Cognitive Science: An Introduction/Cognitive Architectures Implemented As Software

As an architecture for mind is supposed to describe the basics of how all (human) minds work, a software architecture is a implementation of such a theory. The idea is that using the architecture, a scientist will build a model of a particular phenomenon. For example, doing arithmetic is not a part of human nature, it's learned. So a scientist making a model of arithmetic might build a piece of software that does arithmetic the same way a person does. But each modeling shouldn't have to create a memory system, an attention system, a perceptual system, and so on--they are just interested in arithmetic. So the idea is that the scientist could use an architecture to take care of the basics, and in that architecture create a model of a specific task.

Not all modeling works this way. Many models are made without any architecture at all. These models either replicate functionality in things like memory systems, or simply do not bother to make the memory system realistic. This might be acceptable in some cases, because the way memory works might not be important for a particular task. But a larger problem for the field is that when many of these isolated models are created, without consideration for how they make sense give other things we know about minds, they contribute less to our understanding about how minds work as a whole.

In a way an architecture is like using a high-level programming language.

The Common Model of Cognition

Several cognitive architectures have been created. In the late 2010s the creators of some of the major architectures (such as ACT-R and Soar) got together and created a theoretical, high-level theory of how cognition works, based on what all of these major architectures agreed on. This became known as the Common Model of Cognition.