The machine author: what level of copyright protection is appropriate for fully independent computer-generated works?

Author:Yu, Robert
Position::COMMENT
 
FREE EXCERPT

U.S. copyright law is grounded in a utilitarian philosophy: authors are granted a limited monopoly to incentivize production of original expressive works for the benefit of society as a whole. This philosophy may need to be applied to non-human, machine authors in the very near future. Works of literature, music, and art are increasingly being generated through the execution of software programs, suggesting that these machine-authored works may become the norm rather than the exception. The burgeoning of computer-generated works raises novel and fascinating questions of copyrightability, but the existing literature neglects to address a basic question: does extending copyright protection to machine-authored works promote or hinder the purpose of copyright law?

This Comment makes several contributions to the scholarship on copyright law. First, it poses fundamental questions regarding how the existing copyright framework would be applied to the various players that contribute to machine-authored works and notes the problematic aspects of such application, particularly in identifying the legal author of the work. Second, it evaluates whether--in the case of machine-authored works--the human author should be allocated rights based on the economic incentive theory. It argues that inflexible application of copyright law creates a contribution/rights paradox because the party that contributed most to the creation of the work--its author--is not the party to whom we would like to allocate copyright protection. Finally, the Comment posits that because copyrights provide little economic incentive to the players involved in creating machine-authored works, it would be inappropriate from a social policy standpoint to extend protection to fully independent computer-generated works.

INTRODUCTION 1246 I. PURPOSE AND LEGAL REQUIREMENTS OF COPYRIGHT LAW 1251 II. MACHINE-AUTHORED WORKS 1253 A. Machine-Authored Works Defined 1253 B. Copyrightability of Machine-Authored Works 1255 III. PROBLEMS OF APPLYING TRADITIONAL COPYRIGHT FRAMEWORK 1257 A. Authorship and Ownership 1257 1. Machine as Author 1257 2. Programmer as Author 1258 3. Joint Authorship 1259 4. Joint Authorship 1259 B. The Contribution/ Rights Parabox 1260 C. Social Policy Arguments for Not Protecting 1263 Machine-Authored Works IV. PROTECTION PROPOSALS DESIGNED FOR MACHINE-AUTHORED WORKS 1265 A. Immediate Entry into the Public Domain 1265 B. End-User: The Quasi-Property Treatment 1266 C. Programmer Allocation: One-for-One Matching 1268 CONCLUSION 1269 INTRODUCTION

When a 2.7 magnitude earthquake hit Southern California on March 17, 2014, the Los Angeles Times published a news report on the natural disaster within three minutes. (1) How the L.A. Times managed to publish a report so quickly borders on science fiction. Moments after the quake, an algorithm called Quakebot scraped data from the United States Geological Survey reports, plugged the information into a coded template, and generated the actual text of the article. (2) By the time journalist and programmer Ken Schwencke had been woken by the quake and had walked to his computer, the text was already on the screen, ready for publication at the press of a button. (3)

While the L.A. Times article was revised and updated over the course of the morning by actual human writers, all of their work built off the foundation Quakebot constructed. And Quakebot is not the only program of its kind. From reports on homicide (4) to college sports statistics, (5) software that automatically generates news stories is becoming increasingly prevalent in journalism. Quakebot's article was not particularly sophisticated; it provided only the magnitude of the quake and its geographic location. (6) But, however simple, banal, or nondescriptive they might be, machine-authored works like the article prepared by Quakebot are becoming indistinguishable from their human-authored counterparts. (7)

As defined by this Comment, a "machine-authored work" is a fully independent computer-generated work. The "machine author" is a software program, like Quakebot, designed to generate literary content on command. The "work" is the byproduct of executing the software programming. Machine-authored works distinctly differ from what might historically be considered "machine-aided" works. A movie edited in the video-editing software Final Cut Pro, for example, would be a machine-aided work but not a machine-authored work. Although the machine computed and created the final product, it was only able to do so at the creative direction of its human operator. Thus, the most notable difference between a machine-authored work and a machine-aided work is that in the case of a machine-authored work, there is no distinct human author driving the creative process through composition, arrangement, selection, or direction.

