Method for simulating a conversation by means of an electronic computing device, computer program product, computer-readable storage medium, and electronic computing device

EP4463807A1Pending Publication Date: 2024-11-20SIEMENS AG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
EP2023739479
Authority / Receiving Office
EP · EP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2022-06-30
Filing Date
2023-06-27
Publication Date
2024-11-20

Smart Images

  • Figure 1.1
    Figure 1.1
Patent Text Reader

Abstract

The invention relates to a method for simulating, by means of an electronic computing device, a conversation such as occurs, for example, in the context of the discussion of complex technical problems. The invention also relates to a and participant configuration not only allows discourses among participants to be simulated, but also allows the behavior of the configured system to be observed computer program product, a computer readable storage medium, and an electronic computing device. By means of the present invention, a method for simulating a conversation using a text analyzer and a text generator is presented for the first time. The combination of the two approaches with simulation configuration and the data from this observation to be made usable.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] 202212210 Foreign Version 1 Description Method for simulating a meeting using an electronic computing device, computer program product, computer-readable storage medium, and electronic computing device. The invention relates to a method for simulating a meeting using an electronic computing device, such as one that takes place, for example, in connection with the discussion of complex technical problems. Furthermore, the invention relates to a computer program product, a computer-readable storage medium, and an electronic computing device therefor. It is already known that meetings on any topic, but especially on the discussion of technically complex problems, can depend heavily on individual factors. For example, the prior knowledge of the individual participants, as well as their mood and personal state, always play a decisive role.whether the meeting is focused and goal-oriented or not. There are theories and models that increase the quality of meetings, especially those for finding new ideas or solutions to complex technical problems. For example, there is the so-called "Six Thinking Hats" by De Bono, a creativity technique that uses linked parallel thinking as a tool for group discussions. There is also the TRIZ theory for solving inventive problems, which is based on abstracting concrete problems until they can be subsumed under a general problem for which there is a general solution. Finally, there are techniques such as "brainstorming" and "mind mapping" that do not provide a structure, but simply collect ideas and attempt toTo establish connections. 202212210 Foreign Version 2 However, the personal factors and dynamics of a meeting are and remain unpredictable and are often not only ineffective but also counterproductive. The object of the present invention is therefore to create a method, a computer program product, a computer-readable storage medium, and an electronic computing device by means of which a meeting is simulated. This object is achieved by the subject matter of the present invention, as disclosed in the description, the claims, and the figures. Accordingly, the subject matter of the invention is a method for simulating a meeting by means of an electronic computing device having at least one input and output device, a simulation controller, and a number of virtual participants.wherein each virtual participant can capture the content of the user text input and / or the generated text input of a virtual participant by means of the electronic computing device via a text analyzer, generate a response to this by means of an artificial intelligence (AI), and return it to the simulation controller, such that a user specifies a topic to the simulation controller using an input device, wherein a) at the start of the simulation, the simulation controller selects a participant to whom it forwards the user text input – or the most recent textual response of the last participant, b) the text analyzer of the selected virtual participant analyzes the content, c) the analysis data is transmitted to a neural network with an AI, d) the AI ​​calculates one or more reactions and / or responses and provides them to the 202212210 Foreign Version 3 e) text generator of this virtual participant,f) the text generator of the selected virtual participant outputs these in text form as response(s) to the simulation controller, g) the simulation controller then selects another participant to forward the response in text form and thus continue the simulation with the same process steps a) to g). In particular, according to an advantageous embodiment, the simulation can be monitored by a human user via the output device and / or automatically by means of the electronic computing device. This creates a method for simulating a meeting that can access a wide variety of data sources via the intermediary AI, for example the IoT or a discussion memory in which data on past meetings, in particular on the topic, can be retrieved, and can thus simulate discussions.which offer new ideas as a final result. According to an advantageous embodiment of the invention, the method also includes a step of automated simulation observation, in which the behavior of the configured system is observed by a module of the electronic computing device. If the observation or monitoring of the simulation is automated,Then, according to an advantageous embodiment of the invention, the monitoring data is stored in a discussion memory and / or is available as training data to the artificial intelligence of the system with the electronic computing device. This invention describes a method for simulating conversations. The solution proposed here for the first time combines text generation models from artificial intelligence with