Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
DETAILED ACTION
This office action is in response to correspondence 01/15/26 regarding application 18/412,671, in which claims 1, 8, 9, 13, 14, 20, 21, and 25 were amended. Claims 1-25 are pending in the application and have been considered.
Response to Arguments
The amendments to the specification overcome the objection for a minor informality, and so the objection to the specification is withdrawn.
The amendments to the Abstract overcome the objection for containing legal phraseology to be avoided, and so the objection to the Abstract is withdrawn.
The amendments to claims 1, 13, and 25 overcome the objections for minor informalities, and so they are withdrawn. As Applicant notes correctly on page 7, the objection to claim 9 was intended for claim 13, and no objection for claim 9 was intended.
The amendments to claims 8, 9, 14, 20, and 21 overcome the 35 U.S.C. 112(b) rejections as being indefinite, and so the rejections are withdrawn. Specifically, Applicant’s amendments address the lack of antecedent basis issues identified in claims 8, 14, and 20, and address the claim language identified as not making sense because of missing claim language in claims 9 and 21.
Amended claims 1, 13, and 25 overcome the 35 U.S.C. 101 rejections of claims 1-25 as being directed to an abstract idea without significantly more. Specifically, each of independent claims 1, 13, and 25 now recite deploying artificial intelligence of at least one conversational language model trained on a large corpus of text and using deterministic or statistical techniques for text classification or inspection, which cannot be practically performed as a mental process.
Since the 35 U.S.C. 101 rejections of claims 1-25 are withdrawn as noted above, Applicant’s arguments on pages 8-14 regarding these rejections are moot.
Applicant’s arguments on pages 14-20 regarding the 35 U.S.C. 103 rejections based on Wooters in view of Kelkar have been considered but are moot in view of the new grounds for rejection based in part on the newly discovered reference to Sandrew (US 11990139), necessitated by Applicant’s amendments. However, for clarity of the record and compact prosecution, the examiner offers the following response to Applicant’s arguments regarding Wooters, which is still relied upon in the claim rejections:
On pages 14-15, regarding Wooters, Applicant argues that “Wooters is fundamentally concerned with testing the performance of a chatbot, not with evaluating textual content itself.” In response, as those familiar with chatbots would have been aware, chatbots respond with textual replies, and so testing the performance of a chatbot necessarily requires evaluating the textual content of the chatbot’s replies. Indeed, this is the approach taken in Wooters.
In response to Applicant’s arguments on page 15 regarding Wooters not disclose “inspection questions”, the rejection does not (and the previous rejection incorporating Kelkar did not) rely upon Wooters to teach this claim limitation. One cannot show non-obviousness by attacking references individually when the rejection is based upon a combination.
On page 15, Applicant further argues “Wooters' "recorded transcript" is merely a byproduct of the testing process. Wooters does not teach generating queries by combining inspection questions with parts of textual content for analytical purposes. The so-called "Input" shown in Table 1 of Wooters is an internal formatting example for a language model prompt; it is not a systematic generation of inspection queries derived from textual content segments and inspection questions as recited in claim 1.” In response, the examiner respectfully disagrees. As Applicant acknowledges, the Table 1 of Wooters is input to a language model prompt, i.e. a query for the language model to evaluate the chatbot’s response. Wooters discloses prompting the language model with the test scenario which includes the question, the chatbot’s reply, and the expected reply. Wooters does this for a whole set of scenarios to evaluate the percentage of conversations a bot can successfully resolve (see [0048]). Wooters therefore is considered to fairly disclose generating a plurality of queries, each query formed by combining at least one part of the textual content and one of the plurality of questions.
On page 15, Applicant further argues “… the conversational language model in Wooters is used to determine whether an expected outcome was achieved in a test scenario. That determination is binary or categorical with respect to chatbot task success. Claim 1, by contrast, requires generating a plurality of inference values corresponding to inspection questions applied to textual content, and then processing those inference values to generate an evaluation of the textual content itself. This distinction is not semantic; it goes to the core purpose and operation of the system.”
