INTRODUCTION

People have knowledge stored in their memory. Without that knowledge, intelligent decision making would be impossible. Thus, intelligence draws, in great part, from stored knowledge. 

In the same way, machine intelligence also draws from stored knowledge. Thus, knowledge systems are fundamental to AI.

How do we store knowledge in a machine such that it can be used for intelligent decision making? That’s what we explore in this post. 

Let’s begin. 

VALIDITY OF REASONING

How do we know whether a method of reasoning is valid? Consider the following example:

1. Only if you fail, can you succeed.
2. You have succeeded. 
3. Thus, you must have failed. 

Is this reasoning valid? Yes, it is. Think about it. If 1 and 2 are true, then 3 has to be true. 3 cannot ever be false if 1 and 2 are both true. 

Now, let’s consider another example:

  1. If you cry, you will be sad. 
  2. You did not cry.
  3. Thus, you are not sad.

Is this reasoning valid?

  1. Say that crying does lead to sadness (so 1 is true).
  2. And you did not cry (so 2 is also true).
  3. But does this always mean that you are not sad? No, it doesn’t. You can still be sad for some other reason – such as a failed exam (so 3 can be false even though 1 and 2 are both true).

Thus this example is a case of invalid reasoning. 

You can make it a case of valid reasoning by making one small change:

  1. Only if you cry, you will be sad (instead of “If you cry, you will be sad”)
  2. You did not cry. 
  3. Thus, you are not sad. 

With this change, the implication is that crying is the only way to be sad. There is no other way to be sad. Thus if 1 and 2 are true, then 3 has to be true.

In general, validity of reasoning is judged by the certainty of a conclusion if the statements supporting the conclusion are all true. 

As far as a machine is concerned, valid reasoning is what determines intelligence. If the machine uses invalid reasoning, then it is not very intelligent. 

THE NATURE OF KNOWLEDGE SYSTEMS IN MACHINES

Due to the method used to determine valid or invalid reasoning, knowledge used for intelligent machines must be made of statements that have a truth value (true (1) or false (0)). Without truth values, we cannot determine whether the reasoning is valid or invalid. Thus, we cannot determine whether the machine is intelligent or unintelligent. 

PROPOSITIONS – A TECHNICAL TERM

A proposition is a statement that must be either true or false. In other words, it is a statement that has a truth value. 

“I have green skin” is a proposition with truth value = false, unless you are Gamora or The Hulk. 

“I see a mountain” is a proposition with truth value = true, if there really is a mountain, and truth value = false, if there isn’t one. 

“Where are you from?” is not a proposition at all. It has no truth value. It is a question. 

In AI, knowledge systems are always composed of propositions. 

SUMMARY

In this post, which is the first part of the knowledge systems series, we went over several fundamental concepts.

We began by stating that there can be no intelligence without knowledge. Then, we saw that a machine’s ability to intelligently use knowledge can be judged by its use of valid or invalid reasoning. Lastly, we covered the term proposition. 

Propositions, although vital to any knowledge system, are not the all of it. There are other concepts that are equally important. Slowly, as we go through the series, we will understand more and more about how machines currently implement intelligent reasoning based on knowledge systems.