simulation methods. The system can be configured with various properties and participants. Each participant can be configured to exhibit different properties and behaviors. The simulation can be configured to run without human interaction. However, it can also be configured so that the simulation is interrupted and waits for human input.before it continues. In this way, the human user can control the conversation and steer the simulation in specific directions. The invention is used, for example, in the following cases: - Simulation of discussions to generate new ideas. - Simulation and observation of the behavior of systems with participants who have different characteristics, e.g., participants who write negative or skeptical texts, participants with different participation rates in conversations,Fundamentally technocratic and fundamentally emotionally minded participants, and much more. - Generation of ideas for problem solving. - Creation of training data sets for fine-tuning conversational text models. - Creation of verification and validation data sets for interactive systems. - Collection of basic data on human behavior in the context of a discussion. According to an advantageous embodiment, the characteristics of the virtual participants are varied. Each virtual participant can be configured with different characteristics and / or behaviors according to one embodiment of the invention. For example, De Bono's six hats can be virtually recreated. 202212210 Foreign Version 5 For example, the system comprises six virtual participants, each representing a thinking hat according to DeBono, thus - one virtual participant represents the organizing, moderating thinking of the overview of processes,- another virtual participant represents analytical thinking with associated facts, - another virtual participant contributes thinking with feelings and opinions, - another represents critical questioning, concerns, problems, skepticism, and fears, which is countered, for example, by another virtual participant who specializes in best-case scenarios, and - finally, a virtual participant can be configured to represent divergent, creative, lateral thinking. Through strategic simulation control, a response from an opponent can then be generated. Through simulation observation, it is possible, according to one embodiment, to classify the virtual participants' responses according to various criteria, for example, text length, language analysis, e.g., negative, positive, ironic, or skeptical language,for repetitions and / or obvious gaps. Simulation is used in many areas. It allows the result to be simulated and / or varied with different input parameters and the results to be compared. But it also enables the observation of the behavior of the simulated system. The parameters can be subjected to another simulation with only slight changes, and the result can be compared after modifying the configurations. The simulation observation generates further data, which in turn is sent to the electronic computing device via the AI ​​used by each virtual participant.can be made available. This method allows one or more lines of argument to be discussed virtually during a negotiation, and the results of the virtual, detailed discussion can be made available to the negotiating partners. Speech analysis can also be used to draw attention to repetitions during an ongoing discussion. The method always combines two or more content-generating electronic computing devices and generative networks. The present invention combines the approaches of simulation, text generation, and text analysis. For example, it is still a weakness of text or speech analysis and text or speech generation that right or wrong, irony or seriousness, cannot be distinguished. However, if text analysis and text generation are linked to a simulation,As proposed for the first time by this method, the reactions of the other virtual participants reveal what one is dealing with, and irony is recognized through speech generation after the simulation. With more powerful models, the text generation capabilities of language models improve, see GPT-3. These models were trained with large amounts of text data. The models can be used generically without further training. However, the pre-trained models can also be optimized for specific tasks. For example, GPT-3 is a so-called autoregressive transformer model that was trained with huge amounts of data. When GPT-3 was introduced, it was stated that this API programming interface has a universal "text-in, text-out" interface that can handle almost "any English-language task."instead of the usual individual application case. According to the invention, it is proposed to use this interface for the correction, adaptation, and modification of the rules and / or configuration during the simulation process – preferably in real time. By being able to change the rules and procedures of the simulation during the process and / or adapt them to the results of the simulation, this process uses the latest possible techniques to generate freely adaptable simulations – preferably in real time, during an ongoing negotiation. In addition, the process provides the possibility of rewinding the simulation to a point x, which can be defined temporally or otherwise.and, with an adapted configuration, to generate and / or test a new simulation from this point. For this purpose, the system for implementing the method has a so-called discussion memory. GPT-3 is part of a trend in natural language processing (NLP) systems in which language representations are pre-trained. The quality of the text generated by GPT-3 is so high that it can be difficult to determine whether it was written by a human or not. An electronic computing device comprises, for example, processors, circuits, in particular integrated circuits, as well as other electronic components in order to be able to carry out corresponding method steps. Advantageous embodiments of the method are advantageous embodiments of the computer program product,of the computer-readable storage medium and the