In response, the examiner does not see the alleged distinction. Wooters discloses determining if an expected outcome was achieved based on the test scenario + chatbot response + expected response prompt, [0048], using a generative LLM performing an inference to perform this determination. Moreover, the determination is categorized, i.e. processed by adding a textual label, for example, of “resolved”, “not-resolved”, or unclear, based on the determinations by machine learning model 120, [0048]-[0049]).
Applicant’s arguments on pages 15-17 regarding Kelkar are moot in view of the new grounds for rejection, which does not rely upon Kelkar.
On page 18, Applicant further argues that the amended claims are distinct from Wooters because they now recite that the method is performed in real time, during a conversation with a user, whereas Wooters is directed to offline or batch style testing. In response, it is noted that Wooters specifically describes first machine learning model 110 is configured as a simulated user bot for engaging in conversation with the chatbot 200, [0040], for a series of conversations, [0048], carried out in real time by navigating to the web-site that is hosting the chatbot and clicking on the “Chat” button, and carrying out the conversation just as a human would do, [0045]. In other words, at the very least, the “conversations” conducted by Wooters for the purposes of testing the chat-bot are fairly considered “real-time conversations”. With regard to “conversation with a user”, these arguments are moot in view of the new grounds for rejection based in part on Sandrew.
Applicant’s argument on pages 18-19 regarding Kelkar is moot in view of the new grounds for rejection, which does not rely upon Kelkar.
On page 19, Applicant argues that the conversation in Wooters is between two LMs, not a LM and a user. In response, this argument is moot in view of the new grounds for rejection based in part on Sandrew.
On page 19, Applicant further argues that “the amended claim further requires generating a plurality of queries, each query formed by combining at least one part of the textual content and one of the plurality of inspection questions. This query-generation mechanism is central to the invention and is absent from the asserted prior art. In Wooters, questions are posed to the chatbot to elicit responses; they are not inspection questions applied to portions of user-provided content. The "Input" examples in Wooters are static prompt formats used for outcome evaluation and do not represent a systematic combination of inspection questions with content segments to generate analytical queries. Kelkar does not disclose inspection questions at all, let alone their structured combination with portions of textual content.”
In response, it is noted that Wooters discloses deploying artificial intelligence of at least one conversational language model trained on a large corpus of text and using deterministic or statistical techniques for text classification or inspection because in Wooters, language models may be LLMs such as GPT-4, which is trained on a large corpus of text and uses deterministic and statistical techniques for text classification or inspection, [0040], and Wooters specifically describes the LLM providing evaluations which include a label which may be “resolved”, “not-resolved”, or unclear, based on the determinations by machine learning model 120, [0048]-[0049]), i.e. classification.
On pages 19-20, Applicant further argues “The amended claim further requires generating a plurality of inference values by feeding the generated queries to the conversational language model, and then processing the plurality of inference values to produce an evaluation of the textual content. This multi-stage inference pipeline is distinct from the binary or categorical labelling performed in Wooters, which merely determines whether an expected outcome was achieved. In contrast, the claimed inference values are derived from inspection queries applied to content segments and are aggregated or processed to form an evaluation of understanding.” In response, as noted earlier with regard to Wooter’s inference and processing, Wooters discloses determining if an expected outcome was achieved based on the test scenario + chatbot response + expected response prompt, [0048], using a generative LLM performing an inference to perform this determination. The determination is categorized, i.e. processed by adding a textual label, for example, of “resolved”, “not-resolved”, or unclear, based on the determinations by machine learning model 120, ([0048]-[0049]).
The arguments on page 20 regarding independent claims 13 and 25, as well as dependent claims 2-12 and 14-24 are similar to those addressed above, and are either not persuasive or moot in view of the new grounds for rejection for similar reasons.
