Machine Learning vs. Machine Understanding

Different fields of AI have different strengths and even completely different ways of working. Most everyone knows about Machine Learning. Machine Learning is from the field of Data Science and it focuses on getting computers to learn and act like humans do, and to improve their learning over time in autonomous fashion. They do this by feeding them data and information in the form of observations and real-world interactions. Machine Learning is data-driven and it normally requires a lot of clean data to learn from. It’s important to point out that machine learning is focused on the patterns in the data and not on the semantic meaning of the underlying data.

Machine understanding is the focus of modern AI Expert Systems which come from the field of Knowledge Engineering. Expert Systems use knowledge representation and reasoning standards created for the semantic web as a standardized way to model knowledge and encode the logical reasoning to apply that knowledge. Machine understanding is knowledge-driven, and it normally requires a lot of integrated domain specific knowledge models to understand the domain specific vocabularies used in the underlying information and data that align to the knowledge models. The knowledge models, created with standardized languages like W3C OWLv2 DL, allow the Expert System to understand the meaning of the data and information mapped to the knowledge models. It’s important to remember that this is human knowledge and that both the knowledge models and how the knowledge is applied is fully transparent and explainable.

Most human knowledge can be broken down into sentences. Subject, predicate, and objects are the three different components when breaking down a sentence. The subject is the "who" or "what" of the sentence, the predicate is the verb, and the object is any noun or concept that is part of the action of the subject. When you feed data and information into an Expert System it breaks it down into subject, predicate, and object statements in standards-based formats such as the W3C RDF and stores it in the knowledge base. Each subject, predicate, and object used in the statements is mapped to knowledge models that enables the Expert System to understand the semantic meaning of the subject, predicate, and object statements. Once the statement is understood, it can feed into decision making or other follow on knowledge-driven activities.

Knowledge bases used by Expert Systems are also ideal tools for integrating with Data Science pipelines at for Analytics and Machine Learning: the flexibility of data production and the ability to map between knowledge models (ontologies) dynamically means that many of the big headaches involved in data analytics - de-duplication, cleansing, validation, dimensionality, ensuring consistent meaning in properties and resources, and so forth, the 80-90% of work that most data scientists have to do just to get data into a form that’s useful for analysis, can be done automatically.

Blended AI is thought of AI that includes AI functionality from both Knowledge Engineering and Data Science that enables blended AI to support both inductive statistical inference and deductive logical inference which enables blended AI to reason more closely to humans than either approach can alone. Blended AI is currently a hot topic in Deep Learning circles and is something I'd keep an eye on for those watching the AI market. You will of course have solutions that only need machine learning or machine understanding but we should be aware of the real synergies that happen when they are blended together.