Method for communication between a client and an artificial intelligence agent

Abstract

The invention relates to a method for communication between a client and an artificial intelligence agent associated with at least one artificial intelligence model, the method being executed by the artificial intelligence agent and comprising: - receiving (200), by the artificial intelligence agent, a request from the client, the request comprising a demand for omission of the context of the artificial intelligence agent, - transmitting (201) the request by the artificial intelligence agent to the at least one artificial intelligence model without adding information from the context of the artificial intelligence agent.

Description

Description Title: Communication method between a client and an artificial intelligence agent technical field

[0001] The present invention relates to the field of artificial intelligence and more specifically to the communication protocols between a user client and an artificial intelligence agent.

[0002] The invention aims in particular to allow a customer to have a private query of an artificial intelligence agent in order to provide the customer with more information on all the data taken into account by the artificial intelligence model to respond to a request from the customer. Previous technique

[0003] Artificial intelligence techniques, and in particular large language models (LLMs), are increasingly used in the execution of tasks such as producing a response text to an intention, diagnosing an incident, resolving that incident, resolving a malfunction, and optimizing a function.

[0004] In this context, communications assisted by an artificial intelligence agent, particularly generative artificial intelligence, are multiplying.

[0005] In order to improve the user experience and the content of the generated responses, generative artificial intelligence models personalize responses based, in particular, on the history of the user's requests which are archived in a local context.

[0006] This personalization is generally done in a way that is not transparent to the user, without their consent and without information to understand whether the response provided is biased or not by previous requests.

[0007] According to the previous technique, artificial intelligence agents that run generative models can save all or part of the query and response history for a user, regardless of the medium used to generate the query.

[0008] In particular, when an agent receives a request, also called a "prompt", from a user, it can use a set of contextual data to develop a response to the user's request before passing it on to the artificial intelligence model.

[0009] Thus, a user wishing for an answer not biased by previous interactions with the artificial intelligence agent does not have the possibility of knowing whether the history of their requests is taken into account to generate a new answer and cannot prevent it.

[0010] The present invention aims to overcome the limitations of the prior art. Summary of the invention

[0011] The purpose of this disclosure is to address all or part of the limitations of prior art solutions, including those described above, by proposing a solution that allows a user to send a request to an artificial intelligence agent while explicitly requesting that any history of their requests not be taken into account or be reset.

[0012] To this end, a communication method is proposed between a client and an artificial intelligence agent associated with at least one artificial intelligence model, the method being executed by the artificial intelligence agent and comprising: The artificial intelligence agent receives a request from the client, the request including a request to omit the context from the artificial intelligence agent. a transmission of the request by the artificial intelligence agent to at least one artificial intelligence model without adding information from the context of the artificial intelligence agent.

[0013] This process has the advantage of offering a method of querying an artificial intelligence agent that does not take into account the agent's context, in particular the history of exchanges between the agent and the client.

[0014] According to one aspect of the invention, the method further comprises adding to said response information to validate or not validate the request to omit the context of the artificial intelligence agent.

[0015] Thus, the client is notified that their request to omit the agent's context has been taken into account.

[0016] According to one aspect of the invention, the transmission of the request to at least one artificial intelligence model comprises: A conversion of the query content into at least one piece of data usable by the artificial intelligence model, A transmission of said at least one usable data without adding information to at least one artificial intelligence model.

[0017] Thus, the usable data provided to the artificial intelligence model includes no information other than that from the client's request. The artificial intelligence model formulates its response without bias introduced by external or historical information.

[0018] According to one aspect of the invention, the method includes resetting the context of the artificial intelligence agent upon receipt of said request.

[0019] Setting the context to zero allows the artificial intelligence model to develop a response without bias induced by contextual information, even if the model uses context.

[0020] The invention also relates to a method of communication between a client and an artificial intelligence agent associated with at least one artificial intelligence model, the method being executed by the client or an intermediate device between the client and the artificial intelligence agent and comprising a transmission of a request to the artificial intelligence agent, the request comprising a request to omit the context of the artificial intelligence agent.

[0021] This process has the advantage of allowing a client to formulate a request including an explicit request not to take into account the agent's context when preparing the response to the request.

[0022] According to one aspect of the invention, this method comprises: - the reception by the intermediary device of a request issued by the client to the artificial intelligence agent, - an addition of the request to omit the context of the artificial intelligence agent to the query, - the transmission of the request to the artificial intelligence agent.

[0023] Using an intermediary device allows the task of requesting context omission to be shifted to that device instead of the client itself.

