Knowledge representation issues in Artificial Intelligence (AI) - Everything you need to know
by Naima Zubair· Updated Jan 23, 2023
Human beings have been in charge for centuries because they are good at understanding things and using their brains to make sense of things. This ability makes humans unique from other species and helps them succeed in every walk of life. However, AI technology has improved over the past few decades, and we wonder if machines can learn and understand things as humans do. Well, that’s what we are going to discuss in this article. You will get everything you need to know about knowledge representation in Artificial Intelligence, its types, and issues.
Let’s get into it
What is Knowledge Representation in Artificial Intelligence (AI)
Let’s start with understanding what knowledge representation in Artificial Intelligence (AI) is.
“It refers to feeding machines with the information a human possesses to enhance their capabilities.”
The goal is to represent knowledge of the human world to machines in a way that enables them to make complex decisions and solve critical problems.
Well, it is easier said than done.
The problem is that the information stored in the human brain is complex, making it harder to represent in machines. Moreover, we have various emotions entirely alien to machines, such as conscience, intuitions, common sense, etc.
Having said this, We know many other things like numbers and facts, general knowledge, and different languages. And now we are trying to make machines understand and use this information too.
Making machines intelligent is hard. We call it Knowledge Representation and Reasoning (KR, KRR), and it’s a big challenge for us.
Knowledge – Storing the information
Reasoning – Inferring the stored information
Intelligence – Taking actions based on reason.
Now, let’s talk about what kinds of knowledge we have to represent in Artificial Intelligence (AI) to maximize its intelligence.
AI must represent several types of knowledge, including the following:
AI systems must present the essential knowledge surrounding humans, which are various objects. For example, airplanes fly, buses have seats, the park has swings, etc.
We experience the world through the eyes of the events that take place in it. This knowledge refers to all the events taking place in the world. Examples include the evolution of societies, technological advancement, wars, pandemics, etc.
This knowledge refers to the actions of human beings which they perform with other beings and things.
As is quite evident by the name, it revolves around the factual side of knowledge.
The world is full of secrets and concepts. Some we know, some we don’t. Meta-Knowledge deals with all the information we know about the world.
This knowledge means how much someone knows about a specific subject, like making computer programs.
Now that you understand what knowledge representation in AI is and how it has different forms, we can talk about the different kinds of knowledge.
Different Types of Knowledge
There are five types of knowledge which are as follows:
The Relationship Between Knowledge and Intelligence
Knowledge about the world plays a crucial role in your intelligence, and the same implies to Artificial Intelligence. An AI system responds to the given input if it has a piece of decent knowledge and experience about that input.
Imagine a car that drives itself in the city. It uses special tools to see cars, people, and things in its way. If the car’s brain doesn’t know the rules of the road, signs, and how people usually drive, it can’t make good choices and drive. It wouldn’t be able to do anything on the road. Similarly, an AI system would be unable to work or respond if it doesn’t know the input it receives.
In other words, decision-makers act based on knowledge and sensing the environment. If the knowledge is not there, they wouldn’t be able to decide by sensing.
The cycle of Knowledge Representation in Artificial Intelligence (AI)
Artificial Intelligence (AI) displays intelligent behavior comprising various components.
Let’s discuss those components.
Perception – Perception is when an AI system uses its sensors to learn about its surroundings. This perception can be pictures, sounds, words, and even the time or temperature. It helps the AI understand where it is and how it can interact with its environment.
Learning – The AI system will use what it knows to make its own decisions. It does this by using special tools called deep learning algorithms. These tools help the AI move information from what it sees, hears, and senses to its memory bank, where it can learn from it and improve its decision-making.
Knowledge Representation and Reasoning – This is like how we use what we know to make choices. These blocks in the AI work like that too. They help the AI understand essential information and then use it to make better decisions.
Planning and Execution – Even though these two blocks operate individually, they can also work together. These blocks take the input from the knowledge and reasoning blocks and use it to perform specific actions.
So, representing knowledge is crucial for AI systems to function.
Issues in Knowledge Representation
The main idea behind Knowledge Representation is that the AI can use what it knows to figure things out. But sometimes, it can be challenging for the AI to use this information and make good guesses.
We need to consider five critical issues in knowledge representations.