electronic computing device. The electronic computing device has, in particular, corresponding features to be able to carry out the method. For use cases and / or application situations that may arise during the method and which are not explicitly described here, it can be provided that, according to the method, an error message and / or a request to enter user feedback is issued and / or a default setting and / or a predetermined initial state is set. Further modules of the electronic computing device can include elements such as loudspeakers and / or microphones. These modules have, for example, one or more interfaces (e.g., database interfaces, communication interfaces - e.g., network interfaces,WLAN interface) and / or one or more evaluation unit(s) (e.g., another processor) and / or one or more memory unit(s), such as the discussion memory 34 from Figure 2. For example, data can be exchanged (e.g., received, transmitted, sent, or made available) using the interfaces. For example, data can be compared, checked, processed, assigned, or calculated using the evaluation unit in a computer-aided and / or automated manner. For example, data can be stored in a computer-aided and / or automated manner using the memory unit.be retrieved or provided. An "electronic computing device" is understood to mean a device with at least one processor and / or a neural network. In the context of the invention, a "processor" can be understood to mean, for example, a machine and / or an electronic circuit. A processor can be, in particular, a central processing unit (CPU), a microprocessor, or a microcontroller, for example, an application-specific integrated circuit or a digital signal processor, possibly in combination with a neural network and / or a memory unit for storing program instructions, etc. A processor can also be, for example, an IC (integrated circuit).in particular an FPGA (Field Programmable Gate Array) or an ASIC (Application-Specific Integrated Circuit), or a DSP (Digital Signal Processor) or a GPU (Graphic Processing Unit). A processor can also be understood as a virtualized processor, a virtual machine, or a soft CPU. It can also be, for example, a programmable processor that is equipped with configuration steps for executing the aforementioned method according to the invention or is configured with configuration steps such that the programmable processor implements the inventive features of the method or modules.or other aspects and / or partial aspects of the invention. To carry out the method, the electronic computing unit according to an advantageous embodiment has the following modules: - Simulation control with a simulation configuration in which, for example, the number of simulations, number of virtual participants, number of iterations and / or the type of simulation monitoring is defined. Advantageously, the possibility of a review - in particular feedback in the form of additional input by a human user, etc. is also defined there. If it is intended that a user observes the simulation during the simulation and wishes to change the direction taken, then it can advantageously be provided,that even during the running simulation, it is paused and "corrected" by the user. The simulation control uses the simulation configuration and coordinates the simulation. The simulation control selects which virtual participant is used for the simulation. The simulation control records a history of the simulation, which is returned to the user when the simulation is finished. In addition, according to an advantageous embodiment, it also records statistics about the participants. According to another advantageous embodiment, this data is stored in a discussion memory and, if necessary, made available to a virtual participant as training data. - Virtual participants with participant configuration. Each virtual participant participates in the simulation,when the simulation controller selects it randomly or strategically. It uses a text generator and a text analyzer to generate a response in the form of a new text input. The selected virtual participant verifies that the text it generates conforms to the configuration. If the generated text does not conform to the configuration, another text is generated. The newly generated text is verified again. This is repeated over iterative loops until the generated text is classified as configured by the virtual participant. The participant configuration configures the properties of a virtual participant, for example, attributes about the participant themselves, such as being a strategic thinker who always seeks fundamental principles – comparable to TRIZ theory – as well as attributes about their behavior in the simulation, such as the probabilitythat the 202212210 Foreign Version 11 virtual participant contributes to the simulation without being asked. - Text generator, this module receives text inputs and, based on them, generates a response in the form of a new text input. The original text input can also originate from a human user, e.g., at the start of the simulation, or it originates from another virtual participant during the course of the simulation. The text generator module uses at least one language model, for example, GPT-3. For example, according to an advantageous embodiment, a language model is used in which the generated text is controllable, in particular with regard to the topic, the objects discussed, and / or the tone or politeness of the expression. According to a further advantageous embodiment, the language model(s) is / are based on specific data or areas,especially subject areas are pre-trained and / or fine-tuned. With the help of fine-tuning, a model can be trained, for example, to generate mostly positive-sounding text and / or with a specific language that more or less obviously belongs to a certain field of knowledge. - Text analyzer, a module that analyzes and classifies the generated or input text. According to an advantageous embodiment, the text analyzer can recognize whether the generated text is skeptical, positive, or negative. In addition, the