DETAILED ACTION
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
2. In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
3. The following office action is a Final Office Action in response to the communications received on 04/17/2026.
Claims 1-4, 9-12 and 17-21 have been amended; claims 7 and 15 are canceled. Therefore, claims 1-6, 8-14 and 16-22 are currently pending in this application.
Claim Rejections - 35 USC § 101
4. Non-Statutory (Directed to a Judicial Exception without an Inventive Concept/Significantly More)
35 U.S.C.101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
● Claims 1-6, 8-14 and 16-22 are rejected under 35 U.S.C.101 because the claimed invention is directed to an abstract idea without significantly more.
(Step 1)
The current claims fall within one of the four statutory categories of invention (MPEP 2106.03).
Step 2A [Wingdings font/0xE0] Prong One:
The claim(s) recite a judicial exception, namely an abstract idea, as shown below:
— Considering each of claims 1, 9 and 17 as the representative claim, the following claimed limitations recite an abstract idea:
receive an identifier of a rule based on a request;
identify a subset that includes an identifier of the rule within metadata of the subset;
detect a drift between current chat content associated with the rule and the rule itself based on [analyzing] the subset and the rule itself;
update parameters based on text content described in the rule and the detected drift; and
[use] the updated parameters to [provide] a response during a chat session.
Thus, the limitations identified above recite an abstract idea since the limitations correspond to certain methods of organizing human activity, and/or mental processes, which are part of the enumerated groupings of abstract ideas identified according to the current eligibility standard (see MPEP 2106.04(a)). For instance, the current claims correspond to managing personal behavior. In particular, as a user is discussing about a given topic/rule during a chat session, the content of the chat is evaluated in order to determine whether there is a drift between the chat content associated with the rule and the rule itself—such as, by comparing the content of the chat that relates to the rule with the actual content of the rule, etc., and thus, when a drift is detected, parameters are updated based on text content described in the rule and the detected drift is performed; and, using the updated parameters, a response is presented to the user during a chat session, etc.
Similarly, given the limitations that recites the process of: receiving an identifier of a rule based on a request; identifying a subset that includes an identifier of the rule within metadata of the subset; detecting a drift between current chat content associated with the rule and the rule itself, etc., the claims also correspond to mental processes, i.e., limitations that can be performed in the human mind and/or using a pen and paper (e.g., an evaluation, an observation, and/or a judgment process, etc.).
Step 2A [Wingdings font/0xE0] Prong Two:
The claim(s) recite additional element(s), wherein a computer system, which implements a processor, a memory, etc., is utilized to facilitate the recited functions/steps regarding: collecting information (e.g., receiving an identifier of a rule based on a request); analyzing the collected information using one or more algorithms, which include AI models (e.g., identify a subset of vectors within a vector database that include an identifier of the rule stored within metadata of the subset of vectors; detecting a drift between current chat content associated with the rule and the rule itself based on execution of an artificial intelligence (AI) model on the subset of vectors and text content of the rule); generating one or more content items (e.g., update parameters of a chatbot model based on text content described in the rule and the detected drift, and execute the chatbot model with the updated parameters to output a chatbot response during a chat session), etc.
However, the claimed additional element(s) fail to integrate the abstract idea into a patent-eligible practical application since the additional element(s) are utilized merely as a tool to facilitate the abstract idea. Accordingly, when each of the claims is considered as a whole, the additional element(s) fail to impose meaningful limits on practicing the abstract idea. For instance, when each of the claims is considered as a whole, none of the claims provides an improvement over the relevant existing technology.
The observations above confirm that the claims are indeed directed to an abstract idea.
Step 2B
Accordingly, when the claim(s) is considered as a whole (i.e., considering all claim elements both individually and in combination), the claimed additional elements do not provide meaningful limitations to transform the abstract idea into a patent eligible application of the abstract idea such that the claim(s) amounts to “significantly more” than the abstract idea itself (also see MPEP 2106). The claimed additional elements are directed to conventional computer elements, which are serving merely to perform conventional computer functions.
Accordingly, when each of the current claims is considered as a whole (e.g., see the discussion under Prong Two above regarding such consideration of the claim as a whole), none of the claims recites an element—or a combination of elements—directed to an inventive concept.
It is also worth noting—per the original disclosure—that the claimed invention is directed to a conventional and generic arrangement of the additional elements. For instance, the specification describes a system that comprises one or more commercially available conventional computing devices—such as, a laptop computer, a desktop computer, etc. ([0179]; [0180]); wherein the conventional computing device(s) communicates, over the conventional communication network (e.g., the Internet), with at least one server of a service provider; and thereby, the system provides a user with relevant information based on the analysis of collected interactions ([0188] to [0192]).
Of course, the system above executes one or more known algorithms, including artificial intelligence and/or machine learning algorithms, in order to analyze the collected information/conversations (e.g., see [0037]; [0051] to [0054]).