Claim Rejections - 35 USC § 103
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.
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.
Claims 1-7, 12-19, 24, and 25 are rejected under 35 U.S.C. 103 as being unpatentable over Wooters et al. (US 20250029110) in view of Sandrew (US 11990139).
Consider claim 1, Wooters discloses a method for increasing accuracy of a chatbot or of a virtual assistant in evaluating understanding in a textual content (testing language understanding abilities of a chatbot, [0004], by evaluating its transcripts, [0006], [0008], to identify a suitable configuration that improves accuracy, [0037]), comprising:
in real-time conversations (first machine learning model 110 is configured as a simulated user bot for engaging in conversation with the chatbot 200, [0040], for a series of conversations, [0048], carried out in real time by navigating to the web-site that is hosting the chatbot and clicking on the “Chat” button, and carrying out the conversation just as a human would do, [0045]):
accessing a storage to obtain a plurality of questions (test scenarios including questions to be answered by the chatbot, [0043], in second database 130, [0047], Fig. 1 element 130);
acquiring a textual content, using a virtual human interaction agent (first machine learning model starts a conversation with chatbot 200, a virtual human interaction agent, by asking a question and obtaining an answer, [0043-0046], and recorded transcript is stored in second database 130, [0047]);
using at least one processing circuitry (the systems implemented in CPUs, [0095]) deploying artificial intelligence of at least one conversational language model trained on a large corpus of text and using deterministic or statistical techniques for text classification or inspection (language models may be LLMs such as GPT-4, which is trained on a large corpus of text and uses deterministic and statistical techniques for text classification or inspection, [0040]), for generating:
a plurality of queries, each query formed by combining at least one part of the textual content and one of the plurality of questions (the recorded conversations and expected outcomes are provided to second machine learning model 120 as a request to evaluate whether the question was resolved, [0047-0049], the “Input:” portion shown in Table 1, [0050], considered an example query for one of the scenarios);
a plurality of inference values, each by feeding one of the plurality of queries to the at least one conversational language mode (second machine learning model reviews the transcript of the conversation that occurred between the first machine learning model and the chatbot, and determines if the expected outcome was achieved, [0048], which is a generative LLM, i.e. conversational language model such as GPT-4 or Claude, [0040]); and
at least one evaluation of the textual content by processing the plurality of inference values (the evaluation includes a label which may be “resolved”, “not-resolved”, or unclear, based on the determinations by machine learning model 120, [0048]-[0049]).
Wooters does not specifically mention during a conversation with a user, acquiring a textual content provided by the user, and inspection questions.
Sandrew discloses during a conversation with a user, acquiring a textual content provided by the user (during a mock interview, researcher 210, a user, has a dialog with AI system and types response to questions, Col 12 lines 9-27) and inspection questions (the generated questions asked by AI system used to assess researcher’s preparedness and knowledge, i.e. inspect the breadth and depth of researcher’s knowledge for given selected topics, Col 12 lines 15-34).
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 Wooters by during a conversation with a user, acquiring a textual content provided by the user using inspection questions in order to assist with reviewing or reinforcing material that a person desires to gain a command over, as suggested by Sandrew (Col 1 lines 29-33). Doing so would have led to predictable results of ensuring a user’s comprehension of a topic, as suggested by Sandrew (Col 1 lines 23-25). The references cited are analogous art in the same field of natural language processing.
