IP Library › Granted Patent US 12699910

Granted Patent B2

US 12699910 · App. 17/823,357 · Granted Aug 4, 2026

Explainability for artificial intelligence-based decisions

Inventors: Aishwarya Satish (Santa Clara, CA); Anshuma Chandak (Mountain View, CA); Emmanuel Munguia Tapia (San Jose, CA); Molly Carrene Cho (San Jose, CA)

Assignee: Accenture Global Solutions Limited

G06N5/045 G06F40/56

View Patent ↗

Loading inventors, assignments & file history…

Monitor This Case

Get email alerts when status or documents change.

Order Certified Copies

Most orders are placed with the USPTO same day — all within 24 business hours.

Order via The Patent Place →

Pre-filled with this patent's details

Quick Facts

Patent No.: US 12699910
App. No.: 17/823,357
Granted: Aug 4, 2026
Kind: B2

External Links

USPTO Patent Center↗ Full Patent PDF↗

Abstract

The proposed systems and methods are directed to explainability-augmented AI systems. These systems are configured to automatically identify, based on one or more of metadata associated with labels assigned to sample data and responses to AI-system-related questionnaires, one or more reasons that support the decisions made by an AI model in response to user queries. The proposed systems apply natural language processing (NLP) to transform the explainability data (e.g., metadata and questionnaire data) to generate human reader-friendly output that summarizes the reasoning by which the AI system made a specific decision and offer transparency to the AI-decision-making process.

Claims (87)

1 . A method for providing reasoning for decisions made by an artificial intelligence (AI) system, the method comprising:

assigning, at the AI system, a first label to a first data item of a plurality of data items, wherein the AI system comprises a plurality of components including a metadata module, a questionnaire module, a domain data and model knowledge module, a recommendation module and an array of explainability components, wherein each of the plurality of components communicate with each other over a network, and wherein the explainability components comprise a causal graph, support vector data samples, and an explainability classifier;

associating, at the AI system, using the metadata module, a first metadata to the first data item, the first metadata explaining the first label was assigned to the first data item because the first data item includes a first feature;

receiving, at the AI system, using the questionnaire module, a first query involving a second data item;

determining, at the AI system, the second data item includes a second feature that matches the first feature;

assigning, at the AI system and in response to the second feature matching the first feature, the first label to the second data item;

determining characteristics of the AI system based on an explainability questionnaire dataset;

processing, using the recommendation module of the AI system, the explainability questionnaire dataset to identify a plurality of explainability methods including one or more data-specific explainability methods and one or more model-specific explainability methods;

processing, using the domain data and model knowledge module of the AI system, the explainability questionnaire dataset to identify a historical summary of the plurality of explainability methods;

selecting, by the AI system, at least one explainability method among the plurality of explainability methods for the AI system in a ranked order based on the determined characteristics of the AI system, the one or more data-specific explainability methods, the one or more model-specific explainability methods, and the historical summary, wherein the recommendation module implements a recommendation model to use a questionnaire to create a ranked list of the plurality of explainability methods as explainability options for the AI system in the ranked order;

generating, by the AI system, a first content describing the first metadata using the selected at least one explainability method, wherein generating the first content comprises:

receiving, at the AI system, a first input identifying a use-case domain for the AI system;

configuring, at the AI system, a format and granularity of the first content based on the first input; and

implementing the explainability classifier, an ontology builder, and nearest neighbors algorithms to establish a visual representation of the reasoning for the decision of selecting the at least one explainability method, wherein the visual representation of the reasoning comprises at least one of a general data graph and the causal graph, wherein:

the general data graph is configured to organize the first metadata based on weighted relationships between a plurality of nodes in the recommendation model; and

the causal graph is configured to organize the first metadata based on a cause and an effect, wherein each relationship between the cause and the corresponding effect is weighted;

presenting, by the AI system in response to the first query and via a display of a user device, both the first label and the first content describing the first metadata;

receiving, at the AI system, via the display of the user device, a first feedback in response to the presentation of the first label and the first content; and

retraining the AI system based on the first feedback.

2 . The method of claim 1 , wherein the first metadata corresponds to a type of keyword, free text, bounding box, table, timestamp, and trigger action.

3 . The method of claim 1 , further comprising:

transforming, at the AI system, the first metadata to natural language; and

generating, at the AI system, the first content using the natural language.

4 . The method of claim 1 , further comprising:

receiving, at the AI system, a first response to a first question in the questionnaire; and

selecting a first format for the first content based on the first response.

