I’ve been writing a lot about concepts, conceptual spaces, and my formalization of concepts as fuzzy star-shaped sets. But why do we need concepts in the first place? Why are they useful? I would like to give a short answer to these questions in today’s blog post.
As you might recall from one of my earlier blog posts, a concept is something like the abstract idea of a specific category of things. For instance, the concept of an apple is our mental representation of what apples are like: What they look like, how they taste, what we can do with them, etc.
Concepts allow us to categorize our world and to make sense out of what we see, hear, feel, and experience. But why exactly are they so crucial?
Well, consider the alternative: You don’t know anything about concepts like “apple”, “school”, or “friendship” – and are only able to observe what you directly perceive. You only see a roundish, reddish thing with a diameter of approximately 7 to 8 centimeters. How useful is this information by itself? What can you infer from this? If you know nothing about apples in general, not much at all!
But if you are able to categorize this thing as “apple”, you all of the sudden know a lot more about it: that you can eat it, that it tastes sweet, that it is considered healthy, that you can make pie or juice out of it, that you can throw it (and that it probably hurts if someone throws one at you), and many more things. All of this information belongs to the concept of an apple – you know what apples typically look like, what you can typically do with them, how heavy they usually are, etc.
Moreover, just by knowing that this thing is an apple, you also know that it must be a fruit. So if someone sent you to bring some fruits, this realization that the object you are seeing is an apple and therefore a fruit is quite helpful.
And there’s even more: Concepts don’t just store information about typical members of this category, they also provide us with information about correlations. For instance, based on the observation that the “apple” object you are looking at is red, you can be quite sure that it is ripe and therefore sweet. If it was green, you would probably conclude that it does not taste sweet, but sour. This information of typical co-occurrences like “red-sweet” and “green-sour” is also part of the “apple” concept.
Thus, concepts help us to organize the world in our minds. There has been plenty of psychological research on concepts, and I think that concepts are also important for artificial intelligence: Given their important role in the human mind, any artificial mind is also likely to need a similar mechanism in order to become really capable and understanding the world and acting in it.
And this is exactly where my PhD research enters the stage: I work on implementing the conceptual spaces framework, which was derived mainly from insights about human conceptualization. I want to find a mechanism for discovering useful concepts and to equip a computer system with the ability to learn about concepts in the world. In my opinion, this is an important step towards artificial general intelligence.