Abstract
Knowledge-based artificial intelligence has been extensively used in numerous application areas leading to the development of a vast number of methods and tools. In recent years, focus has shifted on non-symbolic approaches, and neural networks in particular have achieved human-level performance in various applications where accountability is a very important issue, closely related to the interpretability of artificial intelligence methods in general. Lack of interpretability of neural networks and various machine learning methods has led to the adoption of knowledge-based methods instead, which offer models compliant with explainability and interpretability requirements. In this article, an overview of knowledge-based methods is presented along with the state of the art in this area, offering to the AI practitioner guidance for applying these important methods in practice.
TopBackground
Since the emergence of Artificial Intelligence in 50s, approaches based on logic based reasoning combined with formal encoding of human knowledge have been of the forefront of artificial intelligence research, due to their similarly to the human way of thinking and their suitability for automation (Van Harmelen et al., 2008). Having lost their preeminence since the so called “AI winter” of late 80s and early 90s and currently overshadowed by data intensive Machine Learning (ML) approaches, especially deep learning approaches, the Knowledge Based (KB) approaches are currently not in the focus of attention of the majority of AI community and the public in general to the same degree as in the early days of AI. On the other hand they still remain a crucial building block of AI applications and an important approach for overcoming current problems caused when deploying AI systems.
As machine learning approaches are applied in an ever-increasing number of application domains several issues are raised related to biases of Machine learning generated models, resulting from biases in training datasets, and the explainability of produced models which contradicts the requirement of accountability of deployed Artificial Intelligence Applications. Since AI is getting more important and ML is applied to critical tasks, such as medical diagnosis and self-driving cars, accountability becomes a major issue as exemplified by EU’s regulations (Goodman & Flaxman, 2017) and DARPA’s explainable AI initiative (Gunning & Aha, 2019).
Neural Networks in particular, although achieving human level, or higher, performance on various applications such as image recognition and automatic translation and being the leading paradigm in current AI research, suffer from their lack of explainability. Specifically the structure of Neural Networks is complex and the interpretation of internal nodes and weights of edges is typically missing making them a “black box” approach, non-compatible with accountability requirements. Deploying such non interpretable systems for a critical task contradicts with the accountability requirements typical for critical applications such as autonomous vehicles and medical diagnosis (Tjoa & Guan, 2020).
Key Terms in this Chapter
Machine Learning: Branch of Artificial Intelligence based on building models by learning directly from data while minimizing human effort.
Knowledge Representation: A field of Artificial Intelligence focusing on representing knowledge in such a way that a software system can use directly for completing tasks and solving problems.
Logic Reasoner: A software system that infers logical consequences from a set of asserted facts and axioms.
Logic Programming: A programming paradigm based on formal logic where a program consists of a set of logic rules and facts about an application domain.
Ontology: A formal conceptualization of a domain.
Artificial Intelligence: A branch of computer science focusing on building systems that perform tasks that are typically performed by humans.
Non-Monotonic Logics: Logics that allow of making previous axioms invalid and retracting some inferred facts in addition to adding new inferences.
Modal Logics: Logics that formally represent the notions of necessity and possibility.
Semantic Web: A set of standards and tools for making Web information machine readable.