[0024] According to one aspect of the invention, the request to omit the context of the artificial intelligence agent to the request is conditioned on at least one piece of information characterizing the client.

[0025] For example, this information includes location data, legal information specific to the user, or information indicating the client device's domain affiliation. Therefore, it's possible to restrict the request to omit this context to certain clients while denying this functionality to others.

[0026] According to one aspect of the invention, the method comprises a transmission by the artificial intelligence agent to the client of a response from at least one artificial intelligence model to said request. The response provided by the artificial intelligence model is thus generated without taking the context into account.

[0027] According to one aspect of the invention, the context of the artificial intelligence agent includes a history of requests and / or responses previously exchanged between the client and the artificial intelligence agent. One advantage of omitting this history is that it can introduce unwanted biases into the generated response.

[0028] The invention also relates to a communication device between a client and an artificial intelligence agent associated with at least one artificial intelligence model, the device being implemented by the artificial intelligence agent and comprising a processor coupled to a communication device and a memory in which program instructions are stored for execution by the processor to: Receive a request from the client, the request including a request to omit the context of the artificial intelligence agent, Transmit the request to at least one artificial intelligence model without adding information from the context of the artificial intelligence agent.

[0029] This execution device is capable of implementing in all its modes of embodiment the communication process between a client and an artificial intelligence agent, executed by the artificial intelligence agent described above.

[0030] The invention also relates to a communication device between a client and an artificial intelligence agent associated with at least one artificial intelligence model, the device being implemented by the client or an intermediate device between the client and the artificial intelligence agent and comprising a processor coupled to a communication device and a memory in which are stored program instructions intended to be executed by the processor to transmit a request to the artificial intelligence agent, the request including a request to omit the context of the artificial intelligence agent.

[0031] This execution device is capable of implementing in all its embodiments the communication process between a client and an artificial intelligence agent executed by the client or an intermediate device described above.

[0032] When the above device is implemented by the intermediate device, the program instructions are further intended to be executed by the processor to: - receive a request issued by the client to the artificial intelligence agent, - add the request to omit the context of the artificial intelligence agent to the request, - transmit the request to the artificial intelligence agent.

[0033] The invention also relates to a computer program product comprising a set of program code instructions which, when executed by at least one processor, configure said at least one processor to implement the communication method according to any one of the implementation modes of this disclosure.

[0034] The invention also relates to a computer-readable recording medium on which is recorded a set of program code instructions which, when executed by at least one processor, configure said at least one processor to implement the communication method according to any one of the implementation modes of this disclosure. Brief description of the drawings

[0035] The invention will be better understood upon reading the following description, given by way of non-limiting example, and made with reference to the figures which represent: [Fig. 1]: a schematic representation of an artificial intelligence module according to a first embodiment, [Fig. 2]: a diagram illustrating the main steps of a communication process according to an example, [Fig. 3]: a representation of a communication device according to another example.

[0036] In these figures, identical references from one figure to another designate identical or analogous elements. For clarity, the elements shown are not to scale unless otherwise indicated. Description of the implementation methods

[0037] More generally, it should be noted that the implementation and realization methods considered below are described as non-limiting examples, and that other variants are therefore conceivable.

[0038] Figure 1 schematically illustrates an example of an artificial intelligence module operating as an AI agent (AG).

[0039] Such an AG module is configured to receive a Req or "prompt" request from a client user and directed to an artificial intelligence (AI) model, for example, a generative model. The AG module includes a preprocessing module (PreProc) and a postprocessing module (PostProc). The PreProc and PostProc modules interact with a DTA database, an OTL tools module, and other MDL models. The MDL models are organized as functions that can be queried through an application programming interface (API). In one embodiment, the other models are a subset of the OTL tools module.

[0040] The artificial intelligence model can be any type of known model, for example a generative model such as an LLM model (“large language model”).

[0041] Upon receiving the Req request, it is processed by the PreProc preprocessing module, which transforms the request content into data usable by the artificial intelligence model. For example, in the context of natural language processing (NLP) and multimedia processing, the PreProc preprocessing module transforms the request content into tokens. This "tokenization" operation is used to convert the request content, such as text, into an intermediate representation usable by the AI ​​model.

[0042] The PostProc post-processing module is configured to then transform data it receives from the artificial intelligence model into a response readable by the user client.

[0043] Using a prior art technique, the PreProc preprocessing module also accesses the history of requests previously issued by the same user, which are stored in the DTA database. All or part of this history is added to the token string that is then passed to the AI ​​model. The AI ​​processes the request and produces a response, also in the form of a token string. The token string corresponding to the Rep response is then converted into text by the PostProc postprocessing module and sent to the client user.