In addition, the use of the existing computer/network technology to facilitate the process of providing/updating a relevant information/content to a user(s), based on the analysis of collected of interactions or conversations, including executing one or more artificial intelligence and/or machine learning algorithms to perform the analysis of the collected interaction, etc., is already directed to a well-known, routine, conventional activity in the art (US 2017/0316326; US 2014/0074688; US 2008/0114737, etc.).
Note also that: (a) the use of one or more artificial intelligence (AI) models to adapt the output/result that a computer is generating to a user, including the process of generating training data to retrain one or more of the AI models in order to enhance the accuracy of the results that the computer is generating to the user (e.g., US 8,775,332; US 10,257,225; US 2019/0236417, etc.); (b) the use of a generative artificial intelligence—as an interactive tool—to facilitate a more realistic human-like natural conversations (e.g., US 2018/0020093; US 2018/0376002; also “Deep Reinforcement Learning For Dialog Generation”, the Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1192–1202, Austin, Texas, November 1-5, 2016. ©2016 Association for Computational Linguistics), etc., are already part of the conventional computer/network technology.
The observation above confirms that the current claims fail to amount to “significantly more” than an abstract idea.
It is worth noting that the above analysis already encompasses each of the current dependent claims (i.e., claims 2-6, 8, 10-14, 16 and 18-22). Particularly, each of the dependent claims also fails to amount to “significantly more” than the abstract idea since each dependent claim is directed to a further abstract idea, and/or a further conventional computer element(s) utilized to facilitate the abstract idea.
Accordingly, the findings above demonstrate that none of the claims implements an element—or a combination of elements—directed to an inventive concept (e.g., none of the current claims is reciting an element—or a combination of elements—that provides a technological improvement over the existing/conventional technology).
● Claims 17-20 further fail to comply with 35 U.S.C.101 since these claims are directed to non-statutory subject matter. Particularly, claims 17-20 are directed to a computer readable storage medium.
It is worth noting that a computer readable storage medium broadly covers both statutory and non-statutory subject matter (e.g., signal per se). However, claims 17-20 do not positively exclude the non-statutory subject matter. Also see MPEP 2106.03(I) (emphasis added),
Non-limiting examples of claims that are not directed to any of the statutory
categories include:
Products that do not have a physical or tangible form, such as information (often referred to as “data per se”) or a computer program per se (often referred to as “software per se”) when claimed as a product without any structural recitations;
Accordingly, claims 17-20 further fail to comply with the statutory requirement per section §101.
Note also that the original disclosure also appears to encompass both statutory and non-statutory subject matter, “[a] computer program may be embodied on a computer readable medium, such as a storage medium. For example, a computer program may reside in random access memory ("RAM"), flash memory . . . or any other form of storage medium known in the art” (see [0176], emphasis added).
► Applicant’s arguments directed to section §101 have been fully considered (the response filed on 04/17/2026); However, the arguments are not persuasive at least for the following reasons:
Firstly, regarding Prong One of Step 2A, Applicant is asserting that “[c]laim 1 cannot practically be performed in the human mind because the system relies on computational processing of vectorized chat content stored in a database and the execution of an AI model to detect semantic drift across large volumes of communication data. The system further updates parameters of a deployed chatbot model and then uses the updated chatbot model to provide responses. Thus, Applicant contends that the inquiry into the abstraction of Applicant's claims should end at this first prong as Applicant's claims do not fall into any of the enumerated groupings and therefore cannot be abstract” (emphasis added).
However, Applicant appears to fail to properly apply the inquiry established under Prong One of Step 2A. Similar to the point made in the previous office action, it is once again worth noting that Prong One of Step 2A does not consider any of the claimed computer elements, which are part of the additional elements. Instead, Prong One requires one to identify merely the limitations that recite a judicial exception (such as, an abstract idea); see MPEP 2106.07(a), (emphasis added),
For Step 2A Prong One, the rejection should identify the judicial exception by referring to what is recited (i.e., set forth or described) in the claim and explain why it is considered an exception. For example, if the claim is directed to an abstract idea, the rejection should identify the abstract idea as it is recited (i.e., set forth or described) in the claim and explain why it is an abstract idea.
In contrast, while emphasizing one or more of the additional elements (e.g., the alleged “computational processing of vectorized chat content stored in a database and the execution of an AI model”, etc.), Applicant is attempting to challenge the finding presented under Prong One. Consequently, Applicant’s arguments are not persuasive. This is because the claims do recite limitations that can practically be performed in the human mind and/or using a pen and paper. For instance, while excluding the claimed computer elements, a human—such as a human supervisor and/or agent—can perform the following limitations of claim 1 mentally and/or using a pen and paper:
the supervisor acquires—via a pen and paper—an identifier (e.g., a name, or a numeric identifier, etc.) that relates to a particular rule based on a request (e.g., a verbal or a written request received during interaction between an agent and a customer);
the supervisor subsequently identifies the relevant rule (e.g., a particular page on a manual, and/or a particular document from a collection of documents, etc.) based on the identifier above; and further evaluates the agent’s discussion with the customer in order to determine whether the relevant rule is being applied properly or not—such as, determining a drift between the agent’s narration of the rule and the actual content of the rule (e.g., the text on the manual/document);
the supervisor then updates (e.g., summarizes on a piece of paper, etc.) one or more issues that need to be addressed;
the supervisor provides the above updated summary to the agent; thereby, helping the agent to learn from the updated summary so that the agent will be able to provide proper response during an interaction (a chat session), etc.
