US-12619831-B1 - Systems and methods for generating deliberative committees of age-stratified large language models

Abstract

The techniques described herein relate to systems and methods for generating deliberative committees of age-stratified large language models (LLMs). An example method for generating a dialogue between at least first and second age-stratified LLMs trained to generate simulated responses to questions that respective first/second persons would have provided in a particular life stage comprises receiving a prompt comprising a question, executing the first LLM to generate a first simulated response to the question that the first person would have provided in the first life stage, executing, using the prompt and/or the first simulated response, the second LLM to generate a second simulated response to at least the question that the second person would have provided in the second life stage, and generating, using the response(s), an output representing a third simulated response to the question that the first/second persons would have collectively determined to provide in their particular life stages.

Inventors

• Charles DeLisi

Assignees

• Charles DeLisi

Dates

Publication Date

20260505

Application Date

20250723

Claims (20)

1. 1 . A method for generating a dialogue between at least a first age-stratified large language model (LLM) and a second age-stratified LLM, the first age-stratified LLM trained to generate simulated responses to questions that a first person would have provided in a first life stage of their life, the second age-stratified LLM trained to generate simulated responses to questions that a second person would have provided in a second life stage of their life, the method comprising: using at least one computer hardware processor to perform: receiving, from an electronic device and via at least one communication network, a prompt comprising a question; executing, using the prompt, the first age-stratified LLM to generate a first simulated response to the question that the first person would have provided in the first life stage; executing, using the prompt and the first simulated response, the second age-stratified LLM to generate a second simulated response to at least one of the question or the first simulated response, the second simulated response corresponding to a response that the second person would have provided in the second life stage; and generating, using the first simulated response and the second simulated response, an output representing a third simulated response to the question that the first person and the second person would have collectively determined to provide in their particular life stages, the output to be provided to the electronic device via the at least one communication network.
2. 2 . The method of claim 1 , wherein: executing the first age-stratified LLM comprises executing the first age-stratified LLM in a first virtual machine (VM) having a first private Internet Protocol (IP) address in a software-defined network; and executing the second age-stratified LLM comprises executing the second age-stratified LLM in a second VM having a second private IP address in the software-defined network, and the method further comprising: transmitting, from the first VM and using the first and second private IP addresses, the first simulated response to the second VM over the software-defined network; and transmitting, from the second VM and using the first and second private IP addresses, the second simulated response to the first VM over the software-defined network.
3. 3 . The method of claim 1 , wherein: executing the first age-stratified LLM comprises executing the first age-stratified LLM in a first container having a first private Internet Protocol (IP) address in a software-defined network; and executing the second age-stratified LLM comprises executing the second age-stratified LLM in a second container having a second private IP address in the software-defined network.
4. 4 . The method of claim 3 , further comprising: transmitting, from the first container and using the first and second private IP addresses, the first simulated response to the second container over the software-defined network; and transmitting, from the second container and using the first and second private IP addresses, the second simulated response to the first container over the software-defined network.
5. 5 . The method of claim 3 , wherein a container pod comprises the first container and the second container, and the method further comprising: instantiating a shared volume in the container pod and accessible by the first container and the second container; and wherein: executing the first age-stratified LLM comprises storing the first simulated response in the shared volume; and executing the second age-stratified LLM comprises retrieving the first simulated response from the shared volume and storing the second simulated response in the shared volume.
6. 6 . The method of claim 1 , wherein the first life stage of the first person corresponds to a first age range of the first person's life and is thereby a maturer life stage than the second life stage of the second person that corresponds to a second age range of the second person's life that is less than the first age range, and further comprising: configuring a software-defined network to enable communications from the first age-stratified LLM to the second age-stratified LLM; and transmitting, from the first age-stratified LLM and using the software-defined network, the first output to the second age-stratified LLM.
7. 7 . The method of claim 1 , wherein the second life stage of the second person corresponds to a second age range of the second person's life and is thereby a maturer life stage than the first life stage of the first person that corresponds to a first age range of the first person's life that is less than the second age range, and further comprising: configuring a software-defined network to block data transmissions from the first age-stratified LLM to the second age-stratified LLM; configuring the software-defined network to enable data transmissions from the second age-stratified LLM to the first age-stratified LLM; and transmitting, from the second age-stratified LLM and over the software-defined network, the second simulated response to the first age-stratified LLM.
8. 8 . The method of claim 1 , wherein the question is whether at least the first person and the second person in their particular life stages would vote for a course of action, the first simulated response is a first simulated vote that the first person would have cast in the first life stage, the second simulated response is a second simulated vote that the second person would have cast in the second life stage, and generating the third simulated response comprises: receiving, from at least the first age-stratified LLM and the second age-stratified LLM, at least the first simulated vote and the second simulated vote; and in response to a number of simulated votes in the affirmative meeting a voting threshold, generating the third simulated response comprises generating the third simulated response to be indicative of approval of the course of action.
9. 9 . The method of claim 1 , wherein the first person in the first life stage is not alive when the second person is in the second life stage.
10. 10 . The method of claim 9 , further comprising: accessing, via the at least one communication network, at least one datastore comprising contextual data associated with the question, the contextual data comprising information that would not have been available to the first person in the first life stage, and wherein: executing the first age-stratified LLM comprises executing, using the prompt and the contextual data, the first age-stratified LLM to generate the first simulated response that the first person would have provided in the first life stage having access to the contextual data.