- Important attributes
- Relationship among attributes
- Choosing the granularity of the representation
- Representing a set of objects
- Finding the proper structure as needed
They might look overwhelming at first look, but they’re not actually. Let’s discuss them in detail one by one.
We must remember essential attributes when we try to represent knowledge in a machine. Are they primary attributes? Are they relevant or relatable? Whether we should store them or not. We must consider these questions before storing any information in the knowledge base.
Sometimes, we have basic attributes that we can use in any situation. For example, we can apply the strategies employed in chess to other board games as they share similar concepts and principles.
One thing worth mentioning is that we need to handle these attributes carefully.
Two such attributes, instance and isa are essential because both support inheritance property.
Relationship Among Attributes
Let’s discuss the next issue we encounter while representing knowledge in artificial intelligence (AI). The attributes we use to describe objects themselves act as entities. The relation among different attributes of an object is independent of the specific knowledge they encode and hold properties like
Inverse – We all know that entities in the world have various relationships with each other. —for example, the relationship between a chef, cooking, and a recipe.
Existence in an ISA hierarchy – This refers to the importance of generalizations and specifications for attributes as they support inheritance. In other words, some things are more general, and some things are more specific. This distinction helps the AI understand things better. Like ‘color’ is a particular type of ‘appearance.’ It’s like how ‘red’ is a type of ‘color’. .” The appearance attribute includes physical characteristics such as color, texture, and shape. The color attribute is a specific aspect of appearance and inherits properties from the general attribute.
Techniques for reasoning about values – Here, we infer understood values of attributes that we don’t need to state explicitly. During this process, we use various types of information. For example, weight must be in a unit of mass, and a child’s age cannot be greater than the age of their parents. We specify these values when first establishing a database.
Single value attributes – This ensures that a particular attribute has a singular value. For instance, a football player may only have one specific jersey number and belong to a single club. Knowledge representation systems use different ways to deal with information with only one value, like when there’s only one answer to a question.
Choosing the Granularity of representation
Now, at what level of depth do we want to go with the knowledge representation?
Before we represent knowledge, it is crucial to know the following:
High-level facts may not be adequate for inference, meaning that storing broad facts and figures won’t give you proper reasoning.
On the other hand, low-level primitives need a lot of storage because keeping every nitty gritty detail in the knowledge base would take a lot of work.
It’s better to plan and decide how much detail we need to remember about something and the essential parts of that information. This way, we only need a little memory storage for the AI to use that information to make good guesses.
Let’s try to understand this with an example. If we have the information “Bob is driving a car,” we can represent this as “Bob, agent, is driving, action, a car, object” this way, we can find out what Bob is doing he is driving a car. Still, we can’t tell where he is going. We need to add more information to answer this question, like “Bob is driving to work.”
With this extra information, we can infer that Bob is driving a car to work.
Representing Set of Objects
When we put things together in a group, it’s important to remember that some things might be valid for the whole group but not for each item individually.”
For example, “there are more cows than people in India” and “Hindi speakers exist throughout the country.”
We must attach the assertion to the sets representing people, cows, and Hindi speakers to describe these facts.
It’s more efficient to associate a property with a whole group rather than each member. We can show information smartly or in a way that groups things and makes them easy to understand.
Finding the Right Structure
The question is how to ask for relevant information stored in a database.
This process is about figuring out how to organize information in a way that works well. This method means choosing a good way to start, adding new information when needed, and knowing when to make a new system.
In the end, finding the correct information in a significant computer memory takes work. It needs a lot of work to ensure we choose the right way to organize it, fill in the correct information, and know when to create a new system. But, unfortunately, we can not only use one way to do it. Instead, we have developed different methods to help with some of these problems, but not all. So, we must think about what we need and pick the best way for us. This approach will ensure that we can access the relevant information quickly and efficiently.
Thanks for Reading!
Tech Content Writer
Naima is a skilled and experienced content writer, with a passion for creating high-quality, well-researched articles and blog posts. With her strong writing skills and attention to detail, Naima is able to craft engaging and informative content that resonates with readers.
In addition to her work as a content writer, Naima is also highly interested in technology and artificial intelligence and is always looking for ways to stay up-to-date on the latest trends and developments in these fields.