The observation above confirms that the claims do recite an abstract idea; such as, a mental process. Consequently, Applicant’s arguments are once again not persuasive.
Secondly, regarding Prong Two of Step 2A, Applicant asserts, “the claims impose meaningful, practical, and succinct operations that updates parameters of a chatbot model based on detected drift in conversational content on a message board and then executes the updated chatbot model to generate responses during an active chat interaction. The process improves the response generation behavior of a chatbot, such as a chatbot in the message window, based on detected inconsistencies between organizational rules and observed communication patterns . . . also improves the functioning of the chatbot itself by aligning generated responses with rule content when drift is detected in underlying communication content. Instead of merely analyzing the content, the system uses the results of the AI-based drift detection to update model parameters and control subsequent response generation” (emphasis added).
However, except for the attempt made to summarize the features of the existing computer/network technology, Applicant fails to demonstrate whether any of the current claims is implementing an element—or a combination of elements—that integrates the abstract idea into a patent-eligible practical application. This is because Applicant fails to demonstrate an element (if any) or a combination of elements (if any) that imposes meaningful limits on practicing the abstract idea. It is again worth noting that the claimed (and the originally disclosed) system/method is directed to the computer/network technology. Accordingly, an integration (if any) of the abstract idea into a patent-eligible practical application is demonstrated if any of the claims is implementing at least one element—or a combination of elements—that provides a technological improvement over the existing computer/network technology; such as, improvement (if any) with respect to the functioning of the computer itself (e.g., see Enfish); an advanced technique (if any) of training the AI model (e.g., see Ex Parte Desjardins), etc.
In contrast, while mistaking the features of the existing computer/network technology for an advanced feature, Applicant appears to be attempting to substantiate an alleged technological improvement regarding the current claims. For instance, unlike Applicant’s assertions, it is part of the existing computer/network technology to train a chatbot model when the chatbot starts to provide one or more responses that deviate (or drift) away from an expected response. In particular, as the name already indicates, a machine-learning model continuously learns—and thereby updates—one or more of its parameters based on newly collected or updated information (e.g., a new/updated document that relates to a particular policy, etc.). Of course, even basic common sense dictates that the crux of such machine-learning process is to improve the accuracy of the results that the model is generating. Accordingly, once the chatbot is trained in such manner, it generates relevant and accurate responses to the user during interaction.
Although a reference is not necessarily required to show the fact above, at least one of the references cited as part of the Step 2B analysis already confirms the above fact. For instance, Bentitou (US 2018/0020093) discloses such a system directed to the existing computer/network technology; and this system implements a generative AI to manage one or more conversations with one or more customers (see [0164], emphasis added),
“. . . the AI obtains the details by asking questions or by checking the CallerID and ANI data and asking questions. The questions may be pre-programmed into the AI such that it always asks the same questions. The answers to the questions may be the details the AI wants to obtain. The questions may also be constructed from the responses received from the calling party using a generative model (except the first question that may need to pre-programed if the AI is configured to start the conversation first). The AI may also provide answers to the questions from the calling party. The questions and answers from the AI change according to the answers and questions received from the calling party. The AI can understand or analyze the content in the answers and questions from the calling party (through a trained neural network or trained AI . . .) and provide questions and answers relevant to or based on the content . . . the content may further include answers and questions from the AI. As such, the AI can analyze the content in the answers and questions from the calling party and the content in the answers and questions from the AI. This further inclusion or analysis may allow the AI to provide more accurate or detailed questions and responses”
Thus, it is evident from the excerpt above that the AI model learns not only from questions and answers that others provided, but also from questions and answers it is generating during interactions; and such inclusive and comprehensive learning process allows the generative AI to provide “more accurate . . . responses” (emphasis added). In contrast, despite relying on such features of the existing computer/network technology, Applicant is attempting substantiate an alleged technological improvement regarding the current claims (e.g., see Applicant’s alleged process that supposedly improves “the response generation behavior of a chatbot”, “the functioning of the chatbot itself”, etc.). Consequently, Applicant’s arguments are not persuasive.
