DATE DOWNLOADED: Sat Jul 8 01:09:58 2023

SOURCE: Content Downloaded from HeinOnline

Citations:
Please note: citations are provided as a general guideline. Users should consult their preferred
citation format's style manual for proper citation formatting.

Bluebook 21st ed.

Gia Jung, Do Androids Dream of Copyright?: Examining AI Copyright Ownership, 35
BERKELEY TECH. L.J. 1151 (2020).

ALWD 7th ed.

Gia Jung, Do Androids Dream of Copyright?: Examining AI Copyright Ownership, 35
Berkeley Tech. L.J. 1151 (2020).

APA 7th ed.

Jung, G. (2020). Do androids dream of copyright?: examining ai copyright ownership.
Berkeley Technology Law Journal, 35(4), 1151-1178.

Chicago 17th ed.

Gia Jung, "Do Androids Dream of Copyright?: Examining AI Copyright Ownership,"
Berkeley Technology Law Journal 35, no. 4 (2020): 1151-1178

McGill Guide 9th ed.

Gia Jung, "Do Androids Dream of Copyright?: Examining AI Copyright Ownership" (2020)
35:4 Berkeley Tech LJ 1151.

AGLC 4th ed.

Gia Jung, 'Do Androids Dream of Copyright?: Examining AI Copyright Ownership' (2020)
35(4) Berkeley Technology Law Journal 1151

MLA 9th ed.

Jung, Gia. "Do Androids Dream of Copyright?: Examining AI Copyright Ownership."
Berkeley Technology Law Journal, vol. 35, no. 4, 2020, pp. 1151-1178. HeinOnline.

OSCOLA 4th ed.

Gia Jung, 'Do Androids Dream of Copyright?: Examining AI Copyright Ownership' (2020)
35 Berkeley Tech LJ 1151 Please note: citations are provided as a
general guideline. Users should consult their preferred citation format's style
manual for proper citation formatting.

-- Your use of this HeinOnline PDF indicates your acceptance of HeinOnline's Terms and
Conditions of the license agreement available at
https://heinonline-org-rcldigitallibrary.knimbus.com/HOL/License
-- The search text of this PDF is generated from uncorrected OCR text.
-- To obtain permission to use this article beyond the scope of your license, please use:
Copyright Information
 DOANDROIDS DREAM OF COPYRIGHT?:
EXAMINING Al COPYRIGHT OWNERSHIP
GiaJungt

I. INTRODUCTION
If Shakespeare were an android, would we care? Would it make his work
less worthy of praise, or less important to society? If Shakespeare, reincarnated
as an android by a clever programmer, published a new play today, would we
deny that play copyright? As outlandish as this hypothetical may seem, these
are immediate questions about the state of copyright for works generated by
artistic artificial intelligence (AI). The U.S. Patent and Trademark Office
(USPTO) is aware, but uncertain, of the role and copyrightability of Al works.
Already, programmers and companies have been registering for copyright on
works produced by generative All despite the recent imposition of a "human
authorship requirement" in the Compendium of Copyright Practices.2 Because
the Copyright Office, which administers the registration and recording of
copyright, only needs someone to claim that a work is theirs, companies and
programmers have been taking advantage of the lax investigation into claims
of ownership.3 But Naruto v. Slater" and the newest issue of the Compendium
show that if those copyrights were ever challenged, they would likely be
invalidated. Scholarly arguments as to the best formal regimes and a request by
the USPTO for comments6 characterize, but not clarify, this uncertainty. As it
stands now, there is a disparity between what companies are doing in practice
and what the Copyright Office allows on paper.

DOI: https://doi.org/10.15779/Z383J3922D
©2020 Gia Jung.
t J.D., University of California, Berkeley, School of Law, Class of 2021.
1. Dani Deahl, The USPTO Wants to Know if Artifidal Intelligence Can Own the Content it
Creates, THE VERGE (Nov. 13, 2019, 1:45 PM), https://www.theverge.com/2019/11/13/
20961788/us-government-ai-copyright-patent-trademark-office-notice-artificial-intelligence.
2. U.S. COPYRIGHT OFFICE, COMPENDIUM OF U.S. COPYRIGHT OFFICE PRACTICES
313.2 (3d ed. 2014).
3. See, e.g., infra note 35.
4. Naruto v. Slater, No. 15-cv-04324-WHO, 2016 U.S. Dist. LEXIS 11041, 2016 WL
362231 (N.D. Cal. Jan. 28, 2016), affd, 888 F.3d 418 (9th Cir. 2018).
5. See, e.g., Kalin Hristov, Artifidal Intelligence and the Copyright Dilemma, 57 IDEA 431
(2017).
6. See Request for Comments on Intellectual Property Protection for Artificial
Intelligence Innovation, 84 Fed. Reg. 58,141 (Oct. 30, 2019).
1152 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

This Note looks particularly at how artistic works produced by Al should

be protected and incentivized. As AI-as-an-industry grows and becomes more
advanced, so do its issues surrounding copyright. Rather than banning AI-
produced works to keep the existing regime unchanged, it is time to reexamine
why copyright hinges on human authorship, rather than creativity. Copyright
presents three distinct hurdles for a work to be copyrightable. 7 The first is
originality, which mandates that a work must be independently created or not
copied.8 The next is creativity, which is used not to judge the aesthetic merit
of the work, but to filter against function. This standard ensures that copyright
protects the substance of a work, not its general idea or theme. Finally, and
most important to this Note, is authorship, which is used to establish the
creator and rights-holder of a copyrighted work.10 This hurdle is highest for Al
generated works because traditionally, originality and authorship were so tied
together that authorship implied creativity, and vice versa.11 AI-generated art,
as authorless creative works, breaks the standard "if it looks creative, it must
have a human author" way of understanding authorship and confronts how
non-human authorship has been and can be accommodated.12 This Note
clarifies how and why AI-generated creative works meet these hurdles, and
how incentive-based rationales for copyright protections apply.
This Note seeks to elucidate the policy arguments underlying the copyright
system that support or oppose the adoption of Al authorship of artistic works.
These will be used to explain institutional changes that can best serve the
producers, users, and the future of generative Al. To that end, this Note begins
with a brief description of generative artificial intelligence, arguing that the way
AIs are designed makes output inherently creative, rather than operating as a
tool of the programmer. Next, Part II explores some of the case precedent and
history on machine and non-human copyright. From there, Parts III and IV
proceed through the elements of originality and authorship, distinguishing
copyrightability from ownership and exploring the implications of applying
copyright to artificial intelligence.

7. U.S. COPYRIGHT OFFICE, supra note 2, 302.

8. Id. 308.1.
9. Id 308.2; see also Dennis S. Karjalam, Copyrightand Creativity, 15 UCLA ENT. L. REV.
169, 201 (2008) (arguing that creativity is a necessary, not sufficient condition for copyright
such that functionality funnels works either into patent or copyright).
10. U.S. COPYRIGHT OFFICE, supra note 2, 302, 404, 405.
11. See infra Part II (showing how legal jurisprudence that affects this area became
entangled).
12. See Artifidal Intelligence: The Ins and Outs of Copyzght and Al, U.S. PATENT
&

TRADEMARK OFFICE (Jan. 31, 2019), https://rev-vbrick.uspto.gov/#/videos/d6e591c3-

64cf-4d74-ab35-9f387a2da4b2.
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1153

Part V proposes a model of copyright ownership for artistic works by

artificial intelligence. A brief overview of current scholarship shows that
popular proposals are incomplete in their considerations of liability, incentive,
and equitable return on investment. This Note advocates for a model that
draws on the works made for hire doctrine and current U.K. law addressing
computer-generated copyright. Al works would have specific provisions for
term limits and authorship to achieve an equitable balance between effort and
reward. Al would be treated as the author-in-fact and the programmer as the
author-in-law to clearly indicate attribution of risks and benefits. This Note
concludes with a call for a clear model that weighs current issues and upcoming
technological advances, so that the future of copyright may ensure appropriate
accountability and administrability for creative works by artistic Al.
A. WHAT IS ARTIFICIAL INTELLIGENCE?