properties of the configuration can be checked via the text analysis of the text input. - Simulation monitoring module, which monitors the simulation process.observed and evaluated. Above all, the results of the simulation are evaluated for their closeness to reality or their remoteness from reality. This role can, for example, be assumed entirely or partially by the human user, depending on the embodiment. This feedback gives the system the opportunity to learn which context-specific reactions lead to creative and targeted contributions, for example in the form of discussion context, participant attitudes within the discussion context, or similar participant-specific configurations. However, simulation monitoring can also be performed automatically, possibly involving a process step of calculating the probability of realizing an idea generated as a result of the simulation. According to an advantageous embodiment, the electronic computing device comprises at least one or more processors, as well as one or more neural networks.one or more storage units, as well as speech input and speech output devices, interconnected to form a system suitably configured to carry out the method. The advantages of the invention disclosed here are: - Generation of ideas generated entirely by software - also as a basis for human discussions on a problem. - Problem-solving simulations to generate new ideas. - Observation of the behavior of systems with different configurations. The conversation can be observed with different participant behavior, e.g., a few participants talk more than others or predominantly skeptical participants. - Simulation observation by humans during the simulation to control the simulation and provide the simulation with additional information or feedback. - The system can learn from human examples,to further improve the generated conversation and its own behavior during the conversation, and conversely, the system can be used to train humans. 202212210 Foreign Version 13 - The simulation can generate data sets for training language models. - The system can learn by collecting user feedback in the observer role. The following figures show schematic block diagrams of two embodiments of an electronic computing device according to the invention. In the figures, identical or functionally equivalent elements are provided with the same reference numerals. Fig. 1 shows a schematic block diagram according to one embodiment of an electronic computing device 10. The central module is the simulation controller 22, which is arranged centrally accordingly. These are the configurations, firstly the simulation configuration 12 and secondly the participant configuration 14.specified. Using these configurations, the simulation controller 22 selects a participant 20 from the pool of virtual participants 20, to whom the user's input is presented as input. Using a text analyzer and a text generator, the selected virtual participant generates a new text input, which is forwarded from the virtual participant 20 to the simulation controller 22. According to an advantageous embodiment shown here, these new text entries are observed by a simulation monitor 24. This can be performed either by a human user or by machine. In any case, the simulation monitor 24 has the option of intervening in the simulation at any time. The simulation process then stops - the so-called "pause function" - and the currently available text is updated by additions, deletions,The correction is then fed back into the running simulation in a modified form. In addition to the pause function, according to an advantageous embodiment of the invention, there is also a "rewind function." 202212210 Foreign Version 14 This means that the process is manually or automatically rewinded to a selected point – - in terms of time, - in terms of configuration, - in terms of participant composition, - in terms of text, - in terms of mood – according to any parameter recorded by the system, definable and specified. From this point, the simulation restarts one or more times, and the result of the new simulation can then be compared with that of the originally created simulation. Various points can be selected in this regard, at each of which the simulation restarts. During the process, every user input, whether human or artificial, is automatically or deliberately initiated.recorded by a protocol and stored or temporarily stored. Inputs in this regard are, firstly, speech inputs in the sense of utterances that can be read as text via speech recognition. Alternatively and additionally, inputs obtained via audio and / or visual methods for recognizing user emotions can also be considered inputs. These include, for example, utterances that are incorporated into the process based on the content, connotation, emotional coloring, pauses, and / or intonation of the participants. In addition, automatically captured data, such as data via facial recognition, gestures, and physiological data of the participants that can be captured via sensors and / or cameras, such as blood pressure, heart rate, facial expressions, and pitch of speech, are also incorporated into the system as inputs. In addition, there are inputs that are less emotionally recognizable, such as text inputs via keyboard,Mouse clicks. 