In addition, while referring to the decision regarding Ex parte Desjardins, Applicant is asserting that “Claim 1 includes a process of detecting drift between current chat content and textual content described within a rule, and updating parameters of a chatbot model based on the detected drift and the rule content. These operations modify the behavior of the chatbot model itself so that responses generated during a chat session reflect alignment with the rule. Thus, similar to the claims in Ex parte Desjardins, the claim is directed to improving how a machine learning model operates rather than merely applying an abstract algorithm. By dynamically adjusting model parameters based on detected semantic drift between communications and rule content, Claim 1 improves the functioning and reliability of the chatbot system when generating responses during chat interactions” (emphasis added).
However, once again Applicant is emphasizing the features of the existing computer/network technology in an attempt to substantiate an alleged technological improvement. Note that it is one of the existing and fundamental features (or functions) of a machine-learning algorithm to learn from new or updated information. For instance, a ML model (e.g., a chatbot) may provide a given answer—e.g., answer “A”—to a user when the user asks a first question. However, when the model receives new/updated information, it may recognize—based on analyzing the new/updated information—that the previous answer (answer “A”) is no longer accurate to address the first question above. Subsequently, the ML model makes one or more adjustments based on the above discovery; such as: (i) assigning a low confidence value/score to the old answer (i.e., answer “A”) while assigning a high confidence value/score to the newly discovered answer (e.g., answer “B”); (ii) simply discarding the old answer and replacing it with the new answer, etc. Accordingly, such learning process is essentially the process of updating one or more parameters of the ML model. Of course, once updated in such manner, the model—i.e., the chatbot—provides more accurate responses when interacting with users (e.g., if a user asks the same question, the chatbot provides a response based on the new/updated answer, etc.). It is worth noting that Bentitou’s teaching above also demonstrates the fact above. Accordingly, Applicant’s repetitive attempt to show an alleged technological improvement, despite relying merely on such functions of the existing computer technology, is once again not persuasive. In contrast, a new—or an improved—AI technology (AI innovation) typically involves at least one advanced technological feature.
Moreover, it is readily apparent—at least to one skilled in the art—that none of the current claims is analogous to the case of Ex parte Desjardins. In particular, Desjardins is considered to provide a technological improvement since it reduces not only the storage space required to manage the data being processed, but also the complexity of the system. This is because the system is implementing an improved ML training scheme, which allows the ML model to learn new tasks while protecting knowledge about previous tasks (thereby overcoming “the problem of ‘catastrophic forgetting’ encountered in continual learning systems”); and this enables the same ML model to learn multiple tasks.
In contrast, Applicant’s current claims, including the original disclosure, are quite the opposite of the case of Desjardins. For instance, unlike Desjardins, neither the current claims nor the original disclosure contemplates a particular manner of training any of the ML models, much less an advanced training methodology. The fact above is already sufficient to confirm the lack of technological improvement when compared to the existing AI or ML technology. Moreover, again unlike the case of Desjardins, each of the current claims—including the original disclosure—is executing multiple ML models in order to implement the claimed/disclosed chatbot model (e.g., see [0125]; also [0130] to [0140], etc.). Thus, the claimed—and the disclosed process—not only demands a massive storage space, but also increases the burden that the system’s processor is facing (i.e., it adds additional complexity without any technological benefit).
The observation above demonstrates that neither the current claims nor the original disclosure is even remotely analogous to the case of Desjardins. Consequently, Applicant’s attempt to substantiate an alleged technological improvement, while misapplying the case of Desjardins to the current claims, is once again not persuasive.
Note also that Applicant’s assertion directed to Step 2B is also not persuasive. In particular, except for the generic assertion, “under the second step (2B) of Alice the ordered combination of elements in the independent claims are sufficient to ensure that the claim amounts to significantly more than the judicial exception”, Applicant fails to present any specific rationale and/or evidence to substantiate the above assertion. Nevertheless, the analysis presented above, including the findings under Step 2B, already demonstrate that the claimed (and the disclosed) system/method is directed to the conventional and generic arrangement of the additional elements. So far, Applicant fails to challenge—much less negate—the Office’s findings.
Thus, at least for the reasons discussed above, the Office concludes that none of the current claims, when considered as a whole, implements an inventive concept that amounts to “significantly more” than an abstract idea.
Claim Rejections - 35 USC § 103
5. The following is a quotation of 35 U.S.C.103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102 of this title, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating
obviousness or nonobviousness.
Note that the one or more citations (paragraphs or columns) presented in this office action regarding the teaching of a cited reference(s) are exemplary only. Accordingly, such citation(s) are not intended to limit/restrict the teaching of the reference(s) to the cited portion(s) only. Applicant is required to evaluate the entire disclosure of each reference; such as additional portions that teach or suggest the claimed limitations.
● Claims 1-6, 8-14 and 16-22 are rejected under 35 U.S.C.103 as being unpatentable over Voyles 2024/0012841 in view of Ramsey 2023/0419287.
