update execution unit includes: if the management terminal approves the update process, it reflects the feedback information to the update target; and if the management terminal disapproves the update process, it suppresses the reflection of the feedback information to the update target. Gateway system.

10. An information processing method performed by a gateway system using filters set based on basic rule information, The system acquires input information that the user terminal sends to obtain a response from the generative artificial intelligence, generates first information by filtering the input information using a first filter, and inputs it to the generative artificial intelligence. The generative artificial intelligence that receives the first information continuously outputs sequential information as the answer, and acquires this sequential information for each predetermined transmission unit. The second information is generated by sequentially filtering the sequential information using a second filter different from the first filter. Based on the conversation history between the user terminal and the generative artificial intelligence, or the basic rule information, which includes at least one of the first or second pieces of information, the filter update process is performed. By performing natural language processing on the aforementioned conversation history, the target for updating the filter is selected from the feedback information regarding false positives or missed detections contained in the conversation history. The management terminal is requested to approve or disapprove the update process for the aforementioned feedback information. If the management terminal approves the update process, the feedback information is reflected in the update target; if the management terminal disapproves the update process, the reflection of the feedback information to the update target is suppressed. Information processing methods.

11. A program to be executed by a gateway system using filters configured based on basic rule information, A first processing step involves acquiring input information that a user terminal transmits to obtain a response from a generative artificial intelligence, generating first information by filtering the input information using a first filter, and inputting this information to the generative artificial intelligence. A second processing step involves acquiring sequential information that the generative artificial intelligence, upon receiving the first information, continuously outputs as the response, for each predetermined transmission unit, and generating second information by sequentially filtering the sequential information with a second filter different from the first filter; The system includes an update step which performs the filter update process based on the conversation history between the user terminal and the generative artificial intelligence or the basic rule information, which includes at least one of the first information or the second information, The update step involves performing natural language processing on the conversation history to select the target for updating the filter from the feedback information regarding false positives or false negatives contained in the conversation history. The aforementioned update step is: An approval request step that requests the management terminal to approve or disapprove the update process for the aforementioned feedback information, An information processing method is caused to be executed by a computer, which includes an update execution step, in which the management terminal approves the update process and reflects the feedback information to the update target, and the management terminal disapproves the update process and suppresses the reflection of the feedback information to the update target. program.

Images