Table of Contents
Introduction
Have you ever shot a basketball into the hoop? Do you notice how many things are processed to make that one shot? Imagine training a machine to make a shot like that. The amount of knowledge that will be required to present to the computer is immense. There lies the problem. Even simple scenarios like lifting an apple off the desk will need a big set of rules and descriptions.
It is what makes knowledge representation in AI so crucial as well as fun to work with. Knowledge representation plays a role in setting up the environment and gives all the details necessary to the system.
Use of Knowledge Representation in AI Systems
The role of knowledge representation in AI systems can be understood by looking at the methodology followed by AI systems. The process is as follows:
1. Perception block
The perception block can be thought of as a set of senses for the machine. It is the component through which the system can interact with the environment. It can be any type of data, audio, video, temperature, etc.
2. Learning block
It is the part of the system where we train the models necessary for the machine to work on its own. The typical learning algorithms (machine learning, deep learning, etc.) are coded in the learning block. The learning block is connected directly with the perception block to retrieve the information necessary for training.
3. Reasoning – Knowledge representation block
It is the most critical block of the system. It takes in the data from the perception block and filters out what’s important. The reasoning block makes sure the knowledge is available that can be provided to the model or learning agent as and when required.
4. Planning and execution block
This block provides a functional road map to the machine. This block specifies the action to be taken and what results to be expected. This block takes the inputs from the reasoning – knowledge representation block.
Read: AI Project Ideas & Topics
Types of Knowledge
Primarily, we see five types of knowledge in any knowledge representation block in AI systems. The knowledge types are as follows:
1. Declarative: It is the type of knowledge that deals with facts, instances, objects, declared as a statement.
2. Structural: It deals with the type of knowledge that describes the relationship between instances and description.
3. Procedural: It deals with the procedures and rules required for a particular system to work efficiently.
4. Meta: It is the knowledge consisting of the higher-level data of other types of knowledge data.
5. Heuristic: It represents the data that helps in governing decisions.
Learn about: Top 8 Most Popular Google AI Projects You Should Work On
Methods for Knowledge Representation
Once we understand the knowledge to be represented and how it is going to be used, it is necessary to know how to achieve this. Here are the methods available for knowledge representation in AI systems:
1. Procedural rules
Production rules are a system in itself. It consists of a rule applier, a set of rules, and a database (memory). Whenever an input is passed through, the condition is checked through the production rules, and an appropriate rule is selected. The action is carried out based on the rules mentioned.
The whole cycle continues for every single input that is brought through the knowledge representation channel. The production rules system is expressed in terms of natural language and hence is used a lot. The only drawback is that sometimes the rule-based system gets inefficient, as some of the rules may still be active.
2. Semantic network
As the name suggests, this type of representation works with a network of data. In semantic networks, there are two types of relationships. One is the ISA relationship, and the second is the instance relationship. In the network, the blocks define objects, and the edges (or arcs) define the relationships between the blocks. Although semantic networks take more computational time, their use is extensive as the knowledge represented is simple to understand.
3. Representation by logic
Logic can be represented via agreed-upon syntax and objects. It deals with the prepositions and has no ambiguity in meaning or interpretation. This type of representation can help in logical reasoning and have a better representation of facts. However, logical representations can be tricky to work with. The strict rules of syntax and associations may make the process tricky.
4. Representation through frames
A frame is a collection of the attributes and the associated values. Frames are also called slot-filler structures. This is because the slots are the attributes, and they are filled by the values of those attributes which represent the knowledge in the environment. Frames make the grouping of data and different object values easier. But sometimes, the inference mechanism is challenging to implement or use as it is a quite generalized approach.
This is how knowledge representation in AI can be applied. But how to test these systems?
The following properties can assess any knowledge representation system:
1. Inferential adequacy and efficiency: It deals with the system’s ability to infer knowledge on its own. Can it infer knowledge from different relations and do it efficiently, are the two primary questions asked to assess this property.
2. Acquisitional adequacy: It deals with the system’s ability to gain additional knowledge based on the environment provided.
3. Representational adequacy: It deals with the system’s ability to represent all types of knowledge. Is the system versatile enough to be able to represent the data that may or may not be in the domain of previously represented knowledge?
Also Try: AI vs Human Intelligence: Difference Between AI & Human Intelligence
Conclusion
Knowledge representation in AI is going to be an evolving field. Someday it will provide the system that can be integrated, which has near-human perception and reasoning. We hope that the article provides enough to get yourself started on the journey of knowledge representation.
If you’re interested to learn more about machine learning & AI, check out IIIT-B & upGrad’s PG Diploma in Machine Learning & AI which is designed for working professionals and offers 450+ hours of rigorous training, 30+ case studies & assignments, IIIT-B Alumni status, 5+ practical hands-on capstone projects & job assistance with top firms.
What are the issues of knowledge representation in artificial intelligence?
Knowledge representation has been a long pursuit in artificial intelligence. In order to accomplish this, computers have to be able to understand the information that is presented to them. This has been solved for many problems, and there are many case studies in which information was gleaned from unstructured data, such as the human genome project. In order for information to be processed by a computer, it must be structured, and this is where the problem lies in artificial intelligence. In order to validate the process of learning from unstructured information, we need first to be able to define what this means.
What are the two ways to represent knowledge in an AI system?
There are two ways to represent knowledge in an artificial intelligence system: symbolic knowledge and sub-symbolic knowledge. Symbolic knowledge means that we have a model in our minds of what we want to do and we have a lexicon of action names that we can use to express an intention. Sub-symbolic knowledge means that we do not really have a model of what we want to do, but we rather learn skills through demonstrations.
Why is knowledge representation important?
Knowledge representation is central to artificial intelligence. Knowledge representation is all about how systems store and manipulate information. You’ve got to be able to represent these things to get them into a computer and to get the computer to think about them. It’s an exciting field of study because it is foundational. AI without knowledge representation is just not possible. There are a lot of different kinds of knowledge representation, but the most common approach is to have a database that has a set of facts, a set of propositions about the world, and a set of rules. You can make inferences from this database by applying the rules.