The burgeoning of machine-authored works raises novel and fascinating questions of copyrightability. Some questions are more intuitive than others. (8) For example, legal scholars going as far back as the 1980s have expounded on the foundational question of whether machine-authored works are legally entitled to copyright protection. (9) Applying the basic inquiries of originality and authorship, many legal scholars have concluded that machine-authored works should be entitled to full copyright protection.

Yet the existing literature neglects a far more fundamental question: does extending copyright protection to machine-authored works promote or hinder the purpose of copyright law? U.S. copyright law is grounded in a utilitarian philosophy: authors are granted a limited monopoly to incentivize production of original expressive works for the benefit of society as a whole. (10) The rationale behind these measures has been intuitive and comprehensive--without such protection, authors would have less of an incentive to continue creating works and the public would suffer from this lack of creativity. (11) The economic incentive theory has gained greater traction as technological progress places authors at a significant disadvantage to potential imitators. (12) Particularly in the digital age, where the marginal costs and sometimes even the fixed costs of reproduction are effectively zero, imitators have a considerable advantage over creators. (13) Consequently, content producers who bear significant fixed costs in production and face uncertain payouts are placed in an even stronger position to demand enhanced copyright protection. (14)

Yet, even within the economic rationale framework, machine-authored works present notable differences to traditional works. For example, with machine-authored articles, both the fixed and variable costs of producing each copyrightable article are effectively zero, which allows producers to compete with imitators even absent legal protection. Additionally, unlike a human author, the software program that constructs the article cannot be legally or economically incentivized to produce more or fewer works. For example, if an online sports writer discovers that he can generate stable income through his copyrighted articles, he is economically incentivized to write more articles. A software program, on the other hand, will follow its programming and generate articles regardless of such economic rewards.

Two counterpoints might be offered to such an argument: first, that copyright protection is designed to motivate the software creator to create more software and, by extension, more creative works; and second, that copyright protection is designed to motivate users licensing the software to generate more creative works. However, these points are unpersuasive in the context of machine-authored news articles. As to the first, copyrightability can be extended to the software without being extended to the articles generated by the software. And as to the second, because control and profitability for modern electronic news depend on being first to market, copyrightability creates little incentive for the software end-user. Specifically, the value of electronic news peaks within the first six hours and then diminishes significantly. (15) By the time copyright protection is secured, the residual value of the article is minimal. (16)

Consequently, stringently mapping the existing copyright framework onto machine-authored works would implicate much of the cost of copyright protection but little of the benefit. For example, under a regime in which machine-authored works are de facto copyrightable, a single individual could, absent any contractual workarounds, own an indefinite number of copyrights. Such an individual could easily behave in ways that would hinder rather than promote future creative efforts. Consider the extreme hypothetical example of media conglomerate ANS with a machine-authorship program. Finding that a startup, FastNews, which produces articles comparable to its machine-authored works, generates more traffic, ANS leverages several copyright infringement lawsuits at FastNews to shut it down. FastNews sells to ANS at a fraction of its valuation after being rendered illiquid by litigation expenses. A legal framework that permits a single party to aggregate a significant number of copyrights through minimal effort readily invites this type of anticompetitive behavior.

This Comment makes several contributions to the scholarship on copyright law. First, it poses fundamental questions regarding the application of the existing copyright framework to the various players involved in creating machine-authored works--particularly with respect to identifying the legal author of the work. Second, it evaluates whether, in the case of machine-authored works, the human author should be allocated rights based on the economic incentive theory. It argues that inflexible application of copyright law creates a contribution/rights paradox because the party that contributed to the creation of the work--its author--is not...

To continue reading

FREE SIGN UP