Artificially intelligent entities are developed and trained by utilizing data. The data used for the development of AI entities are initially unstructured and need attention before actual utilization can be done. What an AI understands is knowledge and information, based on which it is to act and take up designated tasks. The representation of this knowledge to a computer is essential and can vary based on the requirements of tool development. Knowledge representation is not just a protocol for storing and programming knowledge in some machine. The manner or representation matters greatly. Based on the origin of the knowledge and how the knowledge is represented an Ai entity decides upon its fate. Knowledge representation in AI is thus essential to set the tonality and direction of a task designated to an Ai entity. This article will try to discuss different common knowledge representation methods that are commonplace in the development of AI entities we come across every day.
Why is knowledge representation in AI important?
It is better we understand the importance of knowledge representation as a simple example. For instance, imagine an AI entity designated to feed dogs and cows at the same time. Among food choices, it has access to grass and meat. Now the knowledge encoded in the programming will be utilized for understanding the correct course of action. In this case, the knowledge representation will be inferential. The facts, in this case, are 1. Dogs are carnivores, 2, cows are herbivores. Thus by analysis of this knowledge, the AI can assign the right food to the right animal.
What is represented?
- Objects
The knowledge of objects contains the characteristics and all the necessary aspects of an object. For E.g. a car runs on wheels.
- Events
Events are actions and incidents. Knowledge of events also includes all the aspects of an event or incident. E.g. The sun rises in the east every day and sets in the west.
- Performance data
Knowledge about performance includes the behavior of a group, object, or individual. And it contains knowledge regarding protocols and routines.
- Metaknowledge
Metaknowledge is the knowledge of knowledge itself. It includes what we know about a certain fact or incident.
Intelligence and knowledge
Intelligence is a culmination of different cognitive subsets. It can not be defined by one aspect or a few, all the aspects must be considered if intelligence is to be defined. In general, the efficiency of utilizing knowledge is known as intelligence. More data, more knowledge can be generated, and with more knowledge and accessing capabilities more efficiency can be expected from an artificially intelligent entity. An artificially intelligent entity acts on the perception and the comparison of the same with pre-programmed knowledge. An AI entity analyses the knowledge and analyzes the perception of an object’s performance or incident and then decides upon the best course of action. Thus intelligence is determined by the efficiency in utilization of knowledge.
Different approaches to knowledge representation
Simple relational knowledge
Simple and retinol knowledge representation is the representation of knowledge in the most direct and rational way possible. Subjects with tangible aspects are the ideal candidates for this kind of knowledge representation. This method is thus limited to tangible aspects of a subject and has no room for custom interfaces.
Inheritable knowledge
In this approach to knowledge representation, knowledge is derived in stages. And more knowledge can be derived in a hierarchical manner if need be. For instance, a professional fashion database will initially depict the type of clothing someone is choosing for themselves and then based on the choice the profession and gender can be determined ( of course with exceptions ). This type of multilayered representation is called hierarchical knowledge representation.
Inferential knowledge
In this case of representation, knowledge leads to simple and undeniable facts. Like dogs love to eat meat is knowledge. And inference will most likely be that a dog is a carnivore. Another instance is weather detection. For example, the knowledge can be that the sun is setting an hour early for the past week for an inference that the winter has arrived.
Procedural knowledge
This knowledge representation is perhaps the most complicated and it exploited the if-then rule extensively. In this kind of knowledge representation, the AI is programmed to make decisions based on the perception of recent facts.
