October 2017

Jon Watkins, MJLST Staffer

A great deal of attention has been paid recently to artificial intelligence. This CGPGrey YouTube video is typical of much modern thought on artificial intelligence. The technology is incredibly exciting- until it threatens your job. This train of thought has led many, including the video above, to search for kinds of jobs which are unavoidably “human,” and thereby safe.

However, any feeling of safety that lends may be illusory. AI programs like Emily Howell, which composes sheet music, and Botnik, which writes jokes and articles, are widespread at this point. What these programs produce is increasingly indistinguishable from human-created content- not to mention increasingly innovative. Take, as another example, Harold Cohen’s comment on his AARON drawing program: “[AARON] generates objects that hold their own more than adequately, in human terms, in any gathering of similar, but human-produced, objects. . . It constitutes an existence proof of the power of machines to do some of the things we had assumed required thought. . . and creativity, and self-awareness.”

Thinking about what these machines create brings up more questions than answers. At what point is a program independent from its creator? Is any given “AI” actually creating works by itself, or is the author of the AI creating works through a proxy? The answer to these questions are enormously important, and any satisfying answer must have both legal and technical components.

To make the scope of these questions more manageable, let’s limit ourselves to one specific subset of creative work- a subset which is absolutely filled with “AI” at the moment- curation. Curation is the process of sorting through masses of art, music, or writing for the content that might be worth something to you. Curators have likely been around as long as humans have been collecting things, but up until recently they’ve been human. In the digital era, most people likely carry a dozen curators in their pocket. From Spotify and Pandora’s predictions of the music you might like, to Archillect’s AI mood board, to Facebook’s “People You May Know”, content curation is huge.

First, the legal issues. Curated collections are eligible for copyright protection, as long as they exhibit some “minimal degree of creativity.” Feist v. Rural Telephone Co., 499 U.S. 340, 345 (1991). However, as a recent monkey debacle clarified, only human authors are protected by copyright. This is implied by § 102 of the Copyright Act, which states in part that copyright protection subsists “in original works of authorship.” Works of authorship are created by authors, and authors are human. Therefore, at least legally, the author of the AI may be creating works through a proxy. However, as in the monkey case above, some courts may find there is no copyright-eligible author at all. If neither a monkey, nor a human who provides the monkey with creative tools is an author, is a human who provides a computer with creative tools an author? Goldstein v. California, a 1973 Supreme Court case, has been interpreted as standing for the proposition that computer-generated work must include “significant input from an author or user” to be copyright eligible. Does that decision need to be updated for a different era of computers?

The answer to this question is where a technical discussion may be helpful, because the answer may involve a simple spectrum of independence.

On one end of the spectrum is algorithmic curation which is deeply connected to decisions made by the algorithm’s programmer. If a programmer at Spotify writes a program which recommends I listen to certain songs, because those songs are written by artists I have a history of listening to, the end result (the recommendation) is only separated by two or three steps from the programmer. The programmer creates a rigid set of rules, which the computer implements. This seems to be no less a human work of authorship than a book written on a typewriter. Just as a programmer is separated from the end result by the program, a writer may be separated from the end result by various machinery within the typewriter. The wishes of both the programmer and the writer are carried out fairly directly, and the end results are undoubtedly human works of authorship.

More complex AI, however, is often more independent. Take for example Archillect, whose creator stated in an interview “It’s not reflecting my taste anymore . . .I’d say 60 percent of the things [she posts] are not things that I would like and share.” The process involved in Archillect, as described in the same interview, is much more complex than the simple Spotify program outlined above- “Deploying a network of bots that crawl Tumblr, Flickr, 500px, and other image-heavy sites, Archillect hunts for keywords and metadata that she likes, and posts the most promising results. . .  her whole method of curation is based on the relative popularity of her different posts.”

While its author undoubtedly influenced Archillect through various programming decisions (which sites to set up bots for, frequency of posts, broad themes), much of what Archillect does is what we would characterize as judgement calls if a human were doing the work. Deeply artistic questions like “does this fit into the theme I’m shooting for?” or “is this the type of content that will be well-received by my target audience?” are being asked and answered solely by Archillect, and are answered- as seen above- differently from how Archillect’s creator would answer them.

Even closer to the “independent” end of the spectrum, however, even more complex attempts at machine curation exist. This set of programs includes some of Google’s experiments, which attempt to make a better curator by employing cutting-edge machine learning technology. This attempt comes from the same company which recently used machine learning to create an AI which taught itself to walk with very little programmer interaction. If the same approaches to AI are shared between the experiments, Google’s attempts at creating a curation AI might result in software more independent (and possibly more worthy of the title of author) than any software yet.