Granted Patent B1

US 12,632,666 · App. 19/398,372 · Granted May 19, 2026

Inventor: Ming-Chang Chiu (New York, NY)

Assignee: The Joan and Irwin Jacobs Technion-Cornell Institute

G06F40/35G06F18/22

Abstract

Claims (56)

1 . A method for mitigating generative artificial intelligence errors during inter-agent communications, the method comprising:

populating at least one intent field of a schema with at least one intent value representing an intent of an inter-agent communication session including a first artificial intelligence (AI) agent and a second AI agent, wherein the at least one intent value is determined based on a communication from the first AI agent;

comparing each of a plurality of inputs from the first AI agent and the second AI agent to the schema with respect to the at least one intent value;

detecting a misalignment between a first input of the plurality of inputs and the schema when a dissimilarity between the first input and at least a portion of the schema including the at least one intent value exceeds a threshold, wherein detecting the misalignment further comprises applying at least one machine learning model trained to classify values within communications into classes using a baseline learned via training, wherein the classes include misaligned and aligned; and

performing a mitigation action based on the detected misalignment in order to mitigate an effect of at least one generative artificial intelligence error on the first input.

2 . The method of claim 1 , further comprising:

determining a proposed intent based on the communication from the first AI agent; and

verifying the proposed intent by prompting the first AI agent, wherein the at least one intent value is determined based on the verified proposed intent.

3 . The method of claim 1 , further comprising:

determining a first role of the first AI agent and a second role of the second AI agent; and

populating a first role field and a second role field of the schema with a first role value indicating the first role and a second role value indicating the second role, wherein the misalignment is detected based further on the first role value and the second role value.

4 . The method of claim 1 , further comprising:

classifying the inter-agent session into a classification based on the plurality of inputs; and

selecting a first protocol from among a plurality of protocols based on the classification, wherein the first protocol defines the schema.

5 . The method of claim 1 , further comprising:

generating a vector representation of the at least a portion of the schema including the at least one intent value; and

generating a vector representation of the first input, wherein the dissimilarity between the first input and the at least one intent value is determined based on the vector representation of the at least a portion of the schema including the at least one intent value and the vector representation of the first input.

6 . The method of claim 5 , further comprising:

determining a distance between the vector representation of the at least a portion of the schema including the at least one intent value and the vector representation of the first input, wherein the dissimilarity is determined based on the distance.

7 . The method of claim 1 , wherein the at least one intent value includes a first intent value and a second intent value, wherein the first intent value is determined based on the communication from the first AI agent, wherein the second intent value is determined based on a communication from the second AI agent.

8 . The method of claim 1 , further comprising:

tracking at least a portion of the plurality of inputs by recording transaction data generated based on the plurality of inputs on a blockchain.

9 . The method of claim 1 , wherein the first input is communicated by the first AI agent, wherein performing the mitigation action further comprises:

prompting the first AI agent with text indicating that the first input is misaligned, wherein the first AI agent provides a new first input in response to being prompted with the text indicating that the first input is misaligned.

10 . The method of claim 1 , wherein the at least one machine learning model trained to classify values is at least one non-generative machine learning model.

11 . A non-transitory computer-readable medium having stored thereon instructions for causing a processing circuitry to execute a process, the process comprising:

populating at least one intent field of a schema with at least one intent value representing an intent of an inter-agent communication session including a first artificial intelligence (AI) agent and a second AI agent, wherein the at least one intent value is determined based on a communication from the first AI agent;

comparing each of a plurality of inputs from the first AI agent and the second AI agent to the schema with respect to the at least one intent value;

detecting a misalignment between a first input of the plurality of inputs and the schema when a dissimilarity between the first input and at least a portion of the schema including the at least one intent value exceeds a threshold, wherein detecting the misalignment further comprises applying at least one machine learning model trained to classify values within communications into classes using a baseline learned via training, wherein the classes include misaligned and aligned; and

performing a mitigation action based on the detected misalignment in order to mitigate an effect of at least one generative artificial intelligence error on the first input.

12 . A system for mitigating hallucinations during inter-agent communications, comprising:

a processing circuitry; and

a memory, the memory containing instructions that, when executed by the processing circuitry, configure the system to:

populate at least one intent field of a schema with at least one intent value representing an intent of an inter-agent communication session including a first artificial intelligence (AI) agent and a second AI agent, wherein the at least one intent value is determined based on a communication from the first AI agent;

compare each of a plurality of inputs from the first AI agent and the second AI agent to the schema with respect to the at least one intent value;

detect a misalignment between a first input of the plurality of inputs and the schema when a dissimilarity between the first input and at least a portion of the schema including the at least one intent value exceeds a threshold, wherein detecting the misalignment further comprises applying at least one machine learning model trained to classify values within communications into classes using a baseline learned via training, wherein the classes include misaligned and aligned; and

perform a mitigation action based on the detected misalignment in order to mitigate an effect of at least one generative artificial intelligence error on the first input.

13 . The system of claim 12 , wherein the system is further configured to:

determine a proposed intent based on the communication from the first AI agent; and

verify the proposed intent by prompting the first AI agent, wherein the at least one intent value is determined based on the verified proposed intent.

14 . The system of claim 12 , wherein the system is further configured to:

determine a first role of the first AI agent and a second role of the second AI agent; and

populate a first role field and a second role field of the schema with a first role value indicating the first role and a second role value indicating the second role, wherein the misalignment is detected based further on the first role value and the second role value.

15 . The system of claim 12 , further comprising:

classify the inter-agent session into a classification based on the plurality of inputs; and

select a first protocol from among a plurality of protocols based on the classification, wherein the first protocol defines the schema.

16 . The system of claim 12 , wherein the system is further configured to:

generate a vector representation of the at least a portion of the schema including the at least one intent value; and

generate a vector representation of the first input, wherein the dissimilarity between the first input and the at least one intent value is determined based on the vector representation of the at least a portion of the schema including the at least one intent value and the vector representation of the first input.

17 . The system of claim 16 , wherein the system is further configured to:

determine a distance between the vector representation of the at least a portion of the schema including the at least one intent value and the vector representation of the first input, wherein the dissimilarity is determined based on the distance.

18 . The system of claim 12 , wherein the at least one intent value includes a first intent value and a second intent value, wherein the first intent value is determined based on the communication from the first AI agent, wherein the second intent value is determined based on a communication from the second AI agent.

19 . The system of claim 12 , wherein the system is further configured to:

track at least a portion of the plurality of inputs by recording transaction data generated based on the plurality of inputs on a blockchain.

20 . The system of claim 12 , wherein the first input is communicated by the first AI agent, wherein the system is further configured to:

prompt the first AI agent with text indicating that the first input is misaligned, wherein the first AI agent provides a new first input in response to being prompted with the text indicating that the first input is misaligned.

Assignments (1)

References Cited (13)