5 . The method of claim 1 , further comprising:

updating information linked with the first data item based on the first feedback.

6 . The method of claim 1 , further comprising applying, at the AI system, a labelling algorithm to a third data item of the plurality of data items, thereby propagating the first metadata to additional test samples for the AI system.

7 . A non-transitory computer-readable medium storing software comprising instructions executable by one or more computers which, upon such execution, cause the one or more computers to:

assign, at an artificial intelligence (AI) system, a first label to a first data item of a plurality of data items, wherein the AI system comprises a plurality of components including a metadata module, a questionnaire module, a domain data and model knowledge module, a recommendation module and an array of explainability components, wherein each of the plurality of components communicate with each other over a network, and wherein the explainability components comprise a causal graph, support vector data samples, and an explainability classifier;

associate, at the AI system, using the metadata module, a first metadata to the first data item, the first metadata explaining the first label was assigned to the first data item because the first data item includes a first feature;

receive, at the AI system, using the questionnaire module, a first query involving a second data item;

determine, at the AI system, that the second data item includes a second feature that matches the first feature;

assign, at the AI system and in response to the second feature matching the first feature, the first label to the second data item;

determine characteristics of the AI system based on an explainability questionnaire dataset;

process, using the recommendation module of the AI system, the explainability questionnaire dataset to identify a plurality of explainability methods including one or more data-specific explainability methods and one or more model-specific explainability methods;

processing, using the domain data and model knowledge module of the AI system, the explainability questionnaire dataset to identify a historical summary of the plurality of explainability methods;

select, by the AI system, at least one explainability method among the plurality of explainability methods for the AI system in a ranked order based on the determined characteristics of the AI system, the one or more data-specific explainability methods, the one or more model-specific explainability methods, and the historical summary, wherein the recommendation module implements a recommendation model to use a questionnaire to create a ranked list of the plurality of explainability methods as explainability options for the AI system in the ranked order;

generate, by the AI system, a first content describing the first metadata using the selected at least one explainability method, wherein to generate the first content, the one or more computers are configured to:

receive, at the AI system, a first input identifying a use-case domain for the AI system;

configure, at the AI system, a format and granularity of the first content based on the first input; and

implement, at the AI system, the explainability classifier, an ontology builder, and nearest neighbors algorithms to establish a visual representation of the reasoning for the decision of selecting the at least one explainability method, wherein the visual representation of the reasoning comprises at least one of a general data graph and the causal graph, wherein:

the general data graph is configured to organize the first metadata based on weighted relationships between a plurality of nodes in the recommendation model; and

the causal graph is configured to organize the first metadata based on a cause and an effect, wherein each relationship between the cause and the corresponding effect is weighted;

present, by the AI system and in response to the first query, both the first label and the first content describing the first metadata;

receive, at the AI system, via the display of the user device, a first feedback in response to the presentation of the first label and the first content; and

retrain the AI system based on the first feedback.

8 . The non-transitory computer-readable medium storing software of claim 7 , wherein the first metadata corresponds to a type of keyword, free text, bounding box, table, timestamp, and trigger action.

9 . The non-transitory computer-readable medium storing software of claim 7 , wherein the instructions further cause the one or more computers to:

transform, at the AI system, the first metadata to natural language; and

generate, at the AI system, the first content using the natural language.

10 . The non-transitory computer-readable medium storing software of claim 7 , wherein the instructions further cause the one or more computers to:

receive, at the AI system, a first response to a first question in the questionnaire;

and

select a first format for the first content based on the first response.

11 . The non-transitory computer-readable medium storing software of claim 7 , wherein the instructions further cause the one or more computers to:

update information linked with the first data item based on the first feedback.

12 . The non-transitory computer-readable medium storing software of claim 11 , wherein the instructions further cause the one or more computers to apply, at the AI system, a labeling algorithm to a third data item of the plurality of data items, thereby propagating the first metadata to additional test samples for the AI system.

13 . A system for providing reasoning for decisions made by an artificial intelligence (AI) system, the system comprising one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to:

assign, at the AI system, a first label to a first data item of a plurality of data items, wherein the AI system comprises a plurality of components including a metadata module, a questionnaire module, a domain data and model knowledge module, a recommendation module and an array of explainability components, wherein each of the plurality of components communicate with each other over a network, and wherein the explainability components comprise a causal graph, support vector data samples, and an explainability classifier;