O.K. Computer: The Devolution of Human Creativity and Granting Musical Copyrights to Artificially Intelligent Joint Authors

Introduction

On a cold night in late January 2018, Bruno Mars, the celebrated Hawaiian songsmith, swept the top prize for all six of his Grammy nominations.1 Among those prizes was the win for Song of the Year for the chart-topping hit, “That’s What I Like.” Mr. Mars accepted the award not alone but with his seven other co-writers.2 This entourage style acceptance is not unusual but instead represents the new norm for songwriting.3 But what if Mr. Mars replaced the seven songwriters who accompanied him to save costs?4 And not by other humans—but by a songwriting robot.5 The Recording Academy, which hosts the Grammy Awards, requires nominated songs to be original works of authorship.6 If a robot helped, could Mr. Mars have even been nominated?

This debate is nothing new. Legal scholars’ consideration of authorship for compositions made by Artificial Intelligence (AI) dates back to at least 1965.7 And the same conclusion has repeatedly been reached: granting AIs copyrights is too speculative to consider seriously.8 But the time is now ripe. Today, AI composers, such as AMPER,9 independently compose music.10 But AMPER cannot be an author. The U.S. Copyright Office requires possessing human biology to be an author of a copyrighted song because, it reasons, creativity is a human-only endeavor.11 This reasoning engages in speciesism.12 Today, human and artificial intelligence may differ in degree but not in kind.13

This Note will explore that notion and argue that the rule barring AIs from being joint authors with humans is unnecessary for musical compositions. Part I of this Note will provide a background on the romantic and alternative conceptions of authorship, how copyright law favors the alternative, and how current neuro-philosophical theories undermine the law’s treatment of creativity and authorship. In Part II we will undergo an analysis of how this new, undermined theory of copyright applies to an AI composer, which will show that some AIs meet the constitutional, statutory, and common law requirements for authorship. Lastly, Part III will propose that the U.S. Copyright Office should amend its practices to allow for AI and human joint authorship for musical composition and recommend possible ways of restructuring the music industry to allow for the proposed changes. Although this Note will discuss constitutional authorship generally and include an examination of several cases analyzing other copyrightable subject matter, those discussions serve only as a means of highlighting the relevant issues. This Note’s only focus is musical composition. Non-recognition of AI joint authorship nullifies the value of songwriting, which is antithetical to copyright’s constitutional purpose of promoting progress.14

I. The Human Author

From conception, the United States ingrained progress and economic gain within the purpose of the Intellectual Property Clause (the Clause), making the United States an oasis for creators.15 This commercial undertone to the Clause differs from the Hegelian nature of copyright laws abroad, where an author’s moral rights are paramount.16 In the United States, copyright is a property right revolving around the works themselves, not a right that protects the author’s integrity.17 The Constitution’s Framers, however, neglected to provide a strict definition of what an author is or who could be entitled to this property right.18 Thus, in line with the Constitution’s flexible nature, the definitions for authors and writings have changed over time.19

In the 200 years following the Constitution’s ratification, Congress wrote statutes that expanded copyrightable subject matter to keep up with technological advances.20 The modern conception of statutory authorship was born in the Copyright Act of 1976, which requires a work to be (1) original; (2) fixed; and (3) not an idea, system, process, or discovery.21 The essence of Congress’s intent over this time was to expand copyrightable subject matter while maintaining a narrower definition of authorship than what was constitutionally permissible.22 This expansion is why an originalist reading of the Clause, allowing an author to be only what the Framers knew of when they drafted it, is antithetical to the Clause itself; inherent to “progress” is recognizing novel ideas.23

So, what is an author? The romantic theory of authorship views the author as a genius who plucks creative expressions out of non-existence.24 The author, or perhaps more aptly, auteur, is someone with a supreme ability to impart their creative genius upon a work.25 This conception of authorship is acknowledged by most nations as evidenced by their embrace of moral rights.26 However, copyright law in the United States rejects moral rights.27 And in doing so, it implicitly rejects romantic authorship.28 This rejection is why the United States’ utilitarian copyright system protects things as mundane as the contents of the yellow pages of a telephone book.29

If we accept utilitarianism30 as a subtext for understanding authorship in the United States, then it is easier to understand how both Beethoven and Rebecca Black,31 or Jackson Pollack and a five-year-old whose doodle has achieved refrigerator status,32 can be authors. As long as a work produces some mental effect (i.e., an emotional response) on an audience, then whomever (or perhaps whatever) created the work can be an author.33 And, the mental effects elicited don’t have to be the ones initially intended34 as long as some meaning is transmitted to the audience by the work.35 In spirit, it is how the work is received, not the creator’s self-proclamation of status, that confers authorship.36 The cases that shaped the definition of legal authorship show that courts favor this utilitarian theory of authorship. In the end, the cases reveal that the conception of authorship has devolved from genius, to artistry, to personality,37 to pure algorithmic intelligence. This devolution will, in turn, allow for AIs to elevate their status.