Consider claim 13, Wooters discloses a system for increasing accuracy of a chatbot or of a virtual assistant in evaluating understanding in a textual content (a system for identifying a suitable configuration that improves accuracy, [0037]) comprising a storage and at least one processing circuitry (second database 130, [0047], and system including processor, [0027]) configured to:
in real-time conversations (first machine learning model 110 is configured as a simulated user bot for engaging in conversation with the chatbot 200, [0040], for a series of conversations, [0048], carried out in real time by navigating to the web-site that is hosting the chatbot and clicking on the “Chat” button, and carrying out the conversation just as a human would do, [0045]):
access the storage to obtain a plurality of questions (test scenarios including questions to be answered by the chatbot, [0043], in second database 130, [0047], Fig. 1 element 130);
acquire a textual content, using a virtual human interaction agent (first machine learning model starts a conversation with chatbot 200, a virtual human interaction agent, by asking a question and obtaining an answer, [0043-0046], and recorded transcript is stored in second database 130, [0047]);
use at least one processing circuitry (the systems implemented in CPUs, [0095]) deploying artificial intelligence of at least one conversational language model trained on a large corpus of text and using deterministic or statistical techniques for text classification or inspection (language models may be LLMs such as GPT-4, which is trained on a large corpus of text and uses deterministic and statistical techniques for text classification or inspection, [0040]), for generating:
a plurality of queries, each query formed by combining at least one part of the textual content and one of the plurality of questions (the recorded conversations and expected outcomes are provided to second machine learning model 120 as a request to evaluate whether the question was resolved, [0047-0049], the “Input:” portion shown in Table 1, [0050], considered an example query for one of the scenarios);
a plurality of inference values, each by feeding one of the plurality of queries to the at least one conversational language mode (second machine learning model reviews the transcript of the conversation that occurred between the first machine learning model and the chatbot, and determines if the expected outcome was achieved, [0048], which is a generative LLM, i.e. conversational language model such as GPT-4 or Claude, [0040]); and
at least one evaluation of the textual content by processing the plurality of inference values (the evaluation includes a label which may be “resolved”, “not-resolved”, or unclear, based on the determinations by machine learning model 120, [0048]-[0049]).
Wooters does not specifically mention during a conversation with a user, acquiring a textual content provided by the user, and inspection questions.
Sandrew discloses during a conversation with a user, acquiring a textual content provided by the user (during a mock interview, researcher 210, a user, has a dialog with AI system and types response to questions, Col 12 lines 9-27) and inspection questions (the generated questions asked by AI system used to assess researcher’s preparedness and knowledge, i.e. inspect the breadth and depth of researcher’s knowledge for given selected topics, Col 12 lines 15-34).
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 Wooters by during a conversation with a user, acquiring a textual content provided by the user using inspection questions for reasons similar to those for claim 1.
Consider claim 25, Wooters discloses one or more computer program products comprising instructions for increasing accuracy of a chatbot or of a virtual assistant in evaluating understanding in a textual content (identifying a suitable configuration that improves accuracy, [0037]), wherein execution of the instructions by one or more processors of a computing system (computer executing instructions using processor from memory, [0087]) is to cause a computing system to:
in real-time conversations (first machine learning model 110 is configured as a simulated user bot for engaging in conversation with the chatbot 200, [0040], for a series of conversations, [0048], carried out in real time by navigating to the web-site that is hosting the chatbot and clicking on the “Chat” button, and carrying out the conversation just as a human would do, [0045]):
access the storage to obtain a plurality of questions (test scenarios including questions to be answered by the chatbot, [0043], in second database 130, [0047], Fig. 1 element 130);
acquire a textual content, using a virtual human interaction agent (first machine learning model starts a conversation with chatbot 200, a virtual human interaction agent, by asking a question and obtaining an answer, [0043-0046], and recorded transcript is stored in second database 130, [0047]);
use at least one processing circuitry (the systems implemented in CPUs, [0095]) deploying artificial intelligence of at least one conversational language model trained on a large corpus of text and using deterministic or statistical techniques for text classification or inspection (language models may be LLMs such as GPT-4, which is trained on a large corpus of text and uses deterministic and statistical techniques for text classification or inspection, [0040]), for generating:
a plurality of queries, each query formed by combining at least one part of the textual content and one of the plurality of questions (the recorded conversations and expected outcomes are provided to second machine learning model 120 as a request to evaluate whether the question was resolved, [0047-0049], the “Input:” portion shown in Table 1, [0050], considered an example query for one of the scenarios);
a plurality of inference values, each by feeding one of the plurality of queries to the at least one conversational language mode (second machine learning model reviews the transcript of the conversation that occurred between the first machine learning model and the chatbot, and determines if the expected outcome was achieved, [0048], which is a generative LLM, i.e. conversational language model such as GPT-4 or Claude, [0040]); and
at least one evaluation of the textual content by processing the plurality of inference values (the evaluation includes a label which may be “resolved”, “not-resolved”, or unclear, based on the determinations by machine learning model 120, [0048]-[0049]).
