Granted Patent B2

US 12,235,903 · App. 17/118,601 · Granted Feb 25, 2025

Inventors: Ambrish Rawat (Dublin, IE); Jonathan Peter Epperlein (Phibsborough, IE)

Assignee: International Business Machines Corporation

G06F16/9035G06F9/3836G06F16/90335G06F16/9536

Abstract

Claims (104)

1. A computer system comprising:

one or more processing devices and at least one memory device operably coupled to the one or more processing devices, the one or more processing devices are configured to:

receive an original query electronically, wherein a response to the original query is to be submitted electronically by a human;

modify the original query, thereby generating a modified query, wherein one or more features of the modified query are preserved, and the modified query is comprehensible by the human;

transmit the modified query to one or more artificial intelligence (AI) agents;

execute, via the one or more AI agents, an attempted labeling operation on the modified query; and

determine that the modified query cannot be responded to through electronic means without human support based on the attempted labeling operation failing to assign a label to the modified query and the modified query can be responded to through electronic means without human support based on the attempted labeling operation assigning a label to the modified query.

2. The computer system of claim 1 , further comprising:

one or more filtering components communicatively coupled to the one or more processing devices, wherein the one or more processing devices are further configured to:

transmit the original query to the one or more filtering components, thereby to implement one or more adversarial hardening transformations to the original query.

3. The computer system of claim 2 , further comprising:

one or more semantic services communicatively coupled to the one or more processing devices, wherein the one or more processing devices are further configured to:

transmit the modified query to the one or more semantic services;

transmit the original query to the one or more semantic services; and

execute, through the one or more semantic services, a similarity assessment between the original query and the modified query.

4. The computer system of claim 3 , wherein the one or more processing devices are further configured to:

determine, through the one or more semantic services, the modified query is comprehensible by the human.

5. The computer system of claim 1 , wherein the one or more processing devices are further configured to one or more of:

assign, through the one or more AI agents, a label to the modified query;

assign, through the one or more AI agents, a confidence value to the modified query; and

generate, through the one or more AI agents, a response to the modified query.

6. The computer system of claim 1 , wherein the one or more processing devices are further configured to one or more of:

fail to assign, through the one or more AI agents, a label to the modified query;

generate, through the one or more AI agents, no response to the modified query; and

generate, through the one or more AI agents, an incorrect response to the modified query.

7. The computer system of claim 1 , wherein:

the original query comprises a modality of one or more of:

an original textual query;

an original pictorial query; and

an original audio query; and

the one or more processing devices are further configured to modify the original query including modification of the modality of the original query including conversion of the original textual query to one or more of:

an at least partially modified audio query including at least some adversarial noise; and

an at least partially modified image query.

8. The computer system of claim 1 , further comprising:

one or more filtering components communicatively coupled to the one or more processing devices;

one or more semantic services communicatively coupled to the one or more filtering components;

access to the one or more AI agents communicatively coupled to the one or more filtering components, wherein the one or more processing devices are further configured to:

transform, through the one or more filtering components, the original query into an at least partially modified query;

transmit, iteratively, the original query and the at least partially modified query to the one or more semantic services;

transmit, iteratively, the at least partially modified query to the one or more AI agents;

transmit, iteratively, the at least partially modified query to the one or more filtering components;

determine, iteratively:

comprehensibility, through the one or more semantic services, of the at least partially modified query by the human; and

the at least partially modified query cannot be responded to through the electronic means without the human support, through the one or more AI agents, thereby to establish the at least partially modified query is a fully modified query; and

transmit the fully modified query to the human.

9. A computer program product, the computer program product comprising:

one or more computer readable storage media; and

program instructions stored on the one or more computer readable storage media, the program instructions comprising:

program instructions to receive an original query electronically, wherein a response to the original query is to be submitted electronically by a human;

program instructions to modify the original query, thereby generating a modified query, wherein one or more features of the modified query are preserved, and the modified query is comprehensible by the human;

program instructions to transmit the modified query to one or more artificial intelligence (AI) agents;

program instructions to execute, via the one or more AI agents, an attempted labeling operation on the modified query; and

program instructions to determine that the modified query cannot be responded to through electronic means without human support based on the attempted labeling operation failing to assign a label to the modified query and the modified query can be responded to through electronic means without human support based on the attempted labeling operation assigning a label to the modified query.

10. The computer program product of claim 9 , further comprising:

program instructions to transform, through one or more filtering components, the original query into an at least partially modified query, thereby implementing one or more adversarial hardening transformations to the original query;

program instructions to transmit, iteratively, the original query and the at least partially modified query to one or more semantic services;

program instructions to transmit, iteratively, the at least partially modified query to the one or more AI agents;

program instructions to transmit, iteratively, the at least partially modified query to the one or more filtering components;

program instructions to determine, iteratively:

comprehensibility, through the one or more semantic services, of the at least partially modified query by the human; and

the at least partially modified query cannot be responded to through the electronic means without the human support, through the one or more AI agents, thereby to establish the at least partially modified query is a fully modified query; and

program instructions to transmit the fully modified query to the human.

11. A computer-implemented method comprising:

receiving an original query electronically, wherein a response to the original query is to be submitted electronically by a human;

modifying the original query, thereby generating a modified query, wherein one or more features of the modified query are preserved, and the modified query is comprehensible by the human;

transmitting the modified query to one or more artificial intelligence (AI) agents;

executing, via the one or more AI agents, an attempted labeling operation on the modified query; and

determining that the modified query cannot be responded to through electronic means without human support based on the attempted labeling operation failing to assign a label to the modified query and the modified query can be responded to through electronic means without human support based on the attempted labeling operation assigning a label to the modified query.

12. The method of claim 11 , wherein the generating a modified query

comprises:

transmitting the original query to one or more filtering components, thereby implementing one or more adversarial hardening transformations to the original query.

13. The method of claim 12 , further comprising:

transmitting the modified query to one or more semantic services;

transmitting the original query to the one or more semantic services; and

executing, through the one or more semantic services, a similarity assessment between the original query and the modified query.

14. The method of claim 13 , wherein the executing the similarity assessment comprises:

determining, through the one or more semantic services, the modified query is comprehensible by the human.

15. The method of claim 11 , wherein executing an attempted labeling operation

on the modified query comprises one or more of:

assigning a label to the modified query;

assigning a confidence value to the modified query; and

generating a response to the modified query.

16. The method of claim 11 , wherein executing an attempted labeling

operation on the modified query comprises one or more of:

failing to assign a label to the modified query;

generating no response to the modified query; and

generating an incorrect response to the modified query.

17. The method of claim 11 , wherein:

the receiving the original query electronically comprises receiving the original query with a modality of one or more of:

an original textual query;

an original pictorial query; and

an original audio query; and

the modifying the original query comprises modifying the modality of the original query comprises converting the original textual query to one or more of:

an at least partially modified audio query including at least some adversarial noise; and

an at least partially modified image query.

18. The method of claim 11 , further comprising:

transforming, through the one or more filtering components, the original query into an at least partially modified query;

transmitting, iteratively, the original query and the at least partially modified query to the one or more semantic services;

transmitting, iteratively, the at least partially modified query to the one or more AI agents;

transmitting, iteratively, the at least partially modified query to the one or more filtering components;

determining, iteratively:

comprehensibility, through the one or more semantic services, of the at least partially modified query by the human; and

the at least partially modified query cannot be properly responded to through the electronic means without the human support, thereby establishing the at least partially modified query is a fully modified query; and

transmitting the fully modified query to the human.

Assignments (1)

Continuity (1)

References Cited (46)