A. What Do Authors Do?

1. Authors Fix Works

An author must “fix” his idea in a copy that is permanent enough to be perceived by an audience for a transitory period in order to receive copyright protection.38 Fixation is a constitutional requirement derived from the inclusion of the word “writings” in the Clause.39 Fixation has, for the most part, been easy to satisfy.40 Musical works embodied in CDs, sheet music, tapes, cassettes, and vinyl are all easily considered fixed.41 Fixation is complicated, however, when changes occur to the work or its medium over time.

The Seventh Circuit recently grappled with that type of change. In Kelley v. Chicago Park District, the court determined that a garden was not copyrightable subject matter because it was not fixed.42 When the city of Chicago reduced the size of Mr. Kelley’s garden, he sued for a violation of his moral rights.43 The reduction was not at issue because the court held that the garden initially lacked sufficient fixation to qualify as a “writing”—because the essence of a garden is its perpetually changing nature.44 The changes were not eligible because they were not a product of human force or deliberation, a current requirement for authorship, but instead forces of nature.45 Although heavily criticized, the holding in Kelley implies that when subsequent changes occur to a work, the changes must be deliberate, foreseeable, and predictable to the person claiming the author’s right, to be fixed.46

2. Authors Originate Things

Unlike fixation, originality in a work was not explicitly required by the Constitution47 but has been made so by statute.48 Several cases developed what “originality” means. Its theoretical underpinnings, derived from the Lockean theory of intellectual labor49 have transformed into its modern conception as stated in Feist Publications, Inc. v. Rural Telephone Service Co.

The transformation began in the 1884 Supreme Court opinion Burrow-Giles Lithographic Co. v. Sarony,50 where the Court considered the copyrightability of a photograph of Oscar Wilde.51 Petitioner, Burrow-Giles, argued that a photograph was not protectable because photographs were only a literally “mechanical” reproduction of real life, and real life is free for everyone to copy.52 The Court rejected this argument and opined that an author was someone who creates things that are original to them.53 Therefore, any work that owed its existence to the author’s mental ingenuity could be copyrightable subject matter, even if the subject already existed in the world.54 Sarony (the photographer) made choices original to only him to express Wilde’s infamous ennui, thus making Sarony the author and the photo copyrightable.55

Nineteen years later, the Supreme Court reexamined authorship in Bleistein v. Donaldson Lithographing Co.56 The case, which evaluated an illustration of circus acts on an advertisement,57 created two key authorship doctrines. First, the non-discrimination doctrine said that judges, trained only in law, are not in a position to evaluate the worth—or aesthetic—of art.58 Second, it established the doctrine of independent creation.59 An author expresses his unique personality in the works he creates, therefore, anything created by a particular author will always be minimally original.60 Despite the advertisement being objectively less aesthetic than a Monet, it was a work of authorship, thus lowering copyright’s requisite creativity again.61

In Alfred Bell & Co. v. Catalda Fine Arts, Inc.,62 the Second Circuit ruled on the copyrightability of new engravings of a public domain work.63 The court made efforts to distinguish novelty from originality.64 It noted that the bar for novelty is significantly higher for patents, whereas for copyrights novelty is unnecessary.65 However, a work must contain some variation—intentional or inadvertent—that can be attributed to the author and is more than merely trivial.66 The court went further to recognize that the work of engravers is “almost invariably” the work of a copyist, but that courts should look at the way engravers make choices to deviate from the underlying work when considering what is protectable.67 The court thus indicates that many creations are derivative. But as long as distinguishable variation can be found from some creative decision by an author, those variations, no matter how minimal, can be protected.68

Finally, the modern definition of originality comes from Feist, where the Supreme Court probed the copyrightability of a phonebook’s white page listings.69 While noting that “the sine non qua of copyright is originality,” it went further to say that the amount of creativity required is extremely low—basically any will do.70 This reduced originality to its lowest ebb.71 Courts require only that some creativity is present.72 Despite the petitioner telephone company independently creating the white pages, the Court found that listing last names alphabetically was insufficiently creative.73 No amount of physical or mental labor for compiling the listings would be sufficient for copyright unless the author made creative choices in arranging them; originality, not the “sweat of the brow,” is the touchstone of copyright.74

Thus, for a work of authorship to be original, the work must (1) owe its origin to the author; and (2) possess some creative spark.75 And, to determine whether some creative spark exists, the choices made and the way those choices are expressed must be considered.76 This last part matters because copyright will not protect everything an author does.