Confusion about what Al is extends beyond the general public into the
field itself. Scholars' and practitioners' definitions vary from broad to narrow
and analogize from aspects of human functions that are also hard to define
such as the ability to learn, consciousness, and self-awareness.13 What is most
generally accepted and has made its way into the Oxford English Dictionary is
that AIs are "computer systems able to perform tasks normally requiring
human intelligence, such as visual perception, speech recognition, decision-
making, and translation between languages."" In traditional forms of simple
Al, programmers write many specific instructions for the program to follow
and hope that their instructions are the best way to achieve their end goal.15
Current modes of generative Al are more complex and independent. 16 This
Note focuses on deep learning and generative adversarial neural networks as
the two dominant modes of generative Al.
Deep learning and neural networks seek to create programs that behave as
though they are interconnected brain cells. Programmers create multiple layers
of processing "neurons" to allow the Al to learn how to recognize a solution
on its own by looking for signature patterns of appearance, behaviors, or
features at varying scales. 17 Each level of neurons provide insight and passes

13. Matthew U. Scherer, RegulatingArtifidalnel/4jgentSystems:Risks, Challenges, Competendes,

and Strategies, 29 HARV. J.L. & TECH. 353, 360 (2016).
14. ArtifidalIntellgence, OXFORD ENGLISH DICTIONARY,
https://en.oxforddictionaries.com/definition/artificial intelligence (last visited Feb. 9, 2020).
15. Chris Rodley, Deep Dinosaur, CHRIS RODLEY (June 19, 2017), https://
chrisrodley.com/2017/06/19/dinosaur-flowers/.
16. Id
17. Id.
1154 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

that information to the next, more senior level. 18 After learning to recognize
something based off of labeled data or supervised self-learning, deep learning
programs can then apply that knowledge to recognize raw information and
produce outcomes based on those understandings.19
Generative adversarial neural networks (GANs) set two neural networks
against each other to improve the quality of the results. One generates
imitations of the target material, and one evaluates whether or not the
productions are forgeries. Learning in tandem, as the latter gets better at
detecting forgeries, the former must become better at creating them.20 In
contrast to deep learning, GANs do not rely on a large amount of training data,
as the interaction between the two networks itself creates a large amount of
data from a limited source that continues to be cross-checked against each
other."
1. Artistic and Creative GenerativeAl
By their very design, deep learning and GANs (hereinafter generative AI)
seek to imitate the process of human learning and creativity. Consequently,
they produce creative works that are outside the control of the original
programmer. Some scholars, including Professor Jane Ginsburg, dismiss the
concept of Al authorship outright, claiming that Al is a mere tool, and its
results are absolutely dependent on the inputs of the programmer.22 If this
were the case, the extensive scholarship, debate, and calls for input by the
USPTO would be moot; copyright already protects machine created works
where there is creative input or intervention from a human author.23 The issue
at hand arises because the machines described here are generating works with
an increasing level of independence from human intervention. And, as others
have noted, the incorporation of randomness into a generative Al means it is
only constrained to the same parameters as a human author-grammar,

18. See Simon L6bfwander, About Artifidal Intelligence, Neural Networks, & Deep Learning
AYiMA (Jan. 24, 2017), https://www.ayima.com/blog/artificial-intelligence-neural-networks-
deep-learning.html.
19. Id
20. Bernard Marr, Arifidal Intelligence Explained: What are GenerativeAdversarial Networks
(GANs)?, FORBES (June 12, 2019, 12:23 AM), https://www.forbes.com/sites/bernardmarr/
2019/06/12/ artificial-intelligence-explained-what-are-generative-adversarial-networks-gans
/

#32aad65a7e00.
21. Id
22. Jane C. Ginsburg & Luke Ali Budiardjo, Authors and Machines, 34 BERKELEY TECH.
L.J. 343, 396 (2019).
23. U.S. COPYRIGHT OFFICE, sufpra note 2, 313.2.
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1155

cohesion, and genre-specific devices.24 To that end, it could be said that in this
context, the programmer has as much claim to an AI's work as Shakespeare
does to any iteration of the star-crossed lovers narrative.25
This Note specifically addresses artistic generative Al. These types of Al
will consistently be referred to as "creative" or "artistic" as a way of reflecting
the independent process of recognition, analysis, and generation that the Al
undergoes to produce a unique work. The three below examples illustrate the
generative Al referenced throughout this Note. The first, Chris Rodley's
"Deep Dinosaur" (Figure 1) produces novel and striking images by combining
two sets of inputs-here, vintage flowers and dinosaurs. 26 Rodley and his Deep
Dinosaurs are not the product of a large project like Google's Deepmind, but
an independent creative project undertaken by a PhD candidate at the
University of Sydney. 27 Using a technique known as style transfer or deep style,
the Al learns to recognize characteristics of images, then reproduce them,
synthesizing the two characteristics to produce a recognizable amalgam. 28

24. See Annemarie Bridy, Coding Creativity: Copyright and the Arifidally Intelligent Author,
2012 STAN. TECH. L. REv. 5, 12 (2012).
25. By this, I mean works like High SchoolMusical(instead of feuding families, high-school
cliques) and West Side Story (instead of feuding families, rival New York gangs) wherein the
conceit is the same, but a court would likely not consider them derivative works of Romeo and
Juliet. See also Nichols v. Universal Pictures Corporation, 45 F.2d 119 (2d Cir. 1930).
26. Rodley, supra note 15.
27. Chris Rodley, Me, CHRIS RODLEY, https://chrisrodley.com/me/ (last visited Feb. 9,
2020).
28. Rodley, supra note 15.
1156 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

Figure 1: Deep Dinosaur

The second is "Deep-speare," a deep learning Al that writes sonnets

(Figure 2)." Created by researchers for the 56th Annual Meeting of the
Association for Computational Linguistics, Deep-speare was trained on the
rules, structure, and elements of around 2,600 real sonnets. 30 From this data,
Deep-speare was able to generate sonnet quatrains "nearly indistinguishable
from published poems by humans" when read by the average person.3
'

29. Luke Dormehl, Move Over, Shakespeare: This Sonnet-Writing A.I. is the Poet We Need,
DIGITAL TRENDS (July 30, 2018, 12:32 PM), https://www.digitaltrends.com/cool-tech/ai-
generates-shakespearean-sonnets/.
30. Jey Han Lau, Trevor Cohn, Timothy Baldwin, Julian Brooke, and Adam Hammond,
Deep-speare:A Joint NeuralModel of Poetic Language, Meter and Rhyme, 56 PROC. ANN. MEETING
Ass'N FOR COMPUTATIONAL LINGUISTICS 1948 (2018).
31. Id. (noting that a literature expert was not fooled-the stress, rhyme, and meter were
all human-passable but the Al underperformed on readability and emotion).
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1157

Figure 2: Deep-speare Example Quatrain

"With joyous gambols gay and still array,

no longer when he 'twas, while in his day
at first to pass in all delightful ways
around him, charming, and of all his days."

The third is Endel, an app-based company that uses a generative algorithm

to create "[p]ersonalized sound environments" for focus, relaxation, and
sleep.32 In 2018, Endel partnered with Warner Music Group to produce and
distribute these soundscapes on Apple Music and iTunes. 33 To monetize the
songs, Warner needed to register them for copyright, and requested the
songwriting credits for each. But besides creating the program itself, Endel as
a company did nothing to generate the audio-the Al reacts to inputs like
"rain" and creates the song entirely on its own. 34 Ultimately, the company
"decided to list all six employees at Endel as the songwriters for all 600
tracks."35
The above examples do not begin cover the broad and fascinating world
of generative Al. Rather, they serve as points of reference for thinking through
specific aspects of Al copyright dilemmas. But before investigating how
potential solutions for Al copyright ownership could apply to works like these,
it is important to first understand how and why these issues arise out of the
existing copyright regime.

