How AI and copyright would work

Self-aware robots, androids or call-them-what-you-will have been part of science fiction almost from its beginnings. Recently in science reality, there’s been early, speculative discussion about “creative” works generated by these types of machines, and how copyright would apply.

It’s easy — and tempting! — to get wrapped around the axle when it comes to the prospects for artificial intelligence (AI) programs and their creation of original works. When works created by self-running software applications become more common, the result is both more possibilities and more challenges to existing copyright law. But let’s take a step back and consider what we know already, and then move on to what may soon be coming.

Applying today’s law

Think about programs like Google’s DeepMind, or the natural language generation program Wordsmith. These apparently copyright-susceptible expressions result from the operation of application code and algorithms. Because the programs themselves involve no specific human interference or guidance, the results might be best construed as the intellectual property of those who “worked the machine,” i.e. the users.

“Robots” themselves have no actual agency, let alone self-awareness. If there is any human interference or guidance, like a person setting parameters for how a program operates, then that person may be entitled to a copyright if the person’s contribution rises to the level of copyright protection.

When it comes to the rights of an AI itself, I think the key issue would be intent.

Logically speaking, you can’t have original expression without intent, and you can’t have intent without self-awareness. Despite popular articles warning about the coming of AI, a working program with awareness of its own existence and operations is still a long way off — at least according the more skeptical cadre of those studying AI.

Even if an output has the form and feel of a copyrightable work — for example a script written by a program called “Benjamin” and filmed as Sunspring — it may still fail to qualify if we require for expression. In other words, under our copyright rules as they currently exist, to qualify as protectable by copyright, a work — an original creation — needs to meet the definition spelled out in Section 102(a) of the Copyright Act (U.S. Code Title 17):

“Copyright protection subsists, in accordance with this title, in original works of authorship fixed in any tangible medium of expression, now known or later developed, from which they can be perceived, reproduced, or otherwise communicated, either directly or with the aid of a machine or device.”

Can AI-generated works be reasonably construed as original expression, even though there’s no person behind the work doing the expressing?


Interestingly, the Section goes on, in paragraph (b), to spell out certain works of the mind which are not subject to protection by copyright:

“In no case does copyright protection for an original work of authorship extend to any idea, procedure, process, system, method of operation, concept, principle, or discovery, regardless of the form in which it is described, explained, illustrated, or embodied in such work.” (Such “ideas” etc. may be considered intellectual property which is protectable in other ways, e.g. they may be patentable.)

Thought experiments

The real problem of self-aware AI generating original content with intent has not arrived (yet). But it may, and it is interesting to think about.

The core question about AI-generated works is: Can AI-generated works be reasonably construed as original expression, even though there’s no person behind the work doing the expressing?

The terms of the statute also include the word “communicated,” implying that something has gone from the mind of one person (the creator) to another (the audience, the reader, etc.). It would be a stretch to bring in the information theory criteria for communication, i.e. “communication” occurs any time photons or electrons are caused to move from here to there.

That’s one problem: There’s no “ghost in the machine” — the AI is functioning, not thinking. Function, not consciousness, produces a result. Arguably, it has no discernable intent to express anything. (We don’t know if an AI could be incented to produce — only that no extant ones can.)

Of course, laws can be changed, and copyright laws should be changed when new circumstances warrant. Would it be better for society — would it help achieve the purposes of copyright — if protections were extended to AI-generated works? As it says in the Constitution, the purposes of copyright may be thought of as a quid pro quo — in order “to promote the progress of science” certain rights are “secured” to authors [and other creators].

In that frame, copyright protection has both a public and a private purpose. The text of the clause seeks to strike a balance. “The limited scope of the copyright holder’s statutory monopoly, like the limited copyright duration required by the Constitution, reflects a balance of competing claims upon the public interest: Creative work is to be encouraged and rewarded, but private motivation must ultimately serve the cause of promoting broad public availability of literature, music, and the other arts.” (See also: “The rights conferred by copyright are designed to assure contributors to the store of knowledge a fair return for their labors.”)

Logically speaking, you can’t have original expression without intent, and you can’t have intent without self-awareness.


If achieving a “balance of claims” is the core purpose of copyright, the extension of copyright protection to works generated by AI can be measured against that purpose. On the one hand, this extension would likely increase the number of works available for public enjoyment, edification and other worthy social purposes. There’s been a lot of discussion along these lines already, with Google’s Deep Dream-generated images providing the occasion for much of it.

On the other hand, copyright protection for these new works — even at the stage we are now — might increase the profit of one or more private corporations who controlled them and were thereby able to exploit them commercially, in ads or as part of feature films, games and so forth. This seems like a wash to me; a well-written piece of legislation shouldn’t re-tilt the balance.

Bottom line: In my view, a self-aware, autonomous AI would be the prerequisite for its works to be protectable by copyright. At that time, such a revolution in technology would entail a much greater revolution in society, with the law, including copyright law, changing, as well.