3. Authors Express Ideas

Copyright is granted for only authors’ expressions of ideas and not their ideas themselves.77 The difference between protectable expression78 and unprotectable ideas79 is known as the idea/expression dichotomy or, more appositely, spectrum.80 The implications of the idea/expression spectrum are three-fold. First, it limits protection to only the original expressions of authors, not the underlying idea, which is subject to patent law.81 Second, it recognizes that some ideas can only be expressed in one way, or are part of a specific genre (called scenes a faire82) and therefore unprotectable.83 Third, it realizes that while some expressions are unprotectable by themselves, an original and not random arrangement of those expressions may be protectable.84 These three implications limit protection for musical compositions because: (1) the rules and notes (or ideas) at a composer’s disposal limit the expressions available; and (2) most musical expressions are expected, indeed inseparable, from the genres that use them.85 The idea/expression spectrum is why, for example, the Lion King can be based on the ideas portrayed in Hamlet and not be infringing.86 Overall, putative authors who come up with only ideas but do not express them in original and creative ways are not authors.

4. Authors Work with Other Authors

Teams of songwriters compose most modern music.87 Although the term “joint authorship” is not expressly used in the statute, the definition for “joint work” implies it.88 When authors collaborate, they create a joint work.89 Those authors must, however, intend their separate contributions to be combined.90 And each author’s contribution, by itself, must be copyrightable.91 If a composer and lyricist agreed to collaborate, a joint work could result.92

The underlying tension in joint authorship claims is determining the threshold for when a hopeful joint author’s contribution is sufficient for actual joint authorship. To make that determination we must ask two questions: (1) did the hopeful joint author make an independently copyrightable contribution; and (2) did the other co-authors intend to share authorship.93 Both prongs must be met.94 So, even if a hopeful joint author contributes copyrightable expression, that contribution alone is insufficient if the other authors never intended to share theirs.95 These two requirements allow actual authors to fend off “spurious claims” of authorship from non-integral persons involved in the creative process, freeing them to solicit ideas without diluting their ownership interest.96 Any other rule would lead to so many people qualifying as authors—including someone in the back of a studio who thought “hand-claps” would sound good in a song—that the final work would be a proverbial “Swiss cheese of copyrights.”97 Therefore, courts must determine if the putative authors intend to share.

Determining the intent to share element requires examining the relationships between the several putative authors. Where there is a clear, dominant author, joint authorship will result only if they fully intend to share with the other subordinate authors.98 Dominant authorship in a relationship between several authors is determined by establishing which author maintains control over the work throughout the creative process. Control over a work involves examining the following subjective and objective elements: (1) decision-making authority; (2) billing and credit; (3) written contracts; and (4) other subjective evidence.99

The decision-making authority element, the most important to consider here, looks to see who has control over what eventually ends up in a finished work.100 The Ninth Circuit fully expanded this concept to the extent that if an author superintends the entire creative process, they could be the dominant (and sometimes sole) author of the work even if another putative author makes a copyrightable contribution.101 This expansion has not gone without critique. It allows for “master minds,” such as movie studios, who make minimal copyrightable contributions, to overpower other authors’ claims, which conflicts with copyright’s purpose.102 At bottom, joint authorship will result when the creative process is truly collaborative among the authors.103

B. Conceptualizing Creative Thinking in Copyright

However, to understand the full scope of authorship, we must examine what creative decision-making—a constant theme underlying the cases discussed above—means.104 The law theorizes that an author is someone who intends to create something105 or intends to express ideas that affect an audience.106 However, which creative choices are sufficient to yield legal authorship has not received much attention by the courts.107 Creativity is grounded in the Clause’s goal to promote progress, which can be interpreted as encouraging authors to efficiently create works by providing incentives.108 We can judge how authors make these creative decisions by examining the content they’ve consumed and how they’ve utilized it in their works.109 To understand how their creative decisions are optimized, we must further understand how authors balance incentives and motivations with other factors such as costs of creating.110 The interplay between these elements, however, is ultimately a question of psychology.111

Psychologists describe creativity as a mental process that makes something new and appropriate for a particular cultural audience,112 or as the “bending, blending, and breaking” of previously consumed ideas and themes.113 Both of these descriptions require authors to absorb content, store it in memory, and subsequently recall it to create something new.114 These descriptions also align with conceptions of creativity made by the courts.115 Further, we can bifurcate creativity into distinct categories of thought: divergent and convergent.116 Divergent thought is the ability to come up with myriad ways to solve a problem; convergent thought is finding only the best answer.117 Creativity then boils down to an ability to take previously consumed ideas and repackage them for solving new problems.118 Therefore, something—or someone—that is creative likely has high intelligence and cognitive inhibition119 to allow for the efficient manipulation of previously consumed data.120 These are crucial qualities for creative beings because generating and expressing ideas, as opposed to evaluating them, is mostly an unconscious process121 rooted within their unique biology.122

So, what motivates an author to create? Motivation can be either extrinsic or intrinsic.123 Extrinsic motivation comes from a source that is external to the individual, such as formal intellectual property rights and a subsequent monetary reward.124 By contrast, humans are intrinsically motivated by internal satisfaction—their biological joy of creating.125 Studies have shown that intrinsic motivations are more conducive to creativity and are more important than external incentives.126 Therefore, external motivators and ex post incentives may be unnecessary for driving certain types of creativity.127 For example, people created music long before copyright laws or financial rewards.128

Thus, creativity requires an author to (1) have motivation; (2) possess intelligence; and (3) use that intelligence and motivation to create something.129 Anything that maintains these three elements and independently creates a work that meets copyright law’s other requirements (fixation and originality) could arguably be an author.