Regarding each of claims 1, 9 and 11, Voyles teaches the following claimed limitations: an apparatus comprising: a memory; and a processor coupled to the memory, the processor (“a method”, per claim 9; and a “computer-readable storage medium . . . cause the processor to perform”, per claim 17) (see [0052]; FIG 2: e.g., a computer-based system/method for monitoring and detection of a concept drift related to one or more trained models; such as chatbots; and accordingly, such computer-based system already incorporates a processor and a memory): receive an identifier of a rule based on a request, identify a subset of vectors within a vector database that include an identifier of the rule stored within metadata of the subset of vectors ([0053], [0054]; [0055] lines 1-30: e.g., the system receives at least one request from the user; such as, a query in the form of a natural language; and the system extracts—from the received query—pieces of information representing the various aspects of the user’s query as tokens; such as, a token representing the content of the query, a token representing the context of the query, etc., and thereby, the system creates one or more vectors that are representative of the user’s query. Subsequently, the system attempts to identify, from a plurality of intent classifications stored in its database, one or more intent classifications relevant to the user’s query based on the tokens generated above. Thus, one of the extracted pieces of information or token, such as, the token identifying the context—i.e., the intent/concept—corresponds to the identifier of the rule that the system has received based on the request; whereas, the one or more intent classifications, which the system identifies from its database based on one or more of the tokens, correspond to the subset of vectors within a vector database that include the identifier of the rule stored within metadata of the subset of vectors), detect a drift between current chat content associated with the rule and the rule itself based on execution a [machine-learning] model on the subset of vectors and text content of the rule ([0055] lines 30-37; [0056]; [0061]; [0079]: e.g., the system already implements one or more machine-learning models, including a drift detection system; and thereby, it determines a drift/evolution of the intent/concept, based on comparing the intent/concept raised in the chat with the intent/concept specified per one or more of intent/concept classifications identified from its database. The drift detection system detects a drift when it determines that no relevant intent/concept classification exists that corresponds to the issue noted in the chat content; such as, issue specific to: (i) the tolls that cars pay when traveling over the Golden Gate Bridge, and/or (ii) the new bridge that connects the city of San Francisco to a part of Oakland, etc. The above indicates the process of detecting a drift between current chat content associated with the rule and the rule itself based on execution a [machine-learning] model on the subset of vectors and text content of the rule. Also see below the discussion presented under “Response to Arguments” for further detail), update parameters of a chatbot model based on text content described in the rule and the detected drift, and execute the chatbot model with the updated parameters to output a chatbot response during a chat session ([0080]; [0087]; [0097] lines 1-7; [0100]; [0102] lines 1-20; [0103] lines 10-22; [0105]: e.g., when a drift/evolution is detected as discussed above; namely, by comparing the intent/concept raised in the chat with the intent/concept specified in the classification—i.e., the “text content described in the rule”, the chatbot model is retrained automatically to address the detected drift/evolution; and this includes: updating its training files based on the detected concept drift/evolution, including incorporating one or more new intent classifications based on the detected concept drift/evolution, etc.; and subsequently, based on the update above, the chatbot provides one or more pertinent responses to the user during a chat session—such as: providing a correct response to the user about the new toll policy applied to the Golden Gate Bridge, [0079]; providing a correct response to the use about the new bridge that connects the city of San Francisco to part of Oakland, [0080], etc. Also see below the discussion presented under “Response to Arguments” for further detail).
Although Voyles already contemplates the implementation of one or more machine-learning models as discussed above, Voyles does not expressly describe the use of an artificial intelligence (AI) model for detecting the drift as discussed above.
However, Ramsey discloses a system that allows a user/customer to interact with an automated chatbot, wherein the chatbot implements an AI model to provide pertinent response to the user; and the system also incorporates an intent-extraction system that implements an AI model; wherein the intent-extraction system evaluates the interaction between the chatbot and the user in order to determine whether the response, which the chatbot is providing to the user, is pertinent to user’s query, etc., and; and thereby, the intent-extraction system retrains the chatbot based feedback collected during one or more interactions, including: one or more negative feedback that points out the detected incorrect response(s), one or more resolutions that must be applied to correct the incorrect response(s), etc. ([0054]; [0055]; [0076]; [0077]; [0095]).
Accordingly, given the above teaching, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the invention of Voyles in view of Ramsey; for example, by implementing one or more AI models to upgrade the system’s algorithm(s), wherein each of the drift detection system and the chatbot is supplemented with an AI model; and wherein, based on the drift/evolution detected regarding the concept in the chat, the drift detection system commands one or more of the AI models to customize relevant training data to address the detected drift/evolution (e.g., based on the analysis of: interactions of the chatbot, interactions gathered from other chatbots and/or users, one or more negative feedback concerning inaccurate response(s), one or more resolutions that must be applied to correct the detected incorrect response(s), new or updated policy, etc.); and accordingly, the system would have a supplemental AI-based scheme for retraining and updating the chatbot and/or its model; and this further minimizes the chatbot’s chance of providing an irrelevant and/or inaccurate response regarding one or more issues that a user(s) is raising during interaction with the chatbot.