Wooters does not specifically mention during a conversation with a user, acquiring a textual content provided by the user, and inspection questions.
Sandrew discloses during a conversation with a user, acquiring a textual content provided by the user (during a mock interview, researcher 210, a user, has a dialog with AI system and types response to questions, Col 12 lines 9-27) and inspection questions (the generated questions asked by AI system used to assess researcher’s preparedness and knowledge, i.e. inspect the breadth and depth of researcher’s knowledge for given selected topics, Col 12 lines 15-34).
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 Wooters by during a conversation with a user, acquiring a textual content provided by the user using inspection questions for reasons similar to those for claim 1.
Consider claim 2, Wooters discloses using the at least one evaluation, generating at least one prompt (the evaluation includes a reason as to why a judgement was made for the label and a suggestion, used for making improvements to the chatbot configuration, [0049], by forming synthetic questions, i.e. prompts, [0070]-0071]), and using the at least one prompt for querying a user for an additional textual content (synthetic questions are used to query a dataset of “historical” customer questions, i.e. querying a user for additional textual content, [0071]).
Wooters does not specifically mention using the at least one evaluation to generate at least one prompt.
Sandrew discloses generating at least one prompt (the AI system is guided with prompts that indicate the type of inputs the system should analyze and the type of outputs it should generate, Col 8 lines 50-52).
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 Wooters by using the at least one evaluation (i.e. the label, reason) of Wooters to generate the least one prompt in Wooters (i.e. the synthetic question) according to the technique of Sandrew for reasons similar to those for claim 1.
Consider claim 3, the Wooters-Sandrew combination discloses the method of claim 2, and the at least one prompt (see claim 2). Wooters further discloses a suggestion derived from at least one evaluation (evaluating semantic distance of historical questions from synthetic questions, and using the evaluation to suggest historical questions that are unanswerable with the knowledge base, [0074-0076]).
It would have been further obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the Wooters-Sandrew combination by including a prompt comprising a suggestion derived from the at least one evaluation for reasons similar to those for claim 1.
Consider claim 4, the Wooters-Sandrew combination discloses the method of claim 3, and Wooters further discloses the suggestion is deliberately erroneous (evaluating the chatbot with a non-KB answerable, i.e. deliberately erroneous, question to detect an attempt-to-answer label, indicating chatbots failure, i.e. error, [0075-0076]).
Consider claim 5, the Wooters-Sandrew combination discloses the method of claim 4, and Wooters further discloses an unsolvable puzzle question (a non-KB answerable, i.e. puzzle, question [0075-0076], considered puzzling in the sense that it is not answerable using the KB).
Consider claim 6, the Wooters-Sandrew combination discloses the method of claim 3, and Wooters further discloses the suggestion is a request to rephrase a part of the textual content (a historical question with high semantic similarity to the synthetic question, [0070], considered a “rephrase” because it conveys similar semantic meaning with different phrases).
Consider claim 7, Wooters further discloses the at least one evaluation comprising evaluating presence of items from the at least one prompt in the textual content (e.g. labeling the response as “resolved” because the content contains the information about “limit” requested in the question, [0050]).
Consider claim 12, Wooters further discloses the evaluation comprising using the at least one conversational language model to check correctness and completeness of the textual content (e.g. The chatbot provided the correct daily deposit limit information, [0050], i.e. provided a complete and correct answer).