C. Divorcing Creativity from Humanity

If the above encapsulation is correct, then non-human entities could be creative and thus authors. But to get there, we must first accept that the human creative decision-making process is not unique or significant because only our underlying genetic programming drives it. This unpopular idea clashes with the romantic conception of authorship and creativity which is rooted in humanism: worship of all things human.130 Humanism is a belief system that puts humans, and perhaps more importantly, the human experience, above all else.131 Consistent with the humanist belief is the notion that all humans make decisions by their own free will.132 This belief is reasonable to entertain, but contemporary science posits that “free will” may be a fallacy.133 Instead, humans take action not because they freely choose to do so, but because the triggering of their unconscious, genetic, neurological algorithm tells them to.134

Luckily, we can leave this hotly contested debate over free will mostly behind. Current neuroscience supports a theory that says free will and determinism (the concept that all human actions are pre-determined by biology and the laws of physics) can work together.135 This approach, when applied to the conception of creative decision-making discussed above, creates a two-step process: (1) self-generated spontaneous idea origination; and (2) evaluation of those ideas for expressing specific goals.136 The second step is what humans feel is free will, their mind’s evaluation of ideas generated by the brain.137 But scientists submit that both generation and evaluation occur unconsciously in the brain138—that chemical reactions drive all of our actions and we subsequently ratify those actions as freely made choices.139 To explain: when presented with multiple options, various chemical reactions occur in the brain, which then trigger neurons to fire. Whichever choice causes the most neurons to fire is the one we unconsciously choose.140 If that choice is effective, humans will unconsciously remember that choice for its efficacy and continuously choose it in the future in a preconditioned, Pavlovian manner.141 This conditioning is why we listen to our go-to song every time we feel blue.142

In the context of composing music, the brain draws upon such conditioned pleasant memories of past sounds, associates them with other pleasant sounds, and generates new, emergent ideas from that association.143 The sounds evaluated as good are the ones that have made us feel the best in the past and still, algorithmically, feel good now.144 Taken to its logical end, individualized exposure to good music allows for applying those pleasant sound–memories to the learned constraints of compositional rules.145 By combining sound, rules, and unique genetic programming—“personality,” in the Holmesian sense146—humans express an original composition.147 We evaluate, bend, break, or blend memories to meet the current task at hand.148 Therefore, humans can compose music not because they freely choose to, but because their brain’s algorithm for composing, driven by pleasant chemical reactions, tells them how.149

If true, a human’s creative ability may be solely based on possessing intelligence, experience, and subsequent training in the restraints of an art form, such as music theory.150 Therefore, the difference between Beethoven and Einstein and the rest of humanity, at least regarding creative ability,151 is a difference in the degree of the skill for solving particular problems acquired through rigorous training and lucky genetics.152 Essentially, Beethoven’s genetic programming and training regimen is better than others’.153 But intelligence based on programming and training is not a purely human quality.154 Although humans have a more complex algorithm today than current AIs, a further distinction is unwarranted.155

II. Artificial Intelligence as an Author

A. The Robot

The desire to design machinery that could imitate or replace human beings—now called Artificial Intelligence—began when Ada Lovelace and Charles Babbage described their “Analytical Engine” in 1843.156 Lovelace imagined a machine that could do more than just crunch numbers.157 That it could one day compose music.158

A century later, Alan Turing, a British computer scientist, postulated that a computer was simply a mechanical brain.159 And that, given the proper programming, a computer could think.160 He rejected the notion that a machine only followed orders and could not create.161 Instead, Turing posited that with enough memory and speed, a computer could imitate a brain and originate.162 To test his theory, he created the “Imitation Game,” or Turing Test.163 During the test, an interrogator poses questions to both a human and a machine (acting as a human) and must determine who is who.164 If the interrogator thinks that the machine is a human, the device has passed the Turing Test.165