202212210 Foreign Version 15 These entries each lead to a change in the discussion content and result in an entry in a separate storage unit. This storage unit is thus constantly updated, corresponding to a log, which is preferably stored separately in a storage unit, e.g., referred to as the discussion memory, and can be accessed there according to various aspects, such as content, topic, time, participant-related, etc. An entry in the discussion memory can – depending on the specifications and design of the processor – cause a change in the configuration, e.g. - the number of participants is adjusted, i.e., increased or decreased, - the composition of the discussion participants is changed, i.e., some are muted, inactive participants are reactivated, new participants are added with new features,Permissions and / or knowledge are included or excluded, and / or - the participants' modes are changed, e.g., conservative mode is changed to progressive mode, apathetic mode is changed to empathic, active mode, defensive discussion behavior is changed to aggressive. These configuration changes can, for example, be made based on one or more inputs. These configuration changes can be made subsequently, during rewind, i.e., during the discussion.or in real time. The simulation monitor can pause the discussion at any time and, via the discussion memory log, persist the current discussion state, restore any previous state, and continue the simulation from that point – a so-called 202212210 Foreign Version 16 iteration loop. Additionally, the participant configuration and number can be changed while paused. Furthermore, the discussion content can also be changed through insertions, deletions, or corrections. These pausing and resuming operations create a tree-like storage structure of various discussion threads in the discussion memory, which can also be simulated in parallel if necessary. The recordings from the simulation monitor 24 can be transferred to a discussion memory (not shown here).so that they can be used, for example, as training data. When implementing the procedure, the user first configures the simulation configuration and the participant configuration. In the simulation configuration, for example, the number of iteration loops or the discussion duration (as a termination condition), the number of participants, the selection of participants, whether random or strategic, etc., are configured. In the participant configuration, their properties are configured. Depending on the topic, for example, in the case of a technical problem, a skeptic who expects material defects, etc., is countered by an activist who suggests a quick shortcut to a synthesis route by replacing a boiler with a pipe, etc., without fully mentioning the safety risks. As a further property, for example, a virtual participant who always sees the economic results, i.e., sales, profit, etc.Another virtual participant is programmed to essentially present arguments related to sustainability and environmental impact. Initially, a text input is made into the electronic computing device 10, which starts the simulation. The simulation controller 22 then starts the configured number of virtual participants (20). The simulation controller 22 selects, for example, at random, a virtual participant 22 to whom it provides the text input. The selected participant 22 receives the entered data and uses the text generator to create a text. Using the text analyzer, it analyzes the generated text to determine whether it corresponds to the configuration. This process is repeated until the generated text corresponds to the configuration. The simulation controller uses the generated text and selects another, possibly random, participant 20.to continue the simulation. The simulation control 22 also takes into account, for example, the configured threshold of the participant 20 regarding how often they contribute to the simulation. The simulation monitor 24 observes the results of the simulation. This is repeated until the number of configured simulations is completed. The history of the generated simulations is stored in the discussion memory 34 (shown in Figure 2). In addition, statistics on the interactions between the participants 20 are stored, for example. The system has a clock and keeps a log of the discussion participants' statements. Each entry in this log shows when and by whom it was made and, if applicable, in response to which previous statement it refers. In the log, for example, an assignment takes place,which utterance of which participant with which configuration – as input – took place. At any time, the simulation controller can mute participants or remove them from the meeting – all of this is reflected in the minutes and is stored, for example, as a "participant configuration" in the discussion memory. 202212210 Foreign Version 18 This can be particularly relevant with regard to selecting a suitable point for the simulation to return to this point as a "selected point." By creating multiple instances, this system can be scaled across hardware and used to perform ensemble runs, thus eliminating the described burden of serialization. In another embodiment, the system can be configuredthat certain words or topics are more likely to be used. In another embodiment, the user can observe the system during the simulation and control the simulation. These inputs are used by the system to learn from the user and are used for subsequent simulations. In another embodiment, the structure of the conversation can be configured, e.g., a certain number of messages for idea generation, followed by the discussion of certain topics. Figure 2 again shows an electronic computing device 10, which comprises the following modules: a simulation controller 22, a simulation monitor 24, and a discussion memory 34. Communicatively coupled to the simulation controller is a module for discussion simulation, which comprises two virtual participants 20. Each of these virtual participants has a central controller 36,which is communicatively coupled to a text analyzer 18 on the one hand and to a text generator 16 on the other. A human user 30 can feed a simulation configuration 12 and the participant configuration 14, as well as feedback on the results and / or intermediate results of the simulation observation 24, into the system via an input device 26. The user 30 can experience the simulation observation 24 and the result of the simulation via an output device 28. The present invention presents, for the first time, a method for simulating a meeting with a text analyzer and a text generator. The combination of the two approaches with simulation configuration and participant configuration makes it possible not only to simulate discussions among participants, but also to observe the behavior of the configured system and to utilize the data from this observation.

