XAI Algorithm Selection Using User Interaction Prediction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing AI systems lack the ability to predict when users need explanations, which is crucial for compliance with regulations like GDPR and for providing transparent and ethical decision-making.

Innovation Solution

A system that uses a usage graph structure and node2vec embeddings to predict the need for explainable AI (XAI) by analyzing user interactions, selecting AI algorithms with XAI capabilities based on cosine similarity and user feedback.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If AI algorithms are selected without considering explanation needs, then model selection is simpler and faster, but compliance with regulations like GDPR and model transparency are compromised

Engineering Contradiction:

Improvecompliance with GDPR and model transparencyVSAvoidAI algorithm selection process

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system performs preliminary analysis of user interactions and interaction patterns before final algorithm selection. By predicting explanation needs in advance using embeddings and similarity calculations, the system prepares appropriate algorithms beforehand, ensuring compliance is built into the selection process rather than added as an afterthought

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system introduces an intermediary layer of analysis that sits between user interactions and algorithm selection. This intermediary component uses embeddings, similarity calculations, and prediction models to translate interaction patterns into algorithm selection criteria, mediating between raw user behavior data and final algorithm choices to ensure transparency and compliance

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If explainable AI components are always selected, then user trust and transparency are improved, but system flexibility and adaptability to different user needs are reduced

Engineering Contradiction:

Improveuser trust and transparencyVSAvoidadaptability to different user needs

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system dynamically adjusts algorithm selection based on real-time analysis of user interactions. Rather than statically choosing explainable or non-explainable algorithms, the system continuously monitors interaction patterns, updates embeddings, and adapts its predictions to select the most appropriate algorithm for each specific context, balancing transparency needs with functional requirements

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system changes the parameters of algorithm selection based on predicted explanation needs. By calculating similarity values and updating prediction models based on interaction data, the system adjusts selection criteria dynamically - choosing explainable algorithms when transparency parameters are high and prioritizing performance algorithms when explanatory needs are low, thus adapting to different user contexts

Inventive Principle:

Principle #35Parameter changes

3Measurement precision

If AI systems continuously monitor and analyze user interactions to predict explanation needs, then algorithm selection accuracy is improved, but computational resources and processing time are increased

Engineering Contradiction:

Improvealgorithm selection accuracyVSAvoidcomputational resources and processing time

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system extracts and analyzes only the most relevant features from user interactions - specifically those related to explanation needs - rather than processing all interaction data equally. By using embeddings to capture essential patterns and focusing computational resources on predicting explanation requirements, the system achieves accurate selection while reducing overall computational burden

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The system performs preliminary embedding creation and pattern analysis during off-peak times or in advance, so that when algorithm selection is needed, the computational heavy lifting has already been done. This allows accurate predictions to be made without requiring intensive real-time computation during the actual selection moment

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12626166B2Predicting the need for XAI in artificial intelligence systems

Publication Date: 2026.05.12 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

A method, computer program, and computer system are provided for selecting an artificial intelligence algorithm. Data corresponding to a user interaction with an artificial intelligence system is received. A need by the user for an explanation associated with the artificial intelligence system is identified based on the received data corresponding to the user interaction with the artificial intelligence system. An artificial intelligence algorithm is selected from among a plurality of artificial intelligence algorithms. The selected artificial intelligence algorithm includes an explainable artificial intelligence component. A model is created for the the selected artificial intelligence algorithm corresponding to the received data.