II. LEGAL JURISPRUDENCE ON NON-HUMAN

COPYRIGHT OWNERSHIP
Case precedent displays a grudging willingness to challenge traditional
conceptions of copyrightability in order to accommodate mechanical and non-
human copyright. The U.S. Constitution Article I, Section 8, authorizes

32. ENDEL, https://endel.io/ (last visited Feb. 10, 2020).

33. Endel to Release 20 Algorithm-PoweredAlbums to Help You Sleep, Focus & Relax, ENDEL
(Jan. 22, 2018), https://endel.io/presskit/Endel-PressRelease-20MusicAlbums.pdf
[hereinafter Endel.io].
34. See id.; Sdence, ENDEL, https://endel.io/science/ (last visited Feb. 10, 2020).
35. Dani Deahl, We've Been WarnedAbout AI And Music For Over 50 Years, But No One's
Prepared, THE VERGE (Apr. 17, 2019, 10:30 AM),
https://www.theverge.com/2019/4/17/ 18299563/ai-algorithm-music-law-copyright-
human.
1158 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

Congress "[t]o promote the [p]rogress of [s]cience and useful [a]rts, by securing
for limited [t]imes to [a]uthors and [i]nventors the exclusive [r]ight to their
respective [w]ritings and [d]iscoveries." 36 This order, written intentionally
broadly, 37 does not explicitly preclude non-human authors. However, the
concept of "romantic authorship," in which creativity can only stem from "an
individual creative personality, a solitary originator of stylistically consistent
works," has long been the guiding concept for understanding how authorship
and creativity interact.38 As a result, disputes in the copyright field have
developed over the scope of what constitutes "authors" and "writings" when
it is not clear that the author was entirely human.
A. PHOTOGRAPHY AND TORAH SOFT-EXTENDING COPYRIGHT FROM
MACHINE TO MAN

One of the first cases considering how machine-produced works could be

copyrightable, Burrow-GileslithographicCo. v. Sarony,39 established that works are
copyrightable so long as they are "representatives of original intellectual
conceptions of the author."40 In 1865, President Abraham Lincoln authorized
a bill amending the Copyright Act to extend protection to photographs and
photographic negatives, likely due in part to their "prominent role in bringing
the horrors of the Civil War to the public."" Nearly twenty years later, this
case tested the constitutionality of that 1865 legislation, a seemingly logical
progression from maps, to sketchings and engravings, and now photography.
In considering whether photographs, which were and are purely mechanical
reproductions of their subjects, met the requisites of originality and creativity,
the Court understood the concept of authorship could enable the product of
a machine to be copyrightable.4 2 In concluding the case in favor of the
photographer, Justice Miller commented that "photography is to be treated for
the purposes of the act as an art," such that the author uses the camera to

36. U.S. CONST. art. I, 8, ci. 8.

37. ArtI.S8.C.1.1 Origins and Scope of the Power, CONGRESS.GOV, https://
constitution.congress.gov/browse/essay/I_8_8_1_1/ (last visited Feb. 11, 2020) ("Congress
has broad leeway to determine how best to promote creativity and utility through temporary
monopolies.").
38. Bridy, supra note 24, at 4. See also ZACHARY LEADER, REVISION AND ROMANTIC
AUTHORSHIP, 11 (1999) ("The Romantic author is often portrayed as spontaneous,
extemporizing, otherworldly, and alone ... Romantic attitudes to authorship profess a
preference for what comes naturally, with a concomitant devaluing of secondary processes,
including second thoughts.").
39. 111 U.S. 53 (1884).
40. Id. at 58.
41. WILLIAM F. PATRY, COPYRIGHT LAW AND PRACTICE 244 (1996).
42. Burrow-GilesLUthographic, 111 U.S. at 58.
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1159

implement a creative idea.43 To grant copyright, the Court extended the

creative thought of the author through the camera, a logical inference from the
control wielded by the photographer. 44
Notably, the Court in Burrow-Giles declined to establish whether unstaged
photographs that had no active human author or human intervention in their
composition could engender copyright. 45 However, the dicta implied that an
unstaged photograph, since it did not include creative human participation,
could not validly hold copyright. 46 Though not used explicitly in other cases,
Professor Annemarie Bridy notes that the reasoning behind this dicta
continued to perpetuate an unnecessary "dichotomy between creative and
mechanical labor" in which automation is in opposition to creative
authorship. 47 Because the case's reasoning rested largely on Sarony's
involvement in setting up the photograph, later cases concluded that where
mechanical creations arose, a human artist had to be substantially involved. 48
Following that dichotomy, the National Commission on New
Technological Uses of Copyrighted Works ("Commission")'s 1974 report
expressed strong skepticism regarding precedent linking mechanical works to
authorship.4 9 Specifically directed to study the emergence of new, computer
assisted works, the Commission declined to imagine the future implications of
existing generative word processors. 50 Instead, they decided computers could
only be "an inert instrument," which, like the camera, functions solely as a
creative tool that must be activated and directed by a human. 51 The
Commission thus advised Congress that no change should be made to
accommodate automatic systems.5 2 This recommendation was challenged only
ten years later by the Office of Technology Assessment (OTA). 53 The authors
of the 1986 OTA report recognized that programs autonomously producing

43. Id. at 61.

44. Id. at 60 (deciding that the photograph is art on the basis that the photograph was
given "visible form" by Sarony's "desired expression"').
45. Id. at 59 (on the question of whether the ordinary production of a photograph
without the photographer's intervention, "we decide nothing").
46. Bridy, supra note 24, at 5-6.
47. Id. at 6.
48. See id.
49. NAT'L COMM'N ON NEW TECH. USES OF COPYRIGHTED WORKS, FINAL REPORT
OF THE NATIONAL COMMISSION ON NEW TECHNOLOGICAL USES OF COPYRIGHTED WORKS
4, 108-15 (1978).
50. Id.
51. Id. at 109.
52. Id.
53. U.S. OFFICE OF TECH. ASSESSMENT, INTELLECTUAL PROPERTY RIGHTS IN AN AGE
OF ELECTRONICS AND INFORMATION (1986), 65-73.
1160 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

text or musical works were a far cry from the word processing capabilities
implied by the Commission. 54 In particular, the OTA report referenced the
"blurring of the distinction between the copyrighted work and its product." 55
The report concluded by suggesting that interactive computer programs, if not
considered co-authors of the output produced, at least raised several troubling
questions of copyright ownership and creativity requirements. 56
U.S. case law has not yet dealt with the copyrightability of an artistic work
produced by Al. 57 However, a Southern District of New York case, Torah Soft
Ltd. v. Drosnin,51 comes within the realm of dealing with the output of an
algorithm. Torah Soft, makers of a biblical-code-finding algorithm, sued over
infringement of printouts of output generated by their computer program. 59
According to Bible code researchers, the Hebrew Bible is purported to be
embedded with a code that is revealed by finding words and phrases which
appear in the Bible at equidistant letter skips. 60 This software analyzed and
found these code words and phrases. 61 Interestingly, this court did not concern
itself with authorship. The court suggested in passing that copyright protection
afforded to the computer program may also extend to the output files. If the
program does the "lion's share of the work" in creating the output files and
the end-user's input is "marginal," then it follows that the protection extends. 62
Regardless, the court focused on whether the outputs contained protectable
elements. 6 Because the software was made to comply with religious rules that
were functional in nature, the court determined that the outputs were also
functional and therefore not protectable. 64

54. Id.
55. Id.
56. Id.
57. This statement is accurate as of December 2019. But see BETA WRITER, LITHIUM-
ION BATTERIES: A MACHINE-GENERATED SUMMARY OF CURRENT RESEARCH (2019) (a
machine-generated book that claims copyright but is not registered in the U.S. Copyright
Office's Public Catalog); Leo Kelion, Al system 'shouldbe recognisedas inventor,' BBC NEWS (Aug.
1, 2019), https://www.bbc.com/news/technology-49191645 ("[wo professors from the
University of Surrey have teamed up with the Missouri-based inventor of Dabus Al to file
patents in the system's name with the relevant authorities in the UK, Europe and US.").
58. 136 F. Supp. 2d 276 (S.D.N.Y 2001).
59. Id. at 280.
60. Id.
61. Id.
62. Id. at 283.
63. Id. at 283-84.
64. Id. at 287.
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1161

