I have done a lot of thinking about this particular word as a psychologist and as a technologist. It was enigmatic for a long period of time, but resolved more clearly as I studied both psychology and AI.
I define a concept as a label that is applied to an extracted set of correlated features. It is like creating a new object composed of abstract elements that encompasses more than one external objects or ideas.
Consider the concept of fruit.
Fruit (noun). In botany, a fruit is the seed-bearing structure in flowering plants (also known as angiosperms) formed from the ovary after flowering.1
The extracted correlated features are that it comes from a seed bearing structure in flowering plants and is formed from the ovary. The extracted features are themselves concepts. Different concepts may vary in the degree to which they rely on abstract or concrete elements.
In order to achieve AGI, rich models need to be built and language labels attached to the features in the form of concepts (correlated extracted features of abstract, concrete, or mixed features). One could imagine an AI system that uses unsupervised learning to extract and group features. It would be possible for AI to make up it’s own language labels as concepts to communicate correlated features. That would be less useful and instead we probably want AI to use our same labels.
References:
1). Retrieved online from Wikipedia. Fruit – Wikipedia