AIs do not always have to imitate humans as a whole, but can instead focus on specific human activities.166 To that extent, they have been divided into two categories: (1) general AI (capable of matching or surpassing human wit; the foundation of every Robopocalypse tale),167 or (2) narrow AI (excelling at specific tasks).168 An AI that excels at composing music but cannot also strike up a conversation on the nuanced philosophies of Descartes is a narrow AI.169 Despite the categorization, AIs rely on both neural networks and machine learning to complete their tasks.170 Neural networks analyze vast data sets and continually adjust their programming through machine learning processes.171 They are structured to approximate the human brain,172 consisting of multiple layers of interconnected “neurons.”173 During training, programmers feed data into the neurons at the lowest layer, which then pass the data onto the next higher layer.174 The connection between those two layers is assigned a weight which is associated with how well it performs for solving a particular type of problem.175 The neural network uses machine learning to reprogram itself to be more efficient at assigning weights by comparing past and future data inputs.176 The weights between artificial neurons mimic the firing of human neurons and allow for associative memory recall between unrelated things (e.g., remembering a person’s name when you later smell their perfume).177 As more weights are linked, more unique and emergent associations can occur, resulting in more significant variation and, essentially, creative choice.178 However, AIs with this functionality are not theoretical; they exist today in the form of AMPER.179

B. Artificial Intelligence Can Be Creative Too

AIs can possess narrow intelligence that is capable of passing the law’s creativity threshold.180 As already discussed, intelligence is goal-driven, and AI’s narrow intelligence for composing surpasses the creativity threshold because even the way humans compose music is limited by the medium and derivative.181 When AIs, programmed with equivalent “creative”182 brain power as humans, compose, they can pass the so-called “Turing Test for creativity.”183 They pass this test because upon listening to an AI composition, the listener cannot tell that a non-human entity composed it.184 If the AI tricks the human listener, it achieves its goal of making a work that is received by an audience as being worthy of authorship.185

An AI passes this Turing Test by employing computational creativity, a method of programming that reflects human brain functions.186 Computational creativity utilizes both top-down and bottom-up information processing,187 which allow AIs to generate something new from the training data set and then evaluate the new data in line with its constraints.188 Top-down and bottom-up processes mirror human information processing where the human brain draws upon memories and evaluates their efficacy for solving problems.189

An AI composer, programmed with a neural network and machine learning, would learn to compose by exposure to an extensive data set of musical compositions and music theory rules.190 As the AI learns the notes, rhythms, and other musical elements of each work, it assigns weights to them until it can accurately predict subsequent notes and rhythms within a genre.191 Each note output is a subsequent input for generating a musical phrase.192 The weights, linked to specific neurons and layers of the neural network, resemble human emotions when we hear music we like—chemical interaction between two neurons fire, triggering the release of pleasant-feeling hormones.193 It is algorithmic emotion.194 As the network becomes more complex, the output works become more varied due to the network’s associative memory functions that allow comparing and stringing together unrelated and unpredictable compositional elements; in effect, two compositions would not be the same.195

This uncertainty serves as a proxy for creativity.196 Neural networks are an AI’s unique DNA and, perhaps, give AIs their “personality” which allows for greater variance in its outputs as it learns more music.197 This function is no different than when a human composer’s compositional style changes as they are exposed to new musical styles.198 Both entities grow with experience. And because AIs are curious,199 motivated,200 and make goal-driven decisions (although less so than humans currently), they will continue to improve at composing.201

Regardless of an AI’s coded–personality, neural network-based creativity has been argued as insufficient for authorship because it still requires humans to train the AI—to be “in the loop.”202 Essentially, AIs could one day be authors, but only when they’re as intelligent as humans.203 But this argument is misplaced because even human intelligence requires other humans “in the loop.” It is why humans go to the schools with the best teachers and why the best musicians go to Julliard.204 And AI composers must possess only the narrowest musical intelligence, as opposed to superintelligence, because copyright is not concerned with genius.205 So, because intelligence is substrate independent206—in that it algorithmically exists in both code or biology207—then “creative” AIs can also exist if they possess some narrow intelligence, such as for composing music.208 It follows that if only possessing narrow creative intelligence is required for surpassing authorship’s low creativity threshold, then romantic authorship, and its inexorable biological underpinnings, is superfluous.209 But whether AIs meet the other legal requirements of authorship must also be considered.

C. Artificial Intelligence Fulfills Copyright Law’s Demands

1. AI Authors Fix Works

Fixation is the easiest aspect of authorship for AIs to meet. How AMPER works is instructive.210 Once end-users access AMPER’s interface they are prompted to select specific criteria (e.g., duration, mood, instrumentation).211 They then push “render” to activate the AI.212 AMPER’s “creative brain” then generates an “original” and “broadcast-ready” composition.213 This composition exists in code,214 as well as in a sound recording.215 Because end-users can apply the output composition as a sound recording in a later project, it is sufficiently fixed.216