B. PSYCHOGRAPHY-GRANTING GOD COPYRIGHT

Another interesting line of cases that involve non-human works are those
dealing with psychography, or automatic writing. Psychography is the
production of writing or drawing supposedly by a spiritual agent, in which the
human is merely a scribe. 65 Although parties seeking copyright protection in
these cases claim that authorship actually lies with non-human, usually celestial
or spiritual, beings, courts have found a sufficient nexus to human creativity
to sustain copyright.66
In each of these cases, courts apply Feist Publications, Inc., v. Rural Telephone
Service Co. 67 to determine whether a compilation possesses the requisite
originality for copyright protection. In Feist, the Court established that
copyright demanded a minimal degree of creativity, which did not include an
alphabetically arranged phonebook. 68 Nonetheless, compilations of facts may
be copyrightable if arranged creatively, i.e., beyond a merely functional
arrangement.6' Referencing Burrow-Giles, the Court held that an author who
claims infringement must prove the existence of "intellectual production, of
thought, and conception." 70 This showing can be demonstrated through active
and intentional choices, where changes were not accidental or externally
motivated. 71 Feist thus sets a low bar for determining whether a work is creative
or merely functional.
Along these lines, the Ninth Circuit in UrantiaFound. v. Maaherra72 upheld
73
a copyright granted on behalf of a "non-human" author. There, the claimed
work was dictated by a deity listed as "the Divine Counselor, the Chief of the
Corps of Superuniverse Personalities, and the Chief of the Archangels of
Nebadon." 74 It was transcribed by the human whose name ended up on the
copyright. 75 Pushing past the non-human element, the court found that the
originality requirement necessary for a valid copyright was satisfied because
the human beings who "compiled, selected, coordinated, and arranged" the

65. Psychography, OXFORD ENGLISH DICTIONARY,

https://en.oxforddictionaries.com/definition/psychography (last visited Feb. 11, 2020).
66. Bridy, supra note 24, at 20.
67. 499 U.S. 340 (1991).
68. Id. at 362.
69. Id.
70. Id.
71. Bridy, supra note 24, at 8.
72. 114 F.3d 955 (9th Cir. 1997).
73. Id. at 964.
74. Id. at 957.
75. Id.
1162 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

book did so "in such a way that the resulting work as a whole constitutes an
original work of authorship." 76
A similar case, Penguin Books U.S.A., Inc. v. New Christian Church of Full
Endeavor, Ltd77 held that, "[a]s a matter of law, dictation from a non-human
source should not be a bar to copyright." 78 There, a "Voice" the transcriber
identified as Jesus dictated a manuscript and gave legal advice.79 According to
the author, Jesus advised that the copyright page should not have the author's
name in case people confused the copyright author with the actual author
(Jesus and the Holy Spirit). 80 In a compromise, the copyright registration listed
the work's author as "[Anonymous] (Helen Schucman)." 81 The court held that,
irrespective of independent editorial judgment in the process of recording the
work, copyright could subsist in a non-human work so long as there is a
sufficient human nexus. Helen's acknowledgement of her involvement was
sufficient.82

C. NARUTO V. SLATER-THE MONKEY SELFIE THAT DROVE THE

COPYRIGHT OFFICE BANANAS

As made clear in Naruto v. Slater, however, this liberal view of non-human

authorship does not apply to animals. 83 Now widely known as the "Monkey
Selfie Case," 84 Naruto v. Slater dealt with whether Naruto, a crested macaque,
had standing to state a claim under the Copyright Act over photographs of his
own "independent, autonomous action." 85 The dispute began when Slater
the wildlife photographer who set the cameras up and claimed ownership of
the selfie-asked that the photos be taken off of Wikipedia. Wikipedia, in
return, argued that the photo was uncopyrightable because it was taken by an

76. Id. at 958.

77. Penguin Books U.S.A., Inc. v. New Christian Church of Full Endeavor, Ltd., 96 Civ.
4126 (RWS), 2000 U.S. Dist. LEXIS 10394 (S.D.N.Y. July 21, 2000).
78. Id. at *36.
79. Id. at *7-14 ("Schucman heard from the Voice that copyright registration should be
sought for the Course.").
80. Id. at *16.
81. Id. at *16-17.
82. Id. at *19, *32-33, *63-67.
83. Naruto v. Slater, No. 15-cv-04324-WHO, 2016 U.S. Dist. LEXIS 11041, 2016 WL
362231 (N.D. Cal. Jan. 28, 2016), aqF'd, 888 F.3d 418 (9th Cir. 2018).
84. See, e.g., Joshua Berlinger, PETA, PhotographerReach Settlement In Monkey Selfie' Case,
CNN (Sept. 12, 2017, 6:19 AM), https://www.cnn.com/2017/09/12/asia/monkey-selfie-
settlement/index.html; Andres Guadamuz, Can The Monkey Selfie Case Teach Us Anything About
Copyright Law?, WIPO MAG. (Feb. 2018), https://www.wipo.int/wipomagazine/en/2018/
01/article_0007.html.
85. Naruto, 2016 U.S. Dist. LEXIS 11041, at *3.
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1163

animal." Seeing an opportunity to potentially strike a blow for animal rights,

the People for the Ethical Treatment of Animals (PETA) filed a suit on behalf
of Naruto as his "next friends."87 PETA alleged that Slater and the publishers
of books containing the selfies violated Naruto's copyright. 88
As this case was ongoing, the U.S. Copyright Office released an update to
the Compendium Of U.S. Copyright Office Practices, which issued "The
Human Authorship Requirement." 89 Relying on citations from the Trade-Mark
Cases? and Burrow-Giles," the Copyright Office concluded that it will only
register an original work of authorship provided that the work was "created by
a human being."92 The requirement explicitly notes that a "photograph taken
by a monkey" would not be considered an original work of authorship.93 With
this on the books, the court acknowledged and the Ninth Circuit affirmed that
the Copyright Act does not confer standing upon animals like Naruto and
dismissed the case.94
Naruto's holding that non-human entities cannot be considered "authors
in law" leaves unanswered questions that have broad implications for
ownership of non-human art. Slater's active arrangement of the camera would
seem to meet creativity standards under Feistand Burrow-Giles. However, unlike
the psychography cases, the court did not address whether Slater would be able
to enforce his own copyright on the photos. 95 Instead, Slater and PETA
reached a settlement allowing Slater to continue to use the photos so long as
he donates twenty-five percent of the revenue earned from them to a charity
for crested macaques. 96 However, the ruling and the newest edition of the
Compendium seem to suggest that the photos belong in the public domain. If
that is the case, to what extent must a work be "created by a human being"? 97
How can those with financial interests in the output of non-human entities
protect their investment? What sort of regime that protects non-human

86. Id. at *1-2.

87. Id.
88. Id at *2.
89. U.S. COPYRIGHT OFFICE, supra note 2, 306.
90. 100 U.S. 82 (1879).
91. 111 U.S. 53 (1884).
92. U.S. COPYRIGHT OFFICE, supra note 2, 306.
93. Id. 313.2.
94. See Naruto v. Slater, No. 15-cv-04324-WHO, 2016 U.S. Dist. LEXIS 11041, at *2,
2016 WL 362231, at *1 (N.D. Cal. Jan. 28, 2016), affid, 888 F.3d 418 (9th Cir. 2018).
95. See generally Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018); see supra Section I.A.
96. Paulina Julia Perkal, Monkey businessfinally settled: the 'monkey sefie' disputes,
KLUWER COPYRIGHT BLOG (Feb. 5,2018), http://copyrightblog.kluweriplaw.com/2018/02/
05/monkey-business-finally-settled-monkey-selfie-disputes/.
97. U.S. COPYRIGHT OFFICE, supra note 2, 306.
1164 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

creativity would best suit a tradition that inextricably intertwines the concepts
of originality with authorship? This Note grapples with these questions to
preclude a similar in-court outcome for Al generated works.

