Many believe that knowledge representation and reasoning (KR&R) is playing a central role in the evolution of artificial intelligence (AI). But what is it really, and more importantly, how does it work?
What is Knowledge Representation and Reasoning?
Knowledge representation and reasoning is a field dedicated to coming up with real-world descriptions that a computer can use to solve a complex problem such as developing a dialog using natural language or diagnosing a medical condition. The AI system integrates relevant findings to automate the process of reasoning. As such, the goal of knowledge representation and reasoning is to understand and develop intelligent behavior.
What is the History of KR&R?
Knowledge representation and reasoning is not a new concept. One of the earliest known applications of digital knowledge representation was the general problem solver developed by Allen Newell and Herbert A. Simon in the late 1950s. Their knowledge representation system mostly focused on data structures that started with identifying a goal, which is later on divided into subgoals then strategies that allowed a system to more easily attain the said goal.
The popularity of KR&R reached its peak in the 1970s and 1980s with the successful birth of expert systems. From general problem solvers, the field evolved to include and shift focus toward the creation of expert systems that can effectively match human intelligence to perform a given task. Before we dive into how it works, though, let’s first define what knowledge representation and reasoning is.
How does Knowledge Representation and Reasoning Work?
Unlike humans, machines find it hard to interpret knowledge and mimic reasoning. Let’s demonstrate how knowledge representation and reasoning works with an example. The following lists arguments that an AI system uses:
If I have a history lecture today, then it is probably Tuesday or Thursday.
It is not Tuesday.
I have a history lecture today. Or I have no class today.
If I don’t have a lecture today, then I am sad.
I am not sad.
You need to answer the question, “What day is it today?” If you respond that it is Thursday, then you are right. How did you come up with this inference? You took several pieces of information and combined them to come up with the appropriate answer. A machine would find it hard to do the same thing but is made possible by knowledge representation and reasoning.
How AI systems work depends on the approach the programmer used to build them. In knowledge representation and reasoning’s case, knowledge engineers need to provide a system with:
- Knowledge base: A knowledge base contains factual details, including assertions and propositions, and all other relevant information that a computer can use to dissect formal language.
- Reasoning engine: Reasoning engines help a machine produce contextual results derived from a knowledge base.
In sum, KR&R allows AI systems to come up with human-like inferences by accessing a database of factual information (e.g., Seattle is in Washington and Washington is in the U.S.) and a means to convert the data into logical reasoning (e.g., Seattle is in the U.S.).
Many AI systems rely heavily on knowledge representation and reasoning to create stories, abstract content, and come up with analogies. Interestingly, the concept also has significant applications beyond AI, such as in biomedicine, engineering, and other industries.
While KR&R has many uses, it still has a long way to go before it can perfect the way an AI system processes different types of knowledge. When it comes to natural language processing, knowledge representation and reasoning still needs to fix a few kinks. Knowledge engineers need to distinguish differences in syntax and semantics to use a uniform sentence structure that a machine will understand.