11. 11 . At least one computer-readable storage medium storing processor executable instructions that, when executed by at least one hardware processor, cause the at least one hardware processor to perform a method for generating a dialogue between at least a first age-stratified large language model (LLM) and a second age-stratified LLM, the first age-stratified LLM trained to generate simulated responses to questions that a first person would have provided in a first life stage of their life, the second age-stratified LLM trained to generate simulated responses to questions that a second person would have provided in a second life stage of their life, the method comprising: receiving, from an electronic device and via at least one communication network, a prompt comprising a question; executing, using the prompt, the first age-stratified LLM to generate a first simulated response to the question that the first person would have provided in the first life stage; executing, using the prompt and the first simulated response, the second age-stratified LLM to generate a second simulated response to at least one of the question or the first simulated response, the second simulated response corresponding to a response that the second person would have provided in the second life stage; and generating, using the first simulated response and the second simulated response, an output representing a third simulated response to the question that the first person and the second person would have collectively determined to provide in their particular life stages, the output to be provided to the electronic device via the at least one communication network.
12. 12 . The at least one computer-readable storage medium of claim 11 , wherein: executing the first age-stratified LLM comprises executing the first age-stratified LLM in a first virtual machine (VM) having a first private Internet Protocol (IP) address in a software-defined network; and executing the second age-stratified LLM comprises executing the second age-stratified LLM in a second VM having a second private IP address in the software-defined network, and the method further comprising: transmitting, from the first VM and using the first and second private IP addresses, the first simulated response to the second VM over the software-defined network; and causing transmission, from the second VM and using the first and second private IP addresses, of the second simulated response to the first VM over the software-defined network.
13. 13 . The at least one computer-readable storage medium of claim 11 , wherein: executing the first age-stratified LLM comprises executing the first age-stratified LLM in a first container having a first private Internet Protocol (IP) address in a software-defined network; and executing the second age-stratified LLM comprises executing the second age-stratified LLM in a second container having a second private IP address in the software-defined network.
14. 14 . The at least one computer-readable storage medium of claim 13 , the method further comprising: causing transmission, from the first container and using the first and second private IP addresses, of the first simulated response to the second container over the software-defined network; and causing transmission, from the second container and using the first and second private IP addresses, of the second simulated response to the first container over the software-defined network.
15. 15 . The at least one computer-readable storage medium of claim 13 , wherein a container pod comprises the first container and the second container, and the method further comprising: instantiating a shared volume in the container pod and accessible by the first container and the second container; and wherein: executing the first age-stratified LLM comprises storing the first simulated response in the shared volume; and executing the second age-stratified LLM comprises retrieving the first simulated response from the shared volume and storing the second simulated response in the shared volume.
16. 16 . A system for generating a dialogue between at least a first age-stratified large language model (LLM) and a second age-stratified LLM, the first age-stratified LLM trained to generate simulated responses to questions that a first person would have provided in a first life stage of their life, the second age-stratified LLM trained to generate simulated responses to questions that a second person would have provided in a second life stage of their life, the system comprising: at least one hardware processor; and at least one computer-readable storage medium storing processor executable instructions that, when executed by the at least one hardware processor, cause the at least one hardware processor to perform a method comprising: receiving, from an electronic device and via at least one communication network, a prompt comprising a question; executing, using the prompt, the first age-stratified LLM to generate a first simulated response to the question that the first person would have provided in the first life stage; executing, using the prompt and the first simulated response, the second age-stratified LLM to generate a second simulated response to at least one of the question or the first simulated response, the second simulated response corresponding to a response that the second person would have provided in the second life stage; and generating, using the first simulated response and the second simulated response, an output representing a third simulated response to the question that the first person and the second person would have collectively determined to provide in their particular life stages, the output to be provided to the electronic device via the at least one communication network.
17. 17 . The system of claim 16 , wherein the first life stage of the first person corresponds to a first age range of the first person's life and is thereby a maturer life stage than the second life stage of the second person that corresponds to a second age range of the second person's life that is less than the first age range, and the method further comprising: configuring a software-defined network to enable communications from the first age-stratified LLM to the second age-stratified LLM; and transmitting, from the first age-stratified LLM and using the software-defined network, the first output to the second age-stratified LLM.
18. 18 . The system of claim 16 , wherein the second life stage of the second person corresponds to a second age range of the second person's life and is thereby a maturer life stage than the first life stage of the first person that corresponds to a first age range of the first person's life that is less than the second age range, and the method further comprising: configuring a software-defined network to block data transmissions from the first age-stratified LLM to the second age-stratified LLM; configuring the software-defined network to enable data transmissions from the second age-stratified LLM to the first age-stratified LLM; and transmitting, from the second age-stratified LLM and over the software-defined network, the second simulated response to the first age-stratified LLM.
19. 19 . The system of claim 16 , wherein the question is whether at least the first person and the second person in their particular life stages would vote for a course of action, the first simulated response is a first simulated vote that the first person would have cast in the first life stage, the second simulated response is a second simulated vote that the second person would have cast in the second life stage, and generating the third simulated response comprises: receiving, from at least the first age-stratified LLM and the second age-stratified LLM, at least the first simulated vote and the second simulated vote; and in response to a number of simulated votes in the affirmative meeting a voting threshold, generating the third simulated response comprises generating the third simulated response to be indicative of approval of the course of action.
20. 20 . The system of claim 16 , wherein the first person in the first life stage and the second person in the second life stage do not overlap in time, and the method further comprising: accessing, via the at least one communication network, at least one datastore comprising contextual data associated with the question, the contextual data comprising information that would not have been available to the first person in the first life stage, and wherein: executing the first age-stratified LLM comprises executing, using the prompt and the contextual data, the first age-stratified LLM to generate the first simulated response that the first person would have provided in the first life stage having access to the contextual data.