III. ORIGINALITY
The U.S. Compendium of Copyright Practices, citing Feist, states,
"originality is 'the bedrock principle of copyright' and 'the very premise of
copyright law.' "98 To qualify for copyright protection, a work must be original
to the author, which means that the work must be "independently created by
the author," and it must possess "at least some minimal degree of creativity.""
This next Section shows that works by Al satisfy both elements by the nature
of the process of output and by the underlying policies of copyright law.
A. INDEPENDENTLY CREATED

The threshold for independent creation is very low. Works are

independently created so long as they are not literally copied. 100 A work of
authorship may be original, even though it is not novel, inventive, or if "it
closely resembles other works." 101 This low threshold allows a full gamut of
protectable works while also setting parameters for infringement (direct
copying).10 2 Generative AI's purpose is to create novel outputs based on
prompts, its learned process of recognition, and elements of randomness.1 3
This process of generation usually means that each generated work can be
considered independently derived. If the source material is not incorporated
but is instead cut and pasted wholesale, the program is defective because it is
not actually "generating" anything. 104

B. MODICUM OF CREATIVITY

Courts tend to distort the "modicum of creativity standard" as hinging on

novelty or romantic abstraction. However, rather than artistic or inspired

98. U.S. COPYRIGHT OFFICE, supra note 2, 308.

99. Id.
100. Id 308.1.
101. Feist Publications, Inc. v. Rural Tel. Service Co., 499 U.S. 340, 345 (1991).
102. Id. ("The vast majority of works make the grade quite easily, [if] they possess some
creative spark, 'no matter how crude, humble or obvious it might be."').
103. See Rodley, supra note 15.
104. See Clark D. Asay, Independent Creation in a World of A, FIU L. REV. (forthcoming
2020).
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1165

merit, the modicum of creativity requirement mainly functions as a bar to

functional works.105
Case law has traditionally considered creativity an individualistic and purely
internal activity. Requisite creativity is characterized as the "fruits of intellectual
labor" that "are founded in the creative powers of the mind" 106 and the
"original intellectual conceptions of the author." 107 This conception is rooted
largely in the idea of the romantic Jeffersonian author who, through sheer
intellect, creates original ideas out of nothing. 108 Works produced by a romantic
author are by nature inherently original and creative, having sprung from a
creative mind. Foucault scholars argue that this mythic figure is so central to
our understanding about creativity that it obscures the collective and
cumulative nature of creativity and misrepresents the "actual processes of
cultural production."109 With this mythos on center stage, it becomes
immeasurably difficult for courts and laypeople to detach the notion of
creativity from popular representations. The image of the creative spark hitting
the mind of the author like lightning such that words (or paint, sculpture, etc.)
come pouring out like the final scene of Shakespeare in Love (1998) is
ubiquitous and lauded.110 In actuality, Shakespeare should be considered the
anti-romantic author, since few of his plots are original. 111 It is the reality of
cumulative creativity, rather than the fantasy of the romantic author, that has
driven art. But if creativity in copyright remains tied to the idea of romantic
authorship, an Al work could not be considered original, since it is inherently
algorithmic.
Another stream of cases presents the concept of originality as a stand-in
for market value. 2 Setting aside artistic merit or romantic authorship, in 1839

105. See Karjalam, supra note 9, at 201; Feist, 499 U.S. at 346 (creating the standard that
creativity is a low bar that works to filter out utilitarian works like an alphabetized phone
book).
106. Trade-Mark Cases, 100 U.S. 82, 94 (1879).
107. Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884).
108. Oren Bracha, The Ideology ofAuthorship Revisited: Authors, Markets, andLiberal Values in
Eary Amencan Copynght, 118 YALE L.J. 186, 188 (2008) (investigating the role of romantic
authorship as a pervasive myth that is cyclically invoked or diminished depending on the
interests at stake).
109. Bridy, supra note 24, at 4.
110. SHAKESPEARE IN LOVE (Universal Pictures 1998) ("A blank page. A hand is writing:
TWELFTH NIGHT. We see Will sitting at his table ... Will looks up from the table. 'And
her name will be... 'Viola.' He looks down at the paper, and writes: 'Viola"').
111. J. M. Pressley, Shakespeare's Source Material, SHAKESPEARE RES. CTR.,
http://www.bardweb.net/content/ac/sources.html (last visited Dec. 13, 2019).
112. See Bracha, supra note 108, at 203 (In Emerson v. Davies, "Justice Story turned to the
market as the sole arbiter of value"). See also Peter Jaszi, Toward a Theory of Copynght: The
1166 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

and 1845 Justice Story decided cases based on a more practical understanding
of the interdependent and cumulative nature of creativity." 3 He relied on "the
market as the only criterion for assessing value."" 4 Justice Story saw that works
imitate and build on each other. A work did not have to add anything new or
culturally significant to be beneficial and protectable. If it was in demand, it
was worthwhile. 115
Justice Holmes was likewise unconcerned with romantic authorship. In
Bleistein a. Donaldson Lithographing,116 Holmes affirmed that courts have no role
in making aesthetic judgements. Using content neutrality and market value, the
Court found that copyright had no threshold requirement of objective
aesthetic value. 117 Holmes focused not on the author or the merit of the
advertisement, but on how the approval of the public eye and aesthetic
elements placed them under statutory protection. 118 Bleistein shows that works
need not be particularly novel or skillful to be worthy of incentivization and
protection.
As illustrated above, courts have competing views of originality. There
remains a split on how and where creativity arises, and how important it is to
copyright. Precedent like Feist, Torah-Soft, and Justice Story's cases suggests AI-
generated works meet the creativity bar. The demand for Al systems like Endel
and Deep Dinosaur demonstrate enough market interest to satisfy Justice
Story's and Justice Holmes' tests for creativity.119 Likewise, the choices Al
systems make when generating content meet Feist and Torah-Soft's standards
for creativity in authorship.120 Unlike an alphabetical phonebook or code-

Metamorphoses ofAuthorsh/p, 1991 DUKE L.J. 455,481-85 (1991) (examining cases in which "the
'[a]uthor' vanishes" in the commercialization of cultural production).
113. Bracha, supra note 108, at 204. See, e.g., Gray v. Russell, 10 F. Cas. 1035 (C.C.D. Mass.
1839) (No. 5,728); Emerson v. Davies, 8 F. Cas. 615 (C.C.D. Mass. 1845) (No. 4,436).
114. Bracha, supra note 108, at 204.
115. Id.
116. 188 U.S. 239 (1903) (holding that commercial lithographs are copyrightable beyond
their mere commercial value).
117. Bracha, supra note 108, at 200.
118. See Jaszi, supranote 112, at 482.
119. See, e.g., Andrew Liszewski, A NeuralNetwork Turned a Book of Flowers into Shockingjy
Lovely DinosaurArt, GIZMODO (June 19, 2017, 11:27 AM), https://gizmodo.com/a-neural-
network-turned-a-book-of-flowers-into-shocking-1796221045 ("The estate of M.C. Escher
may have just lost its lucrative stranglehold on the dorm room poster market thanks to ... a
deep learning algorithm."); Endel.io, supra note 33 ("[Endel] will be introduced to a larger
audience through the extensive reach of the Arts Music division's marketing and distribution
resources . . . with investors including Amazon Alexa Fund, Avex Inc., Major Lazer's
Jillionaire, Plus 8 Equity Partners, Kima Ventures, Impulse Ventures, and world-famous DJ
La Fleur.").
120. See supra Section IIB; see also Karjalam, supra note 9, at 172.
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1167

finder where the end-result is necessarily the same, generative Al incorporates

randomness into the choices given by its parameters so that the outcome is
not purely a function of its form."
At the heart of the matter is whether courts and legislatures will make
policy changes to recognize AI-generated works as creative as they have in the
past. In the psychography cases, courts found that a "non-human source
should not be a bar to copyright."" In signing photographs into copyright law
in the wake of the Civil War, Congress and President Lincoln made a policy
judgement that photography itself, as a medium and a means of providing
societal value, should be protected. If not simply for more floral dinosaurs,
encouraging artistic Al will produce innumerable discoveries and solutions
much as photography has. 2 And, as some courts acknowledge, "all creativity
is inherently algorithmic and that works produced autonomously by computers
are therefore less heterogeneous to both their human counterparts and existing
copyright doctrine."" The basic story of Romeo and Juliet has been done
many times over, and yet, as Nichols v. UniversalPictures Corp. teaches us, building
on generalized ideas and clich6s are nonetheless protectable and encouraged.125
As courts have similarly noted, most of Shakespeare's plots came directly from
existing stories-like Deep-speare, he learned and practiced on them to
produce recognizably creative works. 12 6