Consider claim 14, Wooters discloses using the at least one evaluation, generating at least one prompt (the evaluation includes a reason as to why a judgement was made for the label and a suggestion, used for making improvements to the chatbot configuration, [0049], by forming synthetic questions, i.e. prompts, [0070]-0071]), and using the at least one prompt for querying a user for an additional textual content (synthetic questions are used to query a dataset of “historical” customer questions, i.e. querying a user for additional textual content, [0071]).
Wooters does not specifically mention using the at least one evaluation to generate at least one prompt.
Sandrew discloses generating at least one prompt (the AI system is guided with prompts that indicate the type of inputs the system should analyze and the type of outputs it should generate, Col 8 lines 50-52).
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 Wooters by using the at least one evaluation (i.e. the label, reason) of Wooters to generate the least one prompt in Wooters (i.e. the synthetic question) according to the technique of Sandrew for reasons similar to those for claim 1.
Consider claim 15, the Wooters-Sandrew combination discloses the method of claim 14, and the at least one prompt (see claim 14). Woot further discloses a suggestion derived from at least one evaluation (evaluating semantic distance of historical questions from synthetic questions, and using the evaluation to suggest historical questions that are unanswerable with the knowledge base, [0074-0076]).
It would have been further obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the Wooters-Sandrew combination by including a prompt comprising a suggestion derived from the at least one evaluation for reasons similar to those for claim 1.
Consider claim 16, the Wooters-Sandrew combination discloses the method of claim 15, and Wooters further discloses the suggestion is deliberately erroneous (evaluating the chatbot with a non-KB answerable, i.e. deliberately erroneous, question to detect an attempt-to-answer label, indicating chatbots failure, i.e. error, [0075-0076]).
Consider claim 17, the Wooters-Sandrew combination discloses the method of claim 16, and Wooters further discloses an unsolvable puzzle question (a non-KB answerable, i.e. puzzle, question [0075-0076], considered puzzling in the sense that it is not answerable using the KB).
Consider claim 18, the Wooters-Sandrew combination discloses the method of claim 3, and Wooters further discloses the suggestion is a request to rephrase a part of the textual content (a historical question with high semantic similarity to the synthetic question, [0070], considered a “rephrase” because it conveys similar semantic meaning with different phrases).
Consider claim 19, Wooters further discloses the at least one evaluation comprising evaluating presence of items from the at least one prompt in the textual content (e.g. labeling the response as “resolved” because the content contains the information about “limit” requested in the question, [0050]).
Consider claim 24, Wooters further discloses the evaluation comprising using the at least one conversational language model to check correctness and completeness of the textual content (e.g. The chatbot provided the correct daily deposit limit information, [0050], i.e. provided a complete and correct answer).
Claims 8, 10, 11, 20, 22, and 23 are rejected under 35 U.S.C. 103 as being unpatentable over Wooters et al. (US 20250029110) in view of Sandrew (US 11990139), in further view of Lam et al. (US 20210056167).
Consider claim 8, Wooters and Sandrew do not, but Lam discloses at least one evaluation is inferring a mental state induced by the at least one prompt (analyze prompts provided to user devices to determine an emotional state of the user, [0033]).
It would have been further obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Wooters and Sandrew such that at least one evaluation is inferring a mental state induced by the at least one prompt in order to make a chatbot appear more human, as suggested by Lam ([0014]), predictably providing a more pleasant and confidence-inducing conversation with a user, as suggested by Lam ([0014]). The references cited are analogous art in the same field of natural language processing.
Consider claim 10, Wooters and Sandrew do not, but Lam discloses at least one evaluation is inferring a mental condition of a user interacting with the virtual human interaction agent (analyze prompts provided to user devices to determine an emotional state of the user, [0033]).
