Skip to main content

When are categories useful?

The regularities that we observe, whether in a formalized scientific effort or in everyday casual experience, allow us to induce simplified models of processes in the environment.  Such models are the substance of scientific theories, and are what engineers use to plan and execute designs.  These are also what our brains use to negotiate the environment to get what we want.

The physicist Edward Rutherford was famously quoted as saying, \That which is not physics is stamp collecting.\"  By "physics," I take his quote to mean quantitative modeling of the content of our observations, and by "stamp collecting," I take his quote to mean qualitative categorization of that content.

Humans use language and its almost unavoidably categorical nature to communicate, which may incite us to attempt to interpret all of our observations using categories, whether categories are warranted by our observations or not.

I would propose that categories, with their unitary and symbolic nature, represent a useful form of data compression.  Specifically, some of our observations 'bunch up' or 'cluster' in multidimensional spaces of descriptors or measurements.  When there is such unity among many observations, we may put a stake in the ground there and say there is a 'type' or category that can approximate that unity and say that there is a sameness between those observations.  We might even give that category a name.

The risk involved comes from the fact that such forms of data compression are almost always going to be lossy, meaning that some relevant information is thrown away.  In some cases, nearly everything necessary to a wide spectrum of perception and behavior may be lost.

For example, I recently asked a robotics researcher what fraction of a robot's control systems could function on categorical information about its environment alone.  He estimated that it was on the order of 1 percent.  That's a very narrow portal for interacting with a very analog world.

Have the cognitive filters that underpin language made us blind to the other 99 percent of what our nervous systems are doing?

The author's affiliation with The MITRE Corporation is provided for identification purposes only, and is not intended to convey or imply MITRE's concurrence with, or support for, the positions, opinions, or viewpoints expressed by the author.

Blog Name