121. See Rodley, supra note 15.

122. Penguin Books U.S.A., Inc. v. New Christian Church of Full Endeavor, Ltd., 96 Civ.
4126 (RWS), 2000 U.S. Dist. LEXIS 10394 (S.D.N.Y. July 21, 2000).
123. See Sdence Photography, SCITABLE BY NATURE EDUCATION (2014) https://
www.nature.com/scitable/spotlight/science-photography-10285104/ ("Careful observation
of evidence is the heart of modern scientific method; photography has always been valued as
an objective technique of observation, freed from the potential for human error implicit in the
older method of sketching experimental observations. Just as important, photography can
gather data that can't be detected or processed by the human eye. Using technology that
captured the scattered path of invisible x-ray beams, for example, Rosalind Franklin in 1952
was able to reveal the precise structure of intertwined DNA molecules-what we now
recognize as the double helix."); Demis Hassabis, On AI's Potential, ECONOMIST (Nov. 21
2019), https://worldin.economist.com/article/17385/edition2O2Odemis-hassabis-predicts-
ai-will-supercharge-science ("By deepening our capacity to ask how and why, Al will advance
the frontiers of knowledge and unlock whole new avenues of scientific discovery, improving
the lives of billions of people.").
124. Bridy, supra note 24, at 2.
125. 45 F.2d 119, 122 (2d Cir. 1930) (holding no infringement where the same stock
characters and basic plot were used). See also Boucicault v. Fox, 3 F. Cas. 977, 982
(C.C.S.D.N.Y. 1862) (No. 1,691); Emerson v. Davies, 8 F. Cas. 615 (C.C.D. Mass. 1845) (No.
4,436).
126. Boudcaul, 3 F. Cas. at 982 ("[T]he plays of Shak[e]speare are framed out of materials
which existed long before his time, and were gathered by him" such that the skill and judgment
in the selection and exposition constitute the basis of his reputation.).
1168 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

Questioning the value or artistic worth of AI-produced work undercuts

copyright law's goal of encouraging the creation of art. Like any other original
expressive work, Al works introduce new ideas and aesthetics. They also
provide society with an insight into the most pervasive patterns of human
creativity. 127 Humans create Al as a way of furthering creative thought and
demonstrating different or non-human ways of creative expression. From the
"enjoyably strange" Al generated short film Sunprng 12" to the "shockingly
lovely" Deep Dinosaur art,12 9 there already exists widespread public
recognition and demand for Al generated works.

IV. AUTHORSHIP
Like creativity, authorship's reliance on the role of the romantic author has
created discomfort with recognizing machine produced works. The 2017
update to the Compendium on Copyright Practices repeatedly makes very clear
that a human author must have created the work to be copyrighted.1

"
However, prior to Naruto and the 2017 edition of the Compendium,
authorship was always impliedly, but not explicitly, human." Specifically, that
definition states, "the creator of the original expression in a work is its author.
The author is also the owner of copyright unless there is a written agreement
... In cases of works made for hire, the employer ... is considered to be the
author."13 2 Copyright law does not distinguish between individuals,
corporations, or metaphysical beings for the incentivization of aesthetic works.
Al can and should fit into this regime. The next Section explains how lessons
from the works made for hire doctrine can ground Al authorship as a viable
concept.
A. LESSONS FROM THE WORKS FOR HIRE REGIME

The works made for hire (WMFH) regime is an exception to the rule that
only the author can rightfully claim copyright. If a work is made for hire, an

127. See, e.g., BENJAMIN, SUNSPRING (an Al produced science fiction screenplay that
repeats the line "I don't know"').
128. Michael Nordine, Watch 'Sunspring ' a Short Sd-Fi Film Written by an Atifidal Intelligence
Algorithm, INDIEWIRE (June 9, 2016, 3:10 PM), https://www.indiewire.com/2016/06/watch-
sunspring-sci-fi-artificial-intelligence-ai-written-1201687033/.
129. Liszewski, supra note 119.
130. U.S. COPYRIGHT OFFICE, supra note 2, 302, 306, 313.2, 803.5(C), 808.7(C)
(wherever authors are mentioned, the human author requirement is reiterated: the cited
sections are the human authorship requirement itself as well as examples of where it arises
throughout the Compendium).
131. See Definitions, U.S. COPYRIGHT OFFICE, http://www.copyright.gov/help/faq-
definitions.html (last visited Feb. 11, 2020).
132. Id.
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1169

employer is considered the author in which initial copyright vests even if an

employee actually created the work." Early courts decided WMFH cases on
the basis of implied intent rather than who the original author was." Rather
than having copyright vest first in the author and then be subsequently
assigned, the initial allocation of copyright between parties was "implied in the
[employment] relationship between them."135 Those who procured the work
(like book commissioners) or orchestrated its creation (like a theater owner)
were granted copyright on the basis of the nature of the relationship. 136 By the
nineteenth century, courts evolved their rationale. The key consideration was
"the degree of involvement of the employer's representatives in the creative
process [and] the supervision exercised or in the expenditure undertaken by
137
it." In effect, this assumed an employer had ownership over the employee's
work unless the employee could point to an explicit, contractual assignation of
ownership. 138 Rights no longer vested in the artist by the nature of their
romantic authorship. Instead, the work became an entity separated from the
employee's labor under the understanding that their artistry was essentially a
tool to be used by the whims of the employer.1 3 9
WMFH presents several important consequences for an Al copyright
regime. The policy rationale behind WMFH is to incentivize employers and
grant them control over works made on their behalf.140 Like a Hollywood
studio, an Al programmer invests time and resources to develop protectable
systems. This investment should be rewarded in kind. Programs like Deep
Dinosaur and programmers like Chris Rodley serve the general copyright goals
of innovation and culture and should be protected accordingly.
Further, corporations are non-human entities in which copyright vests.
Consequently, WMFH strengthens the psychography cases' conclusion that

U.S. COPYRIGHT OFFICE, CIRCULAR 9: WORKS MADE FOR HIRE 1 (2012).

133.
134. Bracha, supra note 108, at 253.
135. Id.
136. Id at 252-54 (examining Keene v. Wheatley, 14 F. Cas. 180 (C.C.E.D. Pa. 1861) (No.
7644) and Lawrence v. Dana, 15 F. Cas. 26 (Clifford, Circuit Justice, C.C.D. Mass. 1869) (No.
8136)).
137. Id. at 254.
138. Id. at 255.
139. See Robert C. Denicola, Ex Machina: Copyright Protectionfor Computer-GeneratedWorks,
69 RUTGERS U.L. REV. 251, 276 (2016) ("William Patry, in his treatise on copyright law, views
the employer as author rule as an example of an 'instrumental approach' to copyright
authorship.")
140. Shlomit Yanisky-Ravid, Generating Rembrandt: Arifidal Intelligence, Copynght, and
Accountability in the 3A Era-The Human-Like Authors are Already Here-A New Model, 2017
MICH. ST. L. REv. 659, 711 (2017).
1170 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

copyright can subsist in a non-human work."1 By divorcing the artist from

their presupposed copyright, WMFH creates an important distinction between
the author-in-law, with whom the rights vest, and the author-in-fact, who
undertook the creative steps to produce the work. 2 This distinction is an
important and major exception to copyright resting with the actual artist. In so
doing, it is also as a means of recognizing the creator of a work without
allocating copyright to them. As a legal fiction, it allows companies to
maximize the profitable labor of their employees by appealing to the need to
protect investment and created "value" as the basis of copyright.1 4 3 This
adaptation mirrors the evolution of originality as not being intrinsically
meritorious by reason of romantic authorship, but as a representation of
subjective market value over which objective courts have limited, if not
arbitrary, purview.

V. APPROACHING THE FUTURE: MODELS AND

RECOMMENDATIONS FOR THE TREATMENT OF
ARTISTIC AI COPYRIGHT
The current state of U.S. copyright law for Al generated works is uncertain
and needs to be definitively clarified to establish the rights and limits of a
rapidly expanding industry. As it stands, "copyright law is not currently
structured to accommodate the particular authorship matrix of people-who-
write-programs-that-make-art."144 Although the Compendium and its human-
authorship requirement does not have the force of law, it nonetheless reveals
the attitudes and practices of the Copyright Office and is taken to "provid[e]
expert guidance" to courts.1 4 5 Courts are unlikely to follow Torah So fs
reasoning and extend copyright to the software owner where the software does
the "lion's share" of the work. Rather, relying on Naruto and the Compendium,

141. The Ninth Circuit in Naruto v. Slater noted that, although courts have allowed
corporations to sue, corporations "are formed and owned by humans; they are not formed or
owned by animals." Nevertheless, this distinction made by the court only serves to illuminate
how corporations under the WMFH regime are interpreted as non-human human entities
nonetheless entitled to legal rights. 888 F.3d 418, 426 n.9 (9th Cir. 2018). See Bridy, supra note
24, at 26 ("The work made for hire doctrine is a more fitting framework within which to situate
the problem of Al authorship because it represents an existing mechanism for directly vesting
ownership of a copyright in a legal person who is not the author-in-fact of the work in
question.").
142. See Bridy, supra note 24, at 26 ("the work made for hire doctrine acknowledges a
disidentity between the author-in-fact (the employee or contractor) and the author-in-law (the
employer or other person for whom the work was made)").
143. Bracha, supranote 108, at 260-61.
144. Bridy, supra note 24, at 22.
145. U.S. COPYRIGHT OFFICE, supra note 2, at 1.
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1171

a court would likely deny copyright protection for an AI-generated work.

