On the Integration of Knowledge Graphs into Deep Learning Models for a More Comprehensible AI—Three Challenges for Future Research

Abstract

Deep learning models contributed to reaching unprecedented results in prediction and classification tasks of Artificial Intelligence (AI) systems. However, alongside this notable progress, they do not provide human-understandable insights on how a specific result was achieved. In contexts where the impact of AI on human life is relevant (e.g., recruitment tools, medical diagnoses, etc.), explainability is not only a desirable property, but it is -or, in some cases, it will be soon-a legal requirement. Most of the available approaches to implement eXplainable Artificial Intelligence (XAI) focus on technical solutions usable only by experts able to manipulate the recursive mathematical functions in deep learning algorithms. A complementary approach is represented by symbolic AI, where symbols are elements of a lingua franca between humans and deep learning. In this context, Knowledge Graphs (KGs) and their underlying semantic technologies are the modern implementation of symbolic AI—while being less flexible and robust to noise compared to deep learning models, KGs are natively developed to be explainable. In this paper, we review the main XAI approaches existing in the literature, underlying their strengths and limitations, and we propose neural-symbolic integration as a cornerstone to design an AI which is closer to non-insiders comprehension. Within such a general direction, we identify three specific challenges for future research—knowledge matching, cross-disciplinary explanations and interactive explanations.

The Challenge of Deploying Failure Modes and Effects Analysis in Complex System Applications—Quantification and Analysis

“…Applying this technique would then lead to higher costs for the overall FMEA deployment through training and dedicating a specialised team. As indicated in Futia and Vetrò [45], there are also two main challenges for adopting KBS techniques: knowledge matching and explanations from different sources. These challenges would be exacerbated in the practice of FMEA, where there are already difficulties in its applications for identifying inputs include failures modes and other features from different sources and backgrounds as well as the gap in knowledge between Functional, Design and Process FMEAs.…”

A Review on Explainability in Multimodal Deep Neural Nets

“…[216], [203], [53] , [204], [54] Graph-based Explanations are generated through knowledge graphs and scene graphs to make the models more interpretable. [197], [182], [96] , [196] Interactive approach Virtual agents' assistance, algorithms for an explanation of the systems' internal state, and feedback loop to rectify in case of a wrong prediction are widely used.…”

“…A "Multimodal Knowledge-aware Hierarchical Attention Network "in which a knowledge graph with multiple modalities and different features is built for the medical field. In [204], a comprehensive view on the neuro symbolic AI perceptive is provided and integration of VOLUME XX, 2017 knowledge graphs in deep learning models for model interpretability is proposed.…”

A Review of Recent Deep Learning Approaches in Human-Centered Machine Learning

“…Sub-research areas such as these have a number of researchers creating different research avenues [ 11 , 12 , 15 , 17 , 20 , 26 , 28 , 32 , 38 , 39 , 40 , 41 ]. For instance, there are works on developing algorithms and novel DL architectures in XAI to add explainability to the models [ 42 , 43 , 44 , 45 , 46 ]. In comparison, there is also work that considers user experience and user requirements for XAI [ 7 , 8 , 9 , 10 , 47 ], and evaluates algorithms and models with user studies [ 48 ].…”