Written computer source code has to be translated into a format the computer can understand. That format, binary, is often outside the scope of review for copyright assignment. No one can really read binary and as an inherently mathematical description, is excluded from even patent provisions governing systems described mathematically.
Written computer source code is not a system but it is the input into a systematic process. That is its primary purpose for being. An API written in source code is intended to be translated into a system of use or as a method to teach and describe systems. Therefore, its purpose for being excludes it from copyright protection under section 102(b) of the copyright provision.
Brought to my attention in a comment on Ars Technica. A separate thought process does exist that calls into question the applicability of copyright to software API based on a principle of publication in terms of the cost of distribution and shared cost between publisher and author versus how that transpires in a more tangible medium.
Rather, the language of existing copyright provision declares the ineligibility of software API, and indeed all software, to copyright protection, at least in my reading of it. Everyone still puts copyright notices on their software but the intent of distributing software is to automatically share it without restriction except to ensure proper citation of the source of authorship.