While U.S. programmers are without an avenue to rights through litigation,
countries such as Hong Kong (SAR),146 India,147 Ireland,148 New Zealand,'
and the United Kingdom 15 allow AI-produced works to be protected.151 This
discrepancy makes clear that the United States needs to address the status of
artistic Al works in the near future.
A. A BRIEF OVERVIEW OF PROPOSED MODELS OF Al COPYRIGHT
OWNERSHIP BY U.S. LEGAL SCHOLARS

Among U.S. legal scholars, the discourse on copyrightable Al works is

governed by four main models. Each differently reflects concerns of how to
place Al within authorship. Each tries to speak to whether the work is:
(1) copyrightable in the first place (if non-humans can be creative or
"original"); (2) whether and how authorship can vest in a non-human; and
(3) the administrability of the proposed changes. To evaluate the models, this
Section asks three questions that bear on the efficacy and rationales: Who is
this meant to incentivize? Is there a need for this model? How would this affect
the market for these works?
1. Works Remain in the Public Domain

Under the current copyright regime, Al works are in the public domain.
Even though it is the official status quo, enforcing it would likely mean
heightened scrutiny on behalf of the Copyright Office to ensure that
programmers are not copyrighting Al generated works, as they currently are.
Proponents of this model claim that incentive is maintained in two ways.152
First, because programmers can register copyright on the Al software itself,
there is no logical reason for furthering the stream of ownership down to the

146. Hong Kong Copyright Ordinance, (1997) Cap. 528, 40, 44, 11(3), 17(6).
147. Copyright (Amendment) Act, 1994, No. 38, Acts of Parliament, 1994 2(vi) (India).
148. Copyright And Related Rights Act 2000 (Act No. 28/2000) (Ir.) 21(f), available at
http://www.irishstatutebook.ie/eli/2000/act/28/section/21/enacted/en/html#sec21,
archived athttps://perma.cc/XWA4-D7QP.
149. Copyright Act 1994, s 5(2)(a) (N.Z.).
150. Each country's statutory language either replicates or closely matches the United
Kingdom's, where "[i]n the case of a literary, dramatic, musical, or artistic work which is
computer-generated, the author shall be taken to be the person by whom the arrangements
necessary for the creation of the work are undertaken," and only differ in the term granted-
either seventy or fifty years from the date of creation. Copyright, Designs and Patents Act
1988, c. 1, 9(3), 12(2) (Eng.).
151. Andres Guadamuz, Artificial intel/gence and copyright, WIPO MAG. (Oct. 2017),
https://www.wipo.int/wipo-magazine/en/2017/05/article_0003.html.
152. Ana Ramalho, Will Robots Rule the (Artistic) World?A ProposedModelfor the Legal Status
of Creationsby ArtificialIntelgence Systems, 21 J. INTERNET L. 11, 21-22.
1172 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

Al outputs. 153 However, much like a manufacturer makes money by the

product, not the production line, a programmer creating an artistic Al would
be hard-pressed to monetize the program itself. The public domain
proponent's response is that the creators of the Al can also be incentivized
through the right to dissemination. 154 By treating owners of the Al as a
publisher, Al generated works would exist behind a paywall and allow the
existing structure of Al copyright to remain unchanged. 155 Reviews of the
efficacy of paywalls are mixed, but the general consensus is they are only
sustainable if the provider has a high level of reputation and uniqueness of
content. 156 Proponents further argue that this gap may be filled by sites like
Patreon, an "online tip jar" where artists crowdfund directly from fans either
through one-time donations or subscriptions. 157
Theoretically, this model would work best for hobby artists like Chris
Rodley, whose Deep Dinosaur and iterations thereof would likely bring in
revenue for the high-resolution printable versions. 158 However, if those works
are uploaded and available for download on another site, Rodley and others
like him would be unable to enforce an action against the pirating site, since
the works are technically in the public domain. This could incentivize a
generation of Al works that are impossible or difficult to copy, such as
experiential or constantly changing outputs. More likely though, with the
knowledge that the free version will be uploaded elsewhere, fewer patrons
would pay and market incentivization would decline. Additionally, for
companies like Endel, who seek to monetize their investment via streaming
services, this model would make that impossible, since streaming licensing is

153. Id.
154. Id.
155. Id.
156. See Kristen Senz, Are Paywalls Saving Newspapers?, HARVARD Bus. SCH. WORKING
KNOWLEDGE (July 8, 2019), https://hbswk.hbs.edu/item/are-paywalls-saving-newspapers;
see also Mike Masnick, The Media's Paywall Obsession Will End In DisasterForMost, TECHDIRT
(May 8, 2018, 9:37 AM), https://www.techdirt.com/articles/20180506/11501539779/
medias-paywall-obsession-will-end-disaster-most.shtml ("It's not that we think that paywalls
are somehow "bad," but that (1) for most publications, they won't actually work and (2) they
are quite frequently counterproductive.").
157. Adi Robertson, Inside Patreon, The Economic Engine Of Internet Culture, THE VERGE
(Aug. 3, 2017, 11:36 AM), https://www.theverge.com/2017/8/3/16084248/patreon-profile-
jack-conte-crowdfunding-art-politics-culture.
158. See, e.g., Phil Wang, Phil Wang is creating This Person Does Not Exist, PATREON, https://
www.patreon.com/lucidrains (last visited Feb. 11, 2020) (the creator of
thispersondoesnotexist.com currently has seventeen patrons at $37 dollars a month as of
February 2020).
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1173

predicated on copyrightability. 159 Without this assurance, large companies in

those fields, like Warner Music, would be disincentivized to buy into the new
industry. This would shift the market away from investment in music and video
media, since those two areas are largely dominated by streaming services.

2. Al as Both the Author and Owner

The second model would have the Al be both owner and author. The need
for this model is unclear because incentives are meant to entice human
decision makers. Proponents are often thinking about the distant future in
which robots and androids may have fully vested rights and possess the
discretion over whether to produce future works. 160 But until the day that a
robot expresses a desire for copyright on its works, this model only
incentivizes programmers more interested in far-reaching philosophical and
legal implications of granting non-human actors human rights than monetizing
their creations.
This model would have some bizarre impacts on the market. It does not
deal with issues of liability. It is also unclear how programmers who do seek
incentive would reap the rewards of investment if all interests vest in the Al
itself. The only logical element to come out of this model would be a fixed
term limit, since Al do not have "lives." Ultimately, it would likely chill
investment by being even more unmanageable and ambiguous than the current
regime.
3. Modifying the Definition of an "Employee" Under the Work for Hire
Regime
Popular among American scholars is a modified WMFH regime. Overall,
this model is not likely to chill investment, since it establishes all the copyright
protections of a traditional work. This proposal essentially modifies the
definition of an "employee" under 17 U.S.C. 101 to include AL. 161 As in
traditional WMFH, copyright would then vest in the human or corporate entity
that procured the work. Some scholars assume that the programmer is the
employer, 162 but others take the "person for whom the work was prepared" to
mean either the programmer or the end user. 163 When attributing ownership