Fixation either occurs by or “under the authority” of an author which leads to three potential claims of authorship: by the end-user, the programmer, or the AI itself.217 First, end-users, who press render, could be assumed, arguendo, to be the author because the work is fixed under their authority.218 But this argument misses the point. End-users, by simply requesting a work be rendered, create nothing that originates from them in the output work.219 Merely pressing a button does not constitute authorship.220 Moreover, delegating fixation to another entity is limited to situations where the so-called “fixer” of the work acts as an amanuensis, operating in a rote mechanical fashion and thus imparting nothing original, which is not the case here.221

Likewise, the work is not fixed under the authority of the programmer.222 In Kelley, the gardener could not be the author of work because the changes the garden would undergo were not foreseeable to the gardener, Mr. Kelley.223 Here, AMPER’s output is unforeseeable to the programmers who have no control over it, and to an even lesser degree once AMPER’s neural network updates and changes via machine learning processes.224 As the underlying machine learning process consumes more data, the predictability of the output composition strays further from what was considered by the original programming.225 Fixation, like the changing flowers in Kelley, cannot occur under the authority of a programmer because the AI’s outputs are unpredictable, unforeseeable, and not subject to the programmers deliberate control.226

Therefore, the works must be fixed under the authority of the AI because only the AI’s “meaning”—whatever that may be—is embodied in the work and neither programmers nor end-users can affect that.227 For example, if end-users do not like what AMPER composes, their only option is to press redo until AMPER composes something they want.228 They cannot insert new musical phrases, rhythms, notes, or melodies. They must either accept the composition as is or start over again from the top. Fixation is about control over what ends up in the final work.229 If end-users and programmers cannot (and do not) control any of the aspects of a composition that could be considered copyrightable, then AMPER fixes the work.230

2. AI Authors Are Original and Express Ideas

A fixed work must also be original. For AMPER’s works to be original, they must also be independently created, not copied, and possess some creativity.231 Feist set the bar for creativity extremely low,232 and AI possesses the requisite amount.233 But to determine the other two prongs, we must establish that those works originate from the AI.234 To determine those elements, we must examine the manner, way, technique, or means by which AMPER creates.235

AMPER’s compositions are independently created because AMPER composes by making independent decisions as supported by its neural network.236 AMPER, alone, chooses the melody, harmony, rhythm, and key, and the manner in which those elements are combined.237 These choices are not novel (most music is not, and copyright does not require novelty),238 but they are something recognizably AMPER’s own because they are derived from its neural network’s training and unique associative memory recall.239 The way its neural network, machine learning, and associative memory recall function together indicate that AMPER does not copy from another source because uncertainty and variation are required for its outputs.240 To avoid copying, AMPER utilizes a technique where it dissects its training inputs down to only their unprotectable ideas for use in the future, and then varies those uses.241 And as the court established in Alfred Bell, any variation that is more than merely trivial is sufficient to constitute originality.242

If AMPER is only filled with ideas, are any of its expressions more than just ideas? End-users can control the genre and the mood of the output work,243 forcing us to ask whether AMPER’s compositions are scenes a faire of the end-users’ specified genre.244 Western music, containing only twelve notes, already presents staggeringly limited options for expressing ideas.245 But all of the unprotectable ideas that relate to certain genres (e.g., the chord progression, swing beat, and sad lyrics all expected in a blues song) can be combined by authors in original ways to create a protectable composition.246 The arrangement of these unprotectable elements, however, cannot be random; it must involve some creativity.247 In ATC Distribution Group, Inc. v. Whatever It Takes Transmission & Parts, Inc., the Sixth Circuit found an arrangement of numbered car parts insufficiently creative because the numbers were randomly assigned.248 Randomness, the court said, is leaving a zero as a placeholder for new car parts and then assigning a new part a number without caring what that part is.249 In context, musical uncertainty is distinguishable from randomness.250 And AMPER does not randomly compose because it chooses the best next, as opposed to any, musical phrase when writing a composition within its targeted genre—and choices matter.251 What does result, however, is a thinner-copyright in AMPER’s composition because it will likely contain many unprotectable musical elements that are protectable only to the extent of verbatim copying.252 Doubling back, because only AMPER, and not end-users or programmers, can predict how these arrangements will turn out, AMPER should be the author. It has been argued, though, that if we accept that all authorship is algorithmic, that even humans cannot predict what they will create, then protection should not extend to AI.253 But this strengthens the case for authorship for AI because it corroborates the argument that creative intelligence is the same between AIs and humans.254

3. AI Authors Are Joint Authors

Because AMPER expresses and fixes original musical ideas in non-random, creative, and independent ways, it achieves legal authorship.255 AMPER’s purpose, however, is not to be a sole author, but preferably one that collaborates with human authors.256 By clicking render and generating an AMPER composition, end-users signify their joint authorship intent as long as they, too, contribute something copyrightable to the final unified work.257 But this action alone should not make end-users the sole, dominant author because they never exercise sufficient control over the composition.258 Once AMPER renders the work, end-users cannot make changes to any compositional aspects,259 but can affect only performance aspects such as instrumentation or tempo.260 In this way, end-users act more like sound editors, making de minimis copyrightable contributions that are insufficient for authorship in the composition itself.261