Regarding each of claim 2, 10 and 18, Voyles in view of Ramsey teaches the claimed limitations as discussed above per claims 1, 9 and 17 respectively.
The limitation directed to the process of determining a difference between the current chat content of the rule and the text content described within the rule, and generating/updating descriptive content about the difference based on execution of a second AI model, is already addressed above per the modification discussed with respect to claims 1, 9 and 17. In particular, besides implementing one or more AI models (i.e., a second AI model), the training data, which is utilized to retrain the chatbot, already includes a negative feedback that points out the detected inaccurate response during the chat. Accordingly, the AI model already generates text data regarding the difference between (a) the current chat content that relates to the rule and (b) the text of the rule (note also that the modification discussed above applies to each of claims 2, 10 and 18).
Regarding each of claim 3, 11 and 19, Voyles in view of Ramsey teaches the claimed limitations as discussed above per claims 1, 9 and 17 respectively.
Here also the limitation per each of the above claims, which is directed to the process of identifying—from the current chat content—a step of the rule being performed incorrectly, and generating/updating a description of how to correctly perform the step based on the execution of a second AI model, is already addressed per the modification discussed above with respect to claims 1, 9 and 17. Here also, besides implementing one or more AI models (which includes a second AI model), the system also (i) identifies the incorrect response that the chatbot has provided during the interaction (i.e., a step of a rule performed incorrectly), and also (ii) provides one or more resolutions that must be applied in order to correct the detected incorrect response (i.e., generating a description of how to correctly perform the incorrect step detected above).
Note that the modification discussed per claims 1, 9 and 17 also applies to each of claims 3, 11 and 19.
Regarding each of claim 4, 12 and 20, Voyles in view of Ramsey teaches the claimed limitations as discussed above per claims 1, 9 and 17 respectively.
Although the modification above does not expressly address the limitation directed to identifying a step of a rule being omitted from the chat content and generating/updating a description of the step of the rule being omitted, Ramsey already teaches that the chatbot interacts with the user by carrying out proper interaction steps; such as, the chatbot first provides a welcoming/greeting phrase, which is followed by a phrase that politely inquires the user’s intention/goal, etc. ([0075]).
Of course, besides implementing one or more AI models (see above the discussion per each of claim 1, 9 and 17), which already indicates the use of a second AI model, Ramsey also teaches that the intent-extraction system evaluates whether one or more of the responses, which the chatbot is providing to the user, is pertinent to the user’s query, etc., and; and thereby, the intent-extraction system retrains the chatbot based feedback collected during one or more interactions, including: one or more negative feedback that points out the detected incorrect response(s), one or more resolutions that must be applied to correct the incorrect response(s), etc. ([0054]; [0055]; [0076]; [0077]; [0095]).
Accordingly, given the above teaching, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to further modify Voyles’s system by updating the system’s algorithm; so that the drift detection system further evaluates, based on one or more relevant templates in its database, whether the chatbot is generating one or more of the phrases according to one or more desired sequences—such as, generating a greeting phrase, which should be followed by a polite inquiry, etc.; and when the drift detection system detects a drift/evolution in the chat content—such as, detecting that the chatbot is inquiring the user without first greeting the user, the drift detection system commands one or more of the AI models to customize further training data based on: the analysis of the chatbot’s interactions, interactions gathered from other chatbots and/or users, etc., wherein such training data also includes one or more negative feedback that points out the incorrect sequences or steps of interactions, one or more resolutions that must be applied to correct the incorrect sequences/steps of interactions, etc., so that the chatbot is retrained based on such training data; so that, the chatbot’s chance of omitting or skipping one or more desired interaction sequences/steps is minimized; and this helps the user to be more comfortable during interaction.
Regarding each of claim 5 and 13, Voyles in view of Ramsey teaches the claimed limitations as discussed above per claims 1, 9 respectively.
Voyles further teaches, retrieving the text content of the rule from a document stored within a storage device ([0046]; [0055] lines 24-30: e.g., the system already stores intent classifications in its database, which the system uses to retrieve a relevant intent/concept that is relevant to the user’s query in the chat. Accordingly, the above indicates the process of retrieving the text content of the rule from a document stored within the storage device).
Regarding each of claim 6 and 14, Voyles in view of Ramsey teaches the claimed limitations as discussed above per claims 1, 9 respectively.
Voyles further teaches, receive an identifier of a geographic location associated with the rule; and identify the subset of vectors based on a comparison of the geographic location and the metadata of the subset of vectors ([0054] to [0056]: e.g., the system is identifying one or more relevant intent/concept classifications from its database, based on the interaction that the user is making with the chatbot; and wherein such interaction involves phrases and/or numbers that form the query and/or issue that the user is making. For instance, when the user provides the query, “what is the name of the famous bridge in San Francisco”, the system identifies the geographic identifier, namely, San Francisco; and thereby, it attempts to identify one or more intent/content classifications that relevant to the above geographic location. Note also that the intent classifications are stored as vectors).