159. See Music Licensing ModerniZation Act, U.S. COPYRIGHT OFFICE, https://
www.copyright.gov/music-modernization/115/ (last visited Feb. 11, 2020).
160. See Andrew J. Wu, From Video Games to Artificial Intelligence: Assigning Copyright
Ownership to Works Generated by Increasingly Sophisticated Computer Programs, 25 AIPLA Q.J. 131
(1997).
161. Hristov, supranote 5, at 447.
162. See, e.g., id. at 445.
163. 17 U.S.C. 201(b) (2018).
1174 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

to the programmer, the incentive clearly rewards the work done to create the
Al, and also ensures that liability for infringement falls on the person best
placed to prevent it.
But some scholars, like Professor Shlomit Yanisky-Ravid, suggest that
when users interact with generative Al, they should be considered the owners
of the works. 164 For tool-like Al, this makes sense, because the user is doing
the lion's share of the work. But these works can already receive copyright
protection because they involve substantial human creativity. 165 When it comes
to truly generative works, it is unfair to attribute copyright to a user. Unless
mitigated through licensing or contract, an unaware user clicking through a site
like thispersondoesnotexist.com166 could suddenly be on the hook for an
infringing work that they had no part in producing, other than clicking
"generate."
This model also acknowledges the issue of protection term length. A
traditional work is protected by copyright for the life of the author plus seventy
years. 167 The term of copyright protection of a work made for hire is ninety-
five years from the date of publication or one hundred twenty years from the
date of creation, whichever expires first. 168 While still an excessively long
period amount of time for Al works to be protected, this explicit limit
nonetheless speaks to fairness within the copyright system and ensures works
eventually enter the public domain.
This model aims to quickly address protecting Al works using an existing
copyright mechanism. However, the Supreme Court has suggested that the
WMFH doctrine is very limited in scope, and applies only where "Congress
has expressed a clear and explicit intent to override section 102."169 Because
implementing such a substantial change to the doctrine would require new
legislation, it seems ineffective to try and push Al into a regime that is primarily
about assignation of agency. 170 As of yet, Al are not sentient and do not have
agency-so it is incongruous to incorporate them in a regime that requires a

164. Yanisky-Ravid, supra note 140, at 707 ("[W]e should view Al systems as working for
the users, and hence the users should bear accountability for the systems' production, in
addition to the benefits thereof").
165. See supra Section IIB.
166. THIS PERSON DOES NOT EXIST, https://thispersondoesnotexist.com/ (last visited
Feb. 11, 2020) (a web-based GAN that generates a new human face each time the site is
refreshed).
167. U.S. COPYRIGHT OFFICE, WORKS MADE FOR HIRE, supra note 133, at 3.
168. Id.
169. Yanisky-Ravid, supranote 140, at 715.
170. See U.S. COPYRIGHT OFFICE, WORKS MADE FOR HIRE, supranote 133, at 2 (devoting
a section to "Agency Law").
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1175

test to determine whether the work was actually made in an employment

context and the employee was an "agent." An Al and its subsequent works
will always be attributable as a creation to an individual or a team. Instead of
using an existing regime as a red herring to distract legislators from the actual
changes to the law they would implement, it is worthwhile to explicitly and
directly create a carve-out for Al works.
4. Applying Existing U.K Law to the United States

Finally, some proponents advocate for the wholesale adoption of current

U.K. copyright law, which treats computer-generated works as if they
originated directly from the programmer. That law states that "[i]n the case of
a literary, dramatic, musical or artistic work which is computer-generated, the
author shall be taken to be the person by whom the arrangements necessary
for the creation of the work are undertaken." 171 The United Kingdom's
Copyright Act goes on to define a computer-generated work as one that "is
generated by computer in circumstances such that there is no human author
of the work." 172 This provision creates an exception to human authorship by
recognizing the time and resources that go into creating a generative Al, even
if the actual work is undertaken by the machine. 173 Additionally, when it comes
to the copyright term, the United Kingdom limits computer generated works
to fifty years from the end of the year of creation. 174
Like the modified WMFH regime, this model grants full copyright
protection and would therefore not chill investment. Because this regime fits
how the market is functioning now in terms of programmers putting their
names on AI-generated works,175 it would not change applicable streams of
revenue. The added benefit of a shortened term limit would ensure that
programmers do not benefit for longer than is fair. Critics of this model note
the ambiguity of the person who made the "arrangements necessary" for the
work to be generated. 176 Depending on the interpretation, this could be the
programmer or the user of the program. If the engineers or programmers of
the Al are taken to be the copyright owners, return on investment is
guaranteed and liability rests fairly on the creators. 177 However, this should be

171. Copyright, Designs and Patents Act 1988, c. 1, 9(3) (Eng.).

172. Id. c. 10, 178.
173. Guadamuz, supranote 151, at 4.
174. Copyright, Designs and Patents Act 1988, c. 1, 12(7) (Eng.).
175. See, e.g., ENDEL.IO, supranote 35.
176. See Bridy, supra note 24, at 27.
177. Pratap Devarapalli, Machine Learning to Machine Owning: Redefining the Copyright
Ownershipfrom the perspective ofAustralian, U.S., UK and EU law, 40 EUR. INTELL. PROP. REV.
722, 727 (2018).
1176 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151

clearly distinguished within the text of the law to limit disputes arising out of
ambiguity.
B. PROPOSED MODEL: Al AS THE AUTHOR-IN-FACT AND THE
PROGRAMMER AS THE AUTHOR-IN-LAW

This Note's proposed model lists the Al as the author-in-fact and the
programmer as the author-in-law. This distinction serves to explicitly indicate
that Al generated works will be treated differently than traditional
copyrightable works and allows the Copyright Office to easily delineate
between the two. This model would include a term limit-either fifty years to
meet international standards 178 or fewer depending on what is deemed
equitable by the legislature. For fair allocation of benefits and risks, the
programmer, not the user, is explicitly listed as the author. Thus, the person
reaping rights-based incentives also faces liability if their program is faulty. This
model specifically incentivizes creators of generative Al who would be unable
to monetize their software without monetizing the output.
Creators are already copyrighting their Al generated works and profiting
off them. But there is no mechanism for enforcing the right against
infringement nor does the term of life of the programmer plus seventy years
accurately reflect the payoff of time and resources for the process of
generation following the initial creation of the Al. Further, any legal challenge
against a generated work would likely lead to its release into the public domain,
as implied in Naruto v. Slater. A definitive regime needs to be put into place to
clearly delineate the bounds of what rights are and are not applicable to AI-
generated works. As proponents of the WMFH model suggest, it is expedient
to rely on ties to existing U.S. copyright law. This model uses a legal fiction of
authorship parallel to the one put forth by WMFH, in which copyright vests
as a matter of law in someone who is not the author-in-fact. As with the
WMFH and U.K. models, this proposed model would ensure that the market
continues to encourage creators of Al, while the term limit reflects that each
work is less impactful as a human-created work.
Finally, this modification would be made directly to the Copyright Office's
"Copyrightable Authorship" section. Legislation is needed to make any of the
changes listed in the above models (excepting public domain). It is
unproductive for the legislature to decide on an adequate or easy-sounding,
but problematic option, such as WMFH. 179 An independent clause on Al

178. See supranotes 146-50.

179. See, e.g., Hristov, supra note 5, at 452-53; see also Timothy L. Butler, Can a Computer be
an Author-CopyrnghtAspects ofArtfidal Intellgence, 4 HASTINGS COMM. & ENT. L.J. 707, 741-
2020] DO ANDROIDS DREAM OF COPYRIGHT? 1177

would not only clarify the distinctiveness of Al as a copyright matter, it would

also speak to the willingness to address technological revolution head-on, in
the footsteps of President Lincoln and Justice Holmes.

VI. CONCLUSION
It is important to encourage creative, not just utilitarian, Al. Artistic Al not
only develops the full potential and breadth of Al but provides a point of
insight into what creativity means, how humans reflect on creative works, and
how alternative creative sources can drive new understandings and
innovations. As AI-as-an-industry and source of creative works continues to
expand, the need for clarification and a modified regime is pressing. A patch
fix like adding "human" to authorship requirements or modifying the
definition of "employee" to slip Al into an ill-fitting regime is inefficient. The
coming update needs to keep the policies and historical trajectories behind
copyright in mind. Congress and the courts have altered the doctrine and
requirements in the face of new technologies before-this Note and its
proposed changes argues for a continuing willingness to adapt the law to
technological advances. While AIs do not dream of copyright protection, their
creators and proponents do. They need a copyright regime that adequately
reflects and delineates the protection of that dream.

42 (1982) (explaining that the contractual nature of WMFH imposes limitations on redefining
the meaning of "employee").
1178 BERKELEY TECHNOLOGY LAW JOURNAL [Vol. 35:1151