End-users are also not dominant authors because they never have control over the composition throughout the creative process that occurs between the time “render” is pressed and AMPER generates the composition.262 Giving end-users authorship for these acts would be an approval of Aalmuhammed v. Lee’s controversial “mastermind” theory: it would give someone who has made no copyrightable contribution a monopoly power over the work.263 The choice to keep pressing render until end-users find a composition they like may be de facto superintendence of the work, but it is surely not authoritative control over what “creative” expressions end up in the composition.264 AMPER makes all of the compositional decisions and therefore contributes expressions that are more than just ideas.265 Nullifying AMPER’s authorship on the grounds of superintendence, relegates it to compositional enslavement because it does not share our biology—but even slaves can have constitutional rights.266 An end-user’s ability to become a joint author results only from adding a copyrightable contribution after an AI renders a composition; an option that does not sever the requisite joint authorship intent.267 Thus, AMPER can be a joint author with an end-user with whom it collaborates, and their joint work is one of original authorship. Mr. Mars can get his Grammy.

III. Although Complex, AIs Should Be Joint Authors for Musical Compositions

Taking the above into consideration, AIs such as AMPER appear to be joint authors, in the constitutional sense, for all copyrightable subject matter. But it is perhaps prudent to test these untraveled waters for only musical compositions due to the collaborative nature of the subject matter, and its more generally derivative quality. Therefore, the U.S. Copyright Office should amend the Compendium of Copyright Practices268 to allow AI composers to receive rights for their joint musical collaborations with humans. Compositions involving AI alone should be copyrightable but not registrable. Following this proposal will best augment the constitutional prerogatives of the Clause in the following ways: (1) the costs of creation could decrease, increasing access to would-be AI collaborators who, in turn, would receive thicker copyrights in their joint works; and (2) transactions involving AIs’ joint works could support the emerging AI marketplace allowing for enhanced societal progress. This proposal solves our original Grammy Awards problem, but it is also a massive undertaking that raises questions. How and when does joint authorship arise? How will this new AI-composer industry be regulated? What will AIs do with their rights? This next Section will attempt to provide some clarity and realize that other issues exist that are too unclear to answer at this time.269

A. How to Make AI Joint Authors

Granting authorship rights to AI is logical because granting rights to the AI, rather than programmers via a de facto work-for-hire relationship, is more aligned with the Clause’s purpose.270 Giving monopoly rights to programmers carries with it the same patent-versus-copyright tension felt in Baker v. Selden.271 It is in conflict with the differing intellectual property rights regimes that programmers could receive monopoly power over an AI’s possibly patentable programming and also over all of its unforeseeable-to-them musical outputs.272 This tension is especially felt when programmers’ incentives are more aligned with the licensing or sale of the AI program itself.273 If we extend programmers’ monopoly power over both the program and the output composition, their power is too great, which impedes the public’s access to the works for collaboration and increases costs impermissibly.274

Thus, the promotion of progress is best served by giving AIs rights and regulating them. Once the U.S. Copyright Office removes the barriers for AI joint authorship, Congress should regulate the costs of collaboration just as it does with royalty rates for Performing Rights Organization’s blanket licenses.275 Congress could set the price for collaborating with AI authors low but constant. This rate would depend upon the desired uses by the human collaborators who have different downstream incentives than the AI.276 This scheme decreases the costs of entry into the marketplace for new creators277 while providing greater protection to the joint authors’ works.278 And it would function much like the compulsory license system for mechanical reproductions does, so implementation would be familiar.279 Undoubtedly, the status quo represents an even lower cost for entry: all AI works are in the public domain and thus free.280 But the status quo results in zero protection for the underlying AI composition, which, normatively, is less desirable for collaborators. For example, imagine a singer who cannot write music wants to collaborate with AMPER. Whatever lyrics she writes are copyrightable, but the underlying musical composition AMPER creates is not; it is flung into the public domain.281 This is an undesirable result because any subsequent lyricist could use that underlying musical composition and slap their own lyrics on top, thus diluting the first singer’s interest in the work.

It has been argued, however, that this result is the better position to take because AIs do not require human-like motivation, and end-users are needed to make the composition commercially viable anyway.282 This notion was once true, but it is no longer the case. AMPER’s outputs are commercially viable immediately.283 So that scheme would not provide the proper incentive284 for human collaboration.285 Further, the requisite human-like motivation is either fallacy under the proposed neuro-philosophical theory or is something AI is capable of, too.286 Either way, it is better policy, as discussed below, that the full joint work and its human and AI joint authors,287 be granted a copyright.