Regarding each of claim 8 and 16, Voyles in view of Ramsey teaches the claimed limitations as discussed above per claims 1, 9 respectively.
Voyles further teaches that the identifying process comprises querying the vector database with the identifier of the rule to identify the subset of vectors within the vector database ([0053], [0054]; [0055] lines 1-30: e.g., as already discussed per each of claims 1 and 9, the system extracts—from the query it receives from the user—pieces of information representing the various aspects of the user’s query as tokens; and thereby, creates one or more vectors representative of the user’s query. Subsequently, the system identifies from a plurality of intent/concept classifications, which are stored as vectors in the database, one or more intent classifications that are relevant to the user’s query based on the tokens generated above. Thus, the identifying process already comprises querying the vector database with the identifier of the rule to identify the subset of vectors within the vector database).
Regarding claim 21, Voyles in view of Ramsey teaches the claimed limitations as discussed above per claim 1.
The limitation, “the processor is further configured to generate a digital document with a description of actions for correctly implementing the rule embedded therein”, is already addressed per the modification discussed with respect to claim 1.
In particular, responsive to identifying the incorrect responses that the chatbot has provided during the chat interaction, the drift detection system commands one or more of the AI models to generate training data based on collected interactions, which includes one or more resolutions that must be applied in order to correct the detected incorrect response (i.e., the digital training materials already comprise actions for correctly implementing the rule); and wherein the chatbot is retrained based on the training data generated above; so that the chatbot’s chance of providing an irrelevant and/or inaccurate response regarding one or more issues that the user is raising during interaction with the chatbot (i.e., the processor generates a digital document with a description of the actions for correctly implementing the rule embedded therein).
Regarding claim 22, Voyles in view of Ramsey teaches the claimed limitations as discussed above per claim 1.
Voyles further teaches, the processor is further configured to display a warning on a graphical user interface (GUI) of a software application in response to detection of the drift ([0097]; [0098]: e.g., once detecting a drift, the drift detection system generates a drift summary report to an administrator or a developer; wherein the report identifies the concept drift/evolution identified; and wherein the report is displayed via a user interface. The above indicates that the processor is already configured to display a warning on a graphical user interface (GUI) of a software application in response to detection of the drift).
Response to Arguments.
6. Applicant’s arguments have been fully considered (the arguments filed on 04/17/2026); however, the arguments are not persuasive at least for the following reasons:
Firstly, while attempting to summarize Voyles based on the teaching of some of its paragraphs (i.e., [0050], [0051]), Applicant asserts that “Voyles fails to render obvious the features of Claim 1, because Voyles fails to describe or suggest, ‘update parameters of a chatbot model based on text content described in the rule and the detected drift, and execute the chatbot model with the updated parameters to output a chatbot response during a chat session.’ . . . Instead, the system of Voyles monitors a trained model by comparing outputs generated from different datasets to determine whether concept drift has occurred, but the system merely detects statistical differences between datasets and may flag or classify drift conditions . . . fails to describe detecting a drift between chat content and rule text, nor updating a chatbot model based on such a drift. Instead, the drift detection system operates as a monitoring mechanism that analyzes changes in model outputs or dataset distributions, rather than modifying the internal parameters of a conversational model using textual rule content as recited in Claim 1” (emphasis added).
However, Applicant appears to fail to properly construe the teaching of Voyles as applied to the current claims. In particular, Applicant appears to fail to recognize that the so-called “text content described in the rule”, as currently claimed, is itself data (e.g. a document) stored in a database, which the model is evaluating in order to determine whether there is a drift or not. In this regard, as correctly pointed out in the previous office action (also the current office action), Voyles already implements various machine-learning model(s), including a drift detection system; and accordingly, by comparing the intent/concept raised in the chat with an intent/concept specified in at least one intent/concept classification, the system determines whether there is a drift/evolution of the intent/concept (again see [0055] lines 30-37; [0056]; [0061]; [0079]). In this regard, the classification above represents the document that relates to the rule; and the intent/concept that that classification specifies corresponds to the text content of the rule. For instance, if there is a new policy regarding tolls that cars must pay when traveling over the Golden Gate Bridge, a user may ask the chatbot a question about the new toll fee (e.g., “when does the city of San Francisco starts charging the new toll fee?”). Accordingly, once identifying the corresponding classification/rule (e.g., one that relates to toll fees on Golden Gate Bridge), the model attempts to find any information that relates to the new toll fee that the user is requesting (i.e., the chatbot analyzes the text content to determine any description related to new toll fees, etc.). If such information does not exist, it means there is a drift (i.e., “a drift between current chat content associated with the rule and the rule itself”).
