Michael Uzowuihe

INTRODUCTION TO EXPLAINABLE AI The need for interpreting neural networks is necessary because, in every setting where deep learning is adopted, there is a requirement for an accurate and justifiable prediction. In essence, a deep neural network lacks justification, with this Explainable AI (XAI) tends to justify decisions by bridging between accuracy and interpretability without interpreting a model directly. What is XAI Explainable AI is the application of artificial intelligence (AI) technology with methods and techniques so that the results of a solution can be understood by a human expert. Furthermore, is the implementation of the social right to explanation. The XAI decisions face a technological problem of interpretability and info-obesity that boils down that transparency may not be always possible. Hence, in test data, a human can audit rules of XAI to gain insight into how a system would generalize on future real-world data outside a test-set. XAI Antiquity It all started with the following: • • • Symbolic reasoning systems like MYCIN, SOPHIE, etc. that were explored so they can represent, reason and make explanations for machine learning purposes. Truth maintenance systems (TMS) were developed to provide explanations by tracking lines of reasoning and justification for conclusions through rule operations or logical inferences. Researchers study on extracting meaning from rules generated by opaque neural networks gave birth to neural network-powered decision support. After this, sprang up new methods in Modern Complex A1 techniques such as deep learning and genetic algorithms to make models more explainable/interpretable and transparent – these models built are replication on Explainable Artificial Intelligence. XAI Principles As many come to depend on AI-based dynamic systems i.e. XAI, the need for clearer accountability led to a global conference on AI to ensure trust and transparency for decisionmaking processes. Hence, to deal with potential problems stemming from the importance of algorithm implementations, there is a regulation to follow to earn the right to explanation with regards to AI. Divisions of XAI XAI has researched and covered the following sectors where AI is applied: • • • • • • • Neural Network Tank imaging Antenna design (evolved antenna) Algorithmic trading (high-frequency trading) Medical diagnoses Autonomous vehicles Designing feature detectors from optimal computer designs (Computer Vision) Text analytics It is worthy to note that, Explainable AI doesn’t fully explain how a neural network reaches a prediction. In the sense that even though the impact on model predictions is explained, the decision process isn’t. This can be resolved with decision trees using Kryptonite like images4, to obtain accuracy in the process. Invariably, this suggests we can combine neural networks i.e. XAI and decision trees to address the failures to justify and attaining high accuracy.