What Even Is a Computer?

In On the Arc of the Computer, the focus was largely on interfaces: on how humans encounter computers, how those interaction grammars formed, stabilized, and eventually ossified. That focus followed experience. Interfaces are where computers are actually encountered, and where their leverage and limitations register most viscerally. But in concentrating on how we meet computers, the computer itself remained mostly implicit—treated as an understood object rather than something requiring definition.

So, what even is a computer?

Standard definitions describe a computer as a programmable device that can store, retrieve, and process data. This is accurate, but incomplete in a practical sense. When we talk about other tools, we rarely stop at their raw mechanical description. A hammer is not understood merely as "a hand tool consisting of a solid head set crosswise on a handle and used for pounding"; it is understood in relation to nails, wood, and construction. Maybe a bell. Likewise, “data” is not merely an inert substance. It stands in for something else, its a medium.

More precisely:
A computer is a programmable device that maintains and operates on structured models of a domain, providing means to inspect, modify, and extend those models over time.

In this framing, the model is the object of work. Computation exists to transform it. Interfaces exist to access it. Automation exists to manipulate it without continuous human involvement. Any model is necessarily selective—it captures certain aspects of its domain and omits others—but that selectivity is a feature of modeling itself.

This definition holds regardless of surface form or application. A spreadsheet maintains a numerical financial model. A simulation maintains a causal model. A game maintains a model of a multifaceted fictional world with internal rules. A design tool maintains a model of a possible artifact. Interfaces vary, domains vary, levels of realism vary, but the basic underlying function remains the same: the externalization and manipulation of a structured model.

From this perspective, a simple constraint follows. Any system that claims to support thinking, understanding, or decision-making must make clear where its models live, how they persist, and how they can be deliberately worked on over time.