Of course, once such drift is detected as discussed above, the chatbot is retrained automatically to address the detected drift/evolution; and such retraining includes: updating its training files based on the detected concept drift/evolution, including incorporating one or more new intent classifications based on the detected concept drift/evolution, etc. (again see [0080]; [0087]; [0097] lines 1-7; [0100]; [0102] lines 1-20; [0103] lines 10-22; [0105]). It is worth noting that “updating” the training files already includes adding new/updated information (e.g., adding text that specifies the new toll fee, the effective date for the new toll fee, etc.). Of course, when the model is retrained based on such “updated” training files, the training process involves updating one or more parameters of the model since the model is incorporating or modifying various attributes (e.g., the new toll fee; the effective date for the new toll fee, etc.). Thus, once trained in such manner, the chatbot is now able to provide the user(s) with the correct response about the new toll fee policy concerning the Golden Gate Bridge ([0079]).
The observation above confirms that Voyles does teach the limitation, “update parameters of a chatbot model based on text content described in the rule and the detected drift, and execute the chatbot model with the updated parameters to output a chatbot response during a chat session”, as currently claimed. Consequently, Applicant’s arguments are not persuasive.
Applicant further asserts that “Voyles also fails to describe executing a chatbot model with updated parameters to output a chatbot response during a chat session. Voyles primarily discusses monitoring performance of a trained model and detecting concept drift in datasets used for training or evaluation. The system described in Voyles focuses on identifying changes in statistical distributions of data rather than operating an interactive chatbot system that produces responses during a chat session” (emphasis added).
However, besides the attempt made to simply dismiss Voyles’ teaching, Applicant also appears to mischaracterize the crux of the operation of the system. If Applicant assumes that Voyles is not executing the chatbot model with updated parameters, which allow the chatbot to provide proper responses to a user during a chat session, then what is the point of retraining the model? Of course, Applicant fails to address the critical issue above. Instead, Applicant appears to make various conclusory assertions that fail to address the purpose of retraining the model. In contrast, Voyles teaches that retraining of the model is initiated when a drift is detected; such as, when the chatbot is facing difficulty in providing the correct response to a query that the user is making during a chat session (e.g., see [0080], emphasis added),
“. . . the city of San Francisco may decide to construct a new bridge connecting the city to a different part of Oakland, prompting an increase in user queries to the chatbot about the newly-constructed bridge. The training dataset for the chatbot will not have included any user input data labeled with an intent classification relating to the new bridge. Initially, the chatbot may have difficulty categorizing queries about the newly constructed bridge, and may loosely associate the queries with the Golden Gate Bridge intent classification, causing the chatbot to output incorrect or misleading information to travelers using the chatbot. As the chatbot continues to receive additional queries about the newly constructed bridge, the user queries about the bridge may diverge from the user queries about the Golden Gate Bridge, as the queries will include additional or different information that is clearly not relevant to the existing Golden Gate Bridge intent classification. Eventually, the model may be retrained to accommodate a new intent classification associated with the new bridge (e.g., by manually retraining or updating the model with new labeled user input data associated with the bridge, or by an automatic retraining system of the model).”
Accordingly, it is evident from the excerpt above that the chatbot is conducting a chat session with the user, wherein the chatbot is attempting to answer the user’s question about the newly-constructed bridge. Although the chatbot is making an attempt to answer the user’s query above, its current classification (i.e., rule) does not address the newly-constructed bridge even if the rule has some information (text content) about old bridges in the city of San Francisco. Of course, due to its lack of information about the newly-constructed bridge (i.e., there is a drift between the current chat content associated with the rule and the rule itself), the chatbot starts to “output incorrect or misleading information” to the user. Thus, when such drift its detected, the model is retrained in order to address the drift, “. . . the model may be retrained to accommodate a new intent classification associated with the new bridge (e.g., by manually retraining or updating the model with new labeled user input data associated with the bridge, or by an automatic retraining system of the model)” ([0080], emphasis added). Of course, such retraining is essentially the process of updating one or more parameters of the model (e.g., incorporating one or more attributes about the newly-constructed bridge, etc.); and this enables the updated chatbot to provide correct responses to a user during a chat session (i.e., the updated chatbot provides the correct response to a user who may raise a question about the newly-constructed bridge, etc.).
Accordingly, it is evident—at least to a person skilled in the art—that Voyles does execute a chatbot model with updated parameters in order to output a chatbot response during a chat session. Consequently, Applicant’s arguments are not persuasive.
Thus, at least for the reasons discussed above, the Office concludes that the current claims are still obvious over the prior art.
Conclusion
Applicant’s amendment necessitated the new grounds of rejection presented in this final office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filled within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BRUK A GEBREMICHAEL whose telephone number is (571) 270-3079. The examiner can normally be reached from 7:00 AM - 3:00 PM.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, PETER VASAT can be reached on (571) 270-7625. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/BRUK A GEBREMICHAEL/Primary Examiner, Art Unit 3715
Prosecution Insights