Description

RELATED APPLICATIONS This application claims priority under 35 U.S.C. § 119 to U.S. Provisional Application No. 63/788,586, entitled “SYSTEM AND METHOD FOR CONTEXT-AWARE, INTER-TEMPORAL AI SIMULATION OF HISTORICAL FIGURES,” filed on Apr. 14, 2025, and U.S. Provisional Application No. 63/788,599, entitled “GOVERNANCE FRAMEWORK FOR ETHICAL SOCIO-POLITICAL DECISION MAKING,” filed on Apr. 14, 2025, each of which is herein incorporated by reference in their entireties. FIELD The techniques described herein relate generally to artificial intelligence (AI) and, more particularly, to systems, apparatus, articles of manufacture, and methods for generating deliberative committees of age-stratified large language models. BACKGROUND A generative machine learning model can be trained to generate content in response to an input. An example generative machine learning model is a large language model (LLM). An LLM is a trained deep-learning model that can respond to an input prompt using natural language text. The input prompt can be a question posed by a user and the LLM can generate a response to the question using natural language text. SUMMARY In accordance with the disclosed subject matter, systems, apparatus, articles of manufacture, and methods are provided for generating banks of age-stratified large language models. Some embodiments relate to a first method for generating a first bank of age-stratified large language models (LLMs) to simulate responses to prompts that would be provided by a first person in different life stages of their life. The method comprising, using at least one computer hardware processor to perform, receiving, from at least one first datastore and via at least one communication network, first electronic data associated with the first person, processing the first electronic data to generate first life stage data comprising first multiple data portions corresponding to the different life stages of the first person, and generating the first bank of age-stratified LLMs by using the first multiple data portions to further train a reference LLM. The generating comprising generating, for each particular data portion of the first multiple data portions corresponding to a particular life stage of the first person, a respective age-stratified LLM by further training, using the particular data portion, an instance of the reference LLM to generate simulated responses to questions that the first person would have provided in their particular life stage. The method further comprises storing the first bank of age-stratified trained LLMs. Some embodiments relate to a second method for generating a dialogue between at least a first age-stratified large language model (LLM) and a second age-stratified LLM, the first age-stratified LLM trained to generate simulated responses to questions that a first person would have provided in a first life stage of their life, the second age-stratified LLM trained to generate simulated responses to questions that a second person would have provided in a second life stage of their life. The method comprises, using at least one computer hardware processor to perform, receiving, from an electronic device and via at least one communication network, a prompt comprising a question, executing, using the prompt, the first age-stratified LLM to generate a first simulated response to the question that the first person would have provided in the first life stage, executing, using at least one of the prompt or the first simulated response, the second age-stratified LLM to generate a second simulated response to at least one of the question or the first simulated response that the second person would have provided in the second life stage, and generating, using at least one of the first simulated response or the second simulated response, an output representing a third simulated response to the question that the first person and the second person would have collectively determined to provide in their particular life stages and for output to the electronic device via the at least one communication network. Some embodiments relate to an apparatus comprising at least one memory storing processor executable instructions, and at least one hardware processor configured to execute the processor executable instructions to perform any of the aforementioned methods. Some embodiments relate to at least one computer readable storage medium storing processor executable instructions that, when executed by at least one hardware processor, cause the at least one hardware processor to perform any of the aforementioned methods. Some embodiments relate to a system comprising at least one hardware processor, and at least one computer-readable storage medium storing processor executable instructions that, when executed by the at least one hardware processor, cause the at least one hardware processor to perform any of the aforementioned methods. The foregoing summary is not intended to be limiting. Moreover