K is for Knowledge representation
From A to I to Z: Jaid’s Guide to Artificial Intelligence
Knowledge representation is the process of structuring data in a way that makes sense to an AI. In other words, it converts information into a language the AI can understand.
The purpose of the process is to enable the AI to acquire knowledge from that data, and to use this knowledge to solve problems, make decisions, or take some other type of action.
Knowledge representation typically happens in five stages:
- Defining concepts
- Ontology creation. In other words, setting out the relationships between the concepts in step one
- Knowledge formalization, that is translating the concepts and ontology into a language the AI can understand
- Knowledge storage, typically a database the AI can refer to as needed
- Reasoning. This is where the AI puts the knowledge to use.
Imagine you used an AI-powered customer service tool. You want to make it more reliably consistent and accurate, so you use customer service data to help it improve.
The knowledge representation process would look something like this:
First, you’d identify the most important concepts your AI needs to know to serve your customers properly. These would include data about your range of products, common issues your customers inquire about, and information about how your customer service team successfully resolved these issues in the past.
Next, you’d connect the concepts together to make their relationships clear. For example, you’d link each problem to its likely solution, or link each one of your customers to the products they own from your range.
Third, you’d turn this knowledge into a formal language.
This could be rules-based. For example, If X then Y. Or it could be a probabilistic model which suggests the response with the best chance of satisfying the customer.
This knowledge would then be stored, and the AI would draw on it to figure out the nature of customer inquiries and formulate the most appropriate reply.
Allen Newell, Herbert Simon, and J.C Shaw developed the first computerized knowledge representation model — the General Problem Solver — in 1957.
The program used means-ends analysis, a problem-solving technique in which the computer chooses the action most likely to reduce the difference between the current state of affairs and an ideal state of affairs.
Means-ends analysis is used in AI to this day to help computers narrow down their search parameters. It’s also commonly used in design and as a tool to spark creativity.
While its computerization is relatively recent, knowledge representation is over 2,000 years old. The concept can be traced back to Ancient Greek philosopher Aristotle’s syllogism — a logical argument that draws a conclusion based on two premises.
An example of syllogism would be:
- The city collects refuse every Thursday. (premise 1)
- Tomorrow is Thursday. (premise 2)
- Therefore, the city will collect my refuse tomorrow.
Want to know more?
This article discusses means-end analysis in detail, including how it works and the step-by-step process for applying it to AI.
This 2009 academic paper is a fascinating look at the intersection between AI and classical philosophy. The authors predict that philosophers will have an increasingly important role as AI technology continues to advance — a prediction that’s slowly becoming true.
Knowledge representation is the bridge between data and AI. Once data is structured in a way that makes sense to AI, the AI can use its superior processing power to apply that knowledge accurately and reliably to a vast range of use cases — from medical diagnosis to legal analysis and customer support.
Take advantage of Jaid’s AI-powered platform to optimize your business functions and customer data – contact us today!