[0044] As mentioned earlier, a drawback of this approach is that the user doesn't know whether or which elements of their query history are being used by the AI ​​agent to generate a new response. Furthermore, the user cannot query the AI ​​model without relying on the history.

[0045] Figure 2 illustrates the main steps of a communication process between a client user and an artificial intelligence agent AG of the type described in Figure 1, according to one embodiment of the invention.

[0046] The client user is, for example, a terminal connected to a network device. The artificial intelligence agent is implemented, for example, according to the architecture shown in Figure 1, to provide an artificial intelligence service to the client user via the network device.

[0047] The process begins at step 200 with the transmission by the client user to the artificial intelligence agent of a request or "prompt" accompanied by a request to omit the context of the artificial intelligence agent to generate a response to this request.

[0048] Context here refers to any information, data, or rules that the artificial intelligence model can take into account when formulating a response to a user's query. Specifically, context can include data extracted from the history of previous queries and / or responses to the same user. It can also include user-independent rules for generating a response, such as rules concerning the semantic content generated by the AI ​​or rules related to legislation. Finally, it can include data extracted from the history of queries and / or responses exchanged with other users.

[0049] Typically, the client user's request may include or correspond to a text query. The query may also include or correspond to other types of data, for example multimedia information (image, video, sound).

[0050] The request to omit context from the artificial intelligence agent is, for example, a numerical piece of information attached to the query. This numerical information is, for example, integrated into a free-form signaling field or one specifically designed for this purpose. It is, for example, entered as a Boolean value so that the information can take the value TRUE or FALSE depending on whether the user wants the AI ​​model to be queried without context or, conversely, with context.

[0051] In one embodiment, communication between the client and the artificial intelligence agent is end-to-end encrypted. The communication is encrypted via the establishment of an encrypted tunnel, for example, a tunnel implementing the TLS (Transport Layer Security) protocol.

[0052] According to one embodiment, the client connects to the server hosting the artificial intelligence agent via a communication protocol such as http or https.

[0053] According to one embodiment, the customer request is accompanied by other information such as a key identifying the user, or one or more operating parameters of the AI ​​artificial intelligence model such as the temperature to be established during the model inference.

[0054] In one embodiment, the communication architecture includes an intermediary device (router, proxy, gateway, etc.). This intermediary device is located between the network equipment to which the client is connected and the server hosting the artificial intelligence agent. The omission request can be sent by the client user to the intermediary device. The omission request can be appended to the request. In such a scenario, an initial encrypted communication is established between the client and the intermediary device, and a second encrypted communication is established between the intermediary device and the server hosting the artificial intelligence agent. In another example, the intermediary device can insert, or decide to insert, the omission request from the context into the client's request.

[0055] In one embodiment, the intermediate device adds the context omission request to the client's request based on information characterizing the client. For example, this information includes location information, legal information specific to the user, or information indicating the client terminal's membership in a particular domain.

[0056] In step 201 of the process, the artificial intelligence agent handles the request received from the client. If the AI ​​agent supports the request to omit context, it decodes the associated information. If this request is successful, it transmits the content of the request to the AI ​​model without querying the DTA database, which contains the history of requests from the same user, and without adding any other contextual information.

[0057] This operation is performed, for example, in the PreProc preprocessing module. The PreProc module transforms the content of the Req query into usable data for the artificial intelligence model, known as input data (for example, a token string), and transmits it directly to the AI ​​model. In this way, the AI ​​agent does not insert any additional contextual data.

[0058] According to one embodiment, the user's query history contained in the DTA database is deleted, thus resetting the history to zero.

[0059] In step 202, the AI ​​model generates data, called output data (for example, a token string). The output data is then transformed by the PostProc post-processing module into a response readable by the user client. The response can be multimedia content, for example, in text, audio, and / or video format. The generated response is then transmitted to the user client.

[0060] In one embodiment, information indicating whether the AI ​​agent's request to omit context has been taken into account is added to the response. For example, this information is Boolean, with a value of TRUE indicating to the user that the response was generated without considering the context, and a value of FALSE indicating the opposite. In another example, a message or text can be displayed with the response, containing information indicating whether the AI ​​agent's request to omit context has been taken into account. The displayed message could be "no history included".

[0061] In this way, the client user is informed that their request history is not taken into account when generating the response to a current request.

[0062] Figure 3 shows a diagram of an example DIS device configured to implement the process of Figure 2.