[0002] 202212210 Foreign Version 20 List of Reference Symbols 10 Electronic Computing Device 12 Simulation Configuration 14 Participant Configuration 16 Text Generator 18 Text Analyzer 20 Participant Agent 22 Simulation Control 24 Simulation Monitoring 26 Input Device 28 Output Device 30 Human Participant 32 Discussion Simulation+ 34 Discussion Memory 36 Agent State and Control

Claims

202212210 Foreign version 21 Patent claims 1. Method for simulating a meeting by means of an electronic computing device (10) with at least one input and output device (26, 28), a simulation controller (22) and a number of virtual participants (20), wherein each virtual participant (20) can capture the content of the user text input (30, 26) or the generated text input of a virtual participant (20) by means of the electronic computing device (10) via a text analyzer (18), generates a response thereto by means of an artificial intelligence "AI" and returns it to the simulation controller (22), such that a user (30) specifies a topic to the simulation controller (22) by means of an input device (26), wherein a) to start the simulation, the simulation controller (22) selects a participant (20) to whom it transmits the user input - or the most recent response of the last participant (20) in text form - forwards,b) the text analyzer (18) of the selected virtual participant (20) analyzes the content, c) the analysis data is transmitted to a neural network with an AI, d) the AI ​​calculates one or more reactions and / or answers and provides them to the e) text generator (16) of this virtual participant (20), f) the text generator (16) of the selected virtual participant (20) outputs these in text form as answer(s) to the simulation controller (22), g) the simulation controller (22) then selects another participant (20) to forward the answer in text form and thus continue the simulation with the same method steps a) to g).

2. Method according to claim 1, wherein simulation monitoring (24) is carried out. 202212210 Foreign Version 22 3. The method according to claim 2, wherein the simulation monitoring (24) is carried out via the output device (28) and / or feedback from the human user (30).

4. The method according to one of the preceding claims 2 or 3, wherein automated simulation monitoring (24) is carried out by the electronic computing device.

5. The method according to one of the preceding claims 2 to 4, wherein the data of the simulation monitoring (24) and / or the feedback from the human user (30) are stored in a discussion memory (34).

6. The method according to claim 5, wherein the data from the discussion memory (34) are used as training data for the AI.

7. The method according to one of the preceding claims 1 to 6, wherein the selection of a virtual participant (20) by the simulation controller (22) is random.Method according to one of the preceding claims 1 to 7, in which the participant configuration (14, 36) is at least partially based on the theory of De Bono.

9. Method according to one of the preceding claims 1 to 8, in which the participant configuration (14, 36) is at least partially based on the TRIZ theory.

10. Method according to one of the preceding claims 1 to 9, which is carried out as part of an interactive method for conducting a discussion via the at least one input and output device (26, 28).

11. Method according to one of the preceding claims, in which the text generator (16) comprises a language model. 202212210 Foreign Version 23 12. Method according to one of the preceding claims, in which the text generator (16) is designed and / or configured such that GPT-3 is used.

13. Computer program product with program code means which cause an electronic computing device (10) to carry out a method according to one of claims 1 to 12 when the program code means are processed by the electronic computing device (10).

14. Computer-readable storage medium with a computer program product according to claim 13.

15. Electronic computing device (10) for simulating a meeting on a predetermined topic with at least one simulation controller (22), two or more virtual participants (20), each comprising a text analyzer (18) and a text generator (16), wherein the electronic computing device (10) is designed to carry out a method according to one of claims 1 to 12.