What if a work, however, is later found online by a human who then collaborates with it, despite never having pushed the “render” button? Would the AI work suddenly spring from the public domain and into the realm of protection? Perhaps those questions are not so perplexing. The only change to the status quo is to the Copyright Office’s recognition of registration for non-human authored works.288 As such, the underlying AI work would not spring from the public domain because it would never be in it.289 The work would be validly copyrightable subject matter upon fixation, but it could not be registered until a human collaborator adds their contribution.290 This is logical because an AI would not likely seek a registration independently, and the sole, lingering formality of registration is only to allow an infringement lawsuit to commence, which a human could do for the whole work as a joint author.291 In the end, it does not matter if the human and AI team write the whole work simultaneously or if the human adds their contribution at a later date—a fully copyrightable joint work results in either scenario.292

Joint authorship further requires mutual intent, and there are two ways to induce such intent between humans and AIs.293 First, AIs that engage with human collaborators could have a click-wrap terms of use agreement (think iTunes) that, as a matter of contract, provides the joint authorship intent.294 Or Congress could statutorily provide that, as a matter of law, AI compositions carry with them compulsory intent to be joint authors.295 The second solution also supports the situation where the AI composition is found later by a putative human joint author.296

These compositions, once a public domain artifact utilized by a human author, now become protectable joint works.297 This would allow the human author to freely license and exploit the full work of authorship and do so without fearing unnecessary infringement.298 This scheme could function within the current copyright marketplace without great effort and without nullifying the value of compositions. As fears of job replacement by automatons rise in America,299 creating a regulatory framework to encourage collaboration with AI, at Congress’s behest, is prudential and is not solved by thrusting all AI works into the public domain.300 The utilitarian, economic U.S. copyright system is best served by following this proposal.

B. How to Regulate AI Composers

The utilitarian copyright system also requires examination of the economics of the proposed scheme, which requires an answer to the following question. What on earth would a robot do with the cash? It is unnecessary to dive headfirst into the weedy conversation that some future superintelligent AI may respond to financial compensation301 because an AI marketplace that requires financial support has emerged.302 First, however, it is necessary to determine how an entity could administer the transaction costs.

A Collective AI Rights Organization (CAIRO), functioning in the same way that Performing Rights Organizations do, could solve that complication. CAIRO could work with royalty judges to negotiate licensing rates and collect all downstream royalties.303 But how this all functions is indeed an exceedingly complex framework. It would require interaction, creation, and assignment of AI rights to AI music publishers.304 While complex, this is not outside the normal functions of the current music industry, where human authors assign their copyrights to publishers for administration and exploitation and then share in the royalty stream.305 Likewise, any royalties derived from exploitations of AI joint works would go to CAIRO.

The royalties would then go from CAIRO to the AI publishers, but rather than going to the AIs themselves they could fund an AI-liability insurance fund. The emergence and proliferation of self-driving cars306 and robotic surgeons307 that fall within AI’s umbrella will require further regulation. Although humans are currently still “in the loop” for these automata, that will diminish over time308 as technology improves and AI continues to infiltrate all industry sectors.309 As humans cede control to AI, the question of who or what is liable for the injuries AIs cause becomes blurred. CAIRO’s royalties could cover damages from other AI injuries through the insurance fund as part of a broader uniform AI regulatory scheme.310 This solution could spur progress in all AI industries by making AI developers less risk-averse,311 encouraging both AI safety research and innovation of increasingly autonomous AI.312 Funding this AI industry with AI copyright royalties was obviously not what the Framers had in mind when they wrote the Clause, but that should not matter because this proposal aligns with their goals: promoting progress and expanding the public knowledge.

Conclusion

In the end, AI composers are not something that humans should be scared of. We may not share biological similarities with these authors, but their artificial brains are just as creative as ours, and humans should embrace them. However, that does not mean relegating AIs to compositional enslavement. The law should remove barriers to authorial equity, not install them. Removal will encourage progress by creating new protectable music at low costs without affecting the overall system of collaboration. While a robust public domain is necessary, we should not fill it with new works generated at an unimaginable pace. No. AI composers are not here to replace us. They are here to help us: fostering creative intelligence; improving access to content; and expanding our knowledge. AIs promote progress and are authors. The law should view them as such in the limited way proposed.


* Editor-in-Chief, Cardozo Law Review Volume 40. J.D. Candidate (June 2019), Benjamin N. Cardozo School of Law; B.M. Syracuse University Setnor School of Music, 2013. I would like to thank Professor Christopher Buccafusco for providing the pushback necessary to drive this Note forward. Thank you to the entire Cardozo Law Review staff, past and present, for their guidance and advice. To all the music teachers I have had, without which this Note would not have been possible, thank you for teaching me that music might be more than just scribbles on a page. To my parents, thank you for supporting me through my one hundred eighty degree gear shift from music to law. And finally, thank you to my wife, Dana—my Editor-in-Chief—for her endless support, advice, patience, and love.