It would have been further obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Wooters and Sandrew such that at least one evaluation is inferring a mental condition of a user interacting with the virtual human interaction agent for reasons similar to those for claim 8.
Consider claim 11, Wooters and Sandrew do not, but Lam discloses the at least one evaluation is inferring an emotional state of a user interacting with the virtual human interaction agent (analyze prompts provided to user devices to determine an emotional state of the user, [0033]).
It would have been further obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Wooters and Sandrew such that the at least one evaluation is inferring an emotional state of a user interacting with the virtual human interaction agent for reasons similar to those for claim 8.
Consider claim 20, Wooters and Sandrew do not, but Lam discloses at least one evaluation is inferring a mental state induced by the at least one prompt (analyze prompts provided to user devices to determine an emotional state of the user, [0033]).
It would have been further obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Wooters and Sandrew such that at least one evaluation is inferring a mental state induced by the at least one prompt in order to make a chatbot appear more human, as suggested by Lam ([0014]), predictably providing a more pleasant and confidence-inducing conversation with a user, as suggested by Lam ([0014]). The references cited are analogous art in the same field of natural language processing.
Consider claim 22, Wooters and Sandrew do not, but Lam discloses at least one evaluation is inferring a mental condition of a user interacting with the virtual human interaction agent (analyze prompts provided to user devices to determine an emotional state of the user, [0033]).
It would have been further obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Wooters and Sandrew such that at least one evaluation is inferring a mental condition of a user interacting with the virtual human interaction agent for reasons similar to those for claim 8.
Consider claim 23, Wooters and Sandrew do not, but Lam discloses the at least one evaluation is inferring an emotional state of a user interacting with the virtual human interaction agent (analyze prompts provided to user devices to determine an emotional state of the user, [0033]).
It would have been further obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Wooters and Sandrew such that the at least one evaluation is inferring an emotional state of a user interacting with the virtual human interaction agent for reasons similar to those for claim 8.
Claims 9 and 21 are rejected under 35 U.S.C. 103 as being unpatentable over Wooters et al. (US 20250029110) in view of Sandrew (US 11990139), in further view of Jang et al. (“Consistency Analysis of ChatGPT”. arXiv:2303.06273v3 [cs.CL] 14 Nov 2023).
Consider claim 9, Wooters and Sandrew do not, but Jang discloses at least one evaluation pertains to consistency of reactions to a plurality of prompts from the virtual human interaction agent, wherein the plurality of prompts are characterized by equivalence (evaluation of semantic consistency, Figure 1, page 4).
It would have been further obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Wooters and Sandrew such that at least one evaluation pertains to consistency of reactions to a plurality of prompts from the virtual human interaction agent, wherein the plurality of prompts are characterized by equivalence in order to ascertain the reliability and trustworthiness of AI systems, as suggested by Jang (page 1) with predictable results of great influence on diverse academic and industrial fields, as suggested by Jang (page 1). The references cited are analogous art in the same field of natural language processing.
Consider claim 21, Wooters and Sandrew do not, but Jang discloses at least one evaluation pertains to consistency of reactions to a plurality of prompts from the virtual human interaction agent, wherein the plurality of prompts are characterized by equivalence (evaluation of semantic consistency, Figure 1, page 4).
It would have been further obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Wooters and Sandrew such that at least one evaluation pertains to consistency of reactions to a plurality of prompts from the virtual human interaction agent, wherein the plurality of prompts are characterized by equivalence for reasons similar to those for claim 9.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this 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 filed 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 nonprovisional extension fee (37 CFR 1.17(a)) 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 mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Jesse Pullias whose telephone number is 571/270-5135. The examiner can normally be reached on M-F 8:00 AM - 4:30 PM. The examiner’s fax number is 571/270-6135.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Andrew Flanders can be reached on 571/272-7516.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Jesse S Pullias/
Primary Examiner, Art Unit 2655 03/30/25
Prosecution Insights