[0063] The DIS device includes a data processing module comprising a storage space, for example a MEM memory, a processing unit, equipped for example with a PROC microprocessor, and controlled by a computer program PGR whose instructions are configured to implement the communication process as described in relation to Figure 2.

[0064] At initialization, the code instructions of the PGR computer program are loaded into MEM memory before being executed by the processing unit's processor. The PROC microprocessor of the processing unit then implements, according to the instructions of the PGR computer program, the steps of the execution process described with reference to Figure 2.

[0065] To achieve this, in addition to the MEM memory and the PROC processor, the device includes COM communication means, enabling it to exchange messages with a server hosting an artificial intelligence agent. These communication means include, for example, a wired communication bus and can utilize communication methods such as an API, a Web API, and / or a communication protocol such as HTTP or SI P.

Claims

Demands

1. A computer-implemented method for communication between a client and an artificial intelligence agent associated with at least one artificial intelligence model, the method being executed by the artificial intelligence agent and comprising: A reception (200) by the artificial intelligence agent of a request from the client, the request including a request to omit the context from the artificial intelligence agent, a transmission (201) of the request by the artificial intelligence agent to at least one artificial intelligence model without adding information from the context of the artificial intelligence agent.

2. A computer-implemented method of communication according to claim 1, wherein the transmission (201) of the request to at least one artificial intelligence model comprises: - A conversion of the query content into at least one piece of data usable by the artificial intelligence model, - A transmission of said at least one usable data without adding information to at least one artificial intelligence model.

3. A computer-implemented method of communication according to any one of the preceding claims, comprising resetting the context of the artificial intelligence agent upon receipt of said request.

4. A computer-implemented method of communication between a client and an artificial intelligence agent associated with at least one artificial intelligence model, the method being executed by the client or an intermediate device between the client and the artificial intelligence agent and comprising a transmission (200) of a request to the artificial intelligence agent, the request comprising a request to omit the context of the artificial intelligence agent.

5. A computer-implemented method of communication according to claim 4, wherein the method is executed by the intermediate device, the method comprising: - the reception by the intermediary device of a request issued by the client to the artificial intelligence agent, - an addition of the request to omit the context of the artificial intelligence agent to the query, - the transmission of the request to the artificial intelligence agent.

6. A computer-implemented method of communication according to any one of claims 4 or 5, wherein the request to omit context from the artificial intelligence agent to the request is conditional on at least one piece of information characterizing the client.

7. A computer-implemented method of communication according to any one of the preceding claims comprising a transmission (202) by the artificial intelligence agent to the client of a response from at least one artificial intelligence model to said request.

8. A computer-implemented method of communication according to claim 7, further comprising adding to said response information to validate or not validate the request to omit the context of the artificial intelligence agent.

9. A computer-implemented method of communication according to any one of the preceding claims, wherein the context of the artificial intelligence agent includes a history of requests and / or responses previously exchanged between the client and the artificial intelligence agent.

10. A communication device between a client and an artificial intelligence agent associated with at least one artificial intelligence model, the device being implemented by a server hosting the artificial intelligence agent and comprising a processor (PROC) coupled to a communication device (COM) and a memory (MEM) in which program instructions (PGRs) are stored for execution by the processor to: - Receive a request from the client, the request including a request to omit the context of the artificial intelligence agent, - Transmit the request to at least one artificial intelligence model without adding information from the context of the artificial intelligence agent.

11. Communication device between a client and an artificial intelligence agent associated with at least one artificial intelligence model, the device being implemented by a client terminal or an intermediate device between the client terminal and the artificial intelligence agent and comprising a processor (PROC) coupled to a communication device (COM) and a memory (MEM) in which program instructions (PGR) are stored for execution by the processor to transmit a request to the artificial intelligence agent, the request including a request to omit the context of the artificial intelligence agent.

12. Communication device according to claim 11, wherein the device is implemented by the intermediate device and the program instructions are further intended to be executed by the processor for: - receive a request issued by the client to the artificial intelligence agent, - add the request to omit the context of the artificial intelligence agent to the request, - transmit the request to the artificial intelligence agent.

13. Communication device according to any one of claims 10 to 12, wherein the context of the artificial intelligence agent includes a history of requests and / or responses previously exchanged between the client and the artificial intelligence agent.

14. Product computer program comprising program code instructions for implementing a communication method according to any one of claims 1 to 9 when said program is executed by a processor.

15. A recording medium readable by a communication device, on which the program according to claim 14 is recorded.

Images