Hastings Communications and Entertainment Law Journal

Volume 4 | Number 4 Article 11

1-1-1982

Can a Computer be an Author - Copyright Aspects

of Artificial Intelligence
Timothy L. Butler

Follow this and additional works at: https://repository.uchastings.edu/

hastings_comm_ent_law_journal
Part of the Communications Law Commons, Entertainment, Arts, and Sports Law Commons,
and the Intellectual Property Law Commons

Recommended Citation
Timothy L. Butler, Can a Computer be an Author - Copyright Aspects of Artificial Intelligence, 4 Hastings Comm. & Ent. L.J. 707
(1982).
Available at: https://repository.uchastings.edu/hastings_comm_ent_law_journal/vol4/iss4/11

This Note is brought to you for free and open access by the Law Journals at UC Hastings Scholarship Repository. It has been accepted for inclusion in
Hastings Communications and Entertainment Law Journal by an authorized editor of UC Hastings Scholarship Repository. For more information,
please contact [email protected].
Can a Computer be an Author? Copyright
Aspects of Artificial Intelligence

By TIMOTHY L. BUTLER*

Helene watched John and cogitated: A supper with him? Dis-

gusting! A supper would facilitate a dissertation and a disserta-
tion or tale was what John carefully wanted to have. With what
in mind? Wine, otters, beans? No! Electrons! John simply was
a quantum logician; his endless dreams were captivating and
interesting; at all events Matthew, Helene, and Wendy were as-
sisting him in his infuriated tries to broaden himself. Now le-
gions of dreams itched to punch Wendy's consciousness. Yet
John whispered, "Just a minute! Helene's a maid, I'm a quan-
tum logician; can maids know galaxies and even stars or a mul-
titude of galactic systems? ...Can maids realize electrons?1

I
Introduction
Few people would doubt the significance of the computer in
modern United States society, but most would doubt a com-
puter "authored" the above quoted excerpt. Perhaps trivial
and somewhat nonsensical, it nonetheless evidences a signifi-
cant trend in computer science towards more powerful, crea-
tive and autonomous computer programs. In the computer
science field of artificial intelligence (AI), these developments
in computer programming pose unique problems in copyright
protection of computer software. Computer software capable
of automatic programming, inductive analysis and knowledge-
based problem solving will soon challenge the legal concepts of
authorship and originality central to the common law and stat-
utory basis of copyright. As the threshold between man and
machine narrows, courts will have to determine the legal sta-
tus of the apparently creative work product of a machine,
which, if produced by a human, would be afforded copyright
protection.
* Member, Third Year Class; B.A., University of California, Irvine, 1977. A ver-
sion of this note has been entered in the Nathan Burkan Memorial Competition.
1. RACTER, Soft Ions, OMNI,April 1981, at 96, 97.

707
 COMM/ENT [Vol. 4

To evaluate copyright issues presented by artificial intelli-

gence, this note will first examine the field of artificial intelli-
gence and its various themes, current status and indications of
future development. Next, this note will briefly discuss the ba-
sic structure and function of computer programs. Finally,
copyright law will be assessed as it pertains to computer pro-
grams, focusing on the originality and authorship aspects of
copyright law affected by artificial intelligence to decide if a
computer can or should be considered an "author" under the
federal copyright laws and the consequences of such
protection.

II
Artificial Intelligence
The ultimate goals of AI researchers can roughly be divided
into two non-exclusive parts. 2 One group of AI investigators
uses computers in attempts to simulate, and thereby under-
stand, human behavior.3 The other group of researchers uses
AI principles to make computers perform tasks unsuited to
human capabilities, thus extending the power of man in his en-
vironment.4 Both groups use AI machines' as tools to achieve
their respective goals.
Although the field of AI has been filled with attempts to cre-
ate machines which perform human tasks with the speed and
efficiency of man, AI machines are still "remote from achieving
a level of intelligence comparable in complexity to human
thought. ' 6 In early contemplation of machine intelligence,

2. This analytical framework is a subset of a more complete description of the

field of artificial intelligence research dividing research into three parts: (1) under-
standing human behavior; (2) applications using artificial intelligence techniques; and
(3) theoretical research not bound by goals of applications possibilities nor the con-
cept of "human" intelligence per se. See generally Waltz, Artificial Intelligence, Sc.
AM., October 1982, at 118. For a philosophical treatment of issues concerning artificial
intelligence, see generally D. HOFSTADTER & D. DENNETT, THE MIND'S I (1981).
3. H. Bremmerman, Artificial Intelligence: A Study of Intelligence in Machine,
Animal, and Man iii (June 1974) (unpublished manuscript available at U.C. Berkeley).
4. Id. at v35.
5. For the purpose of this note, an "AI machine" is composed of a computer, AI
software and associated input, output, sensing and environment-manipulating devices
like mechanical arms, etc.
6. Firschein, Forecasting and Assessing the Impact of Artificial Intelligence on
Society, Third International Joint Conference on Artificial Intelligence: Advance Pa-
pers of the Conference (August 20-23, 1973) (Stanford University).
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 709

noted British mathematician and logician Alan Turing 7 stated

that, "[I]n the last analysis, the question of whether a com-
puter can 'think' or not can be answered in the affirmative if a
human being, by asking it questions, could not tell from the
answer whether he were interrogating a man or a machine." 8
Prompted by the challenge of the Turing test, programs have
been developed which allow a person to engage in a limited
"conversation" with a machine. Often found in museum exhib-
its, these programs allow "hands on" use of a computer by the
public in a non-threatening, entertaining manner. However, di-
alogues resulting from such programs are usually clever incor-
porations of the user's answers into questions "asked" in a
predetermined sequence by the machine, allowing for branch-
ing9 to different patterns of questions upon input of specific an-
swers. Similarly, computers programmed to play chess,
checkers, backgammon or other games, also employ AI tech-
niques to simulate human behavior.10 The user may tend to
anthropomorphize the machine, but few persons are fooled
into believing the machine is participating as an intellectual
entity.
Other researchers in artificial intelligence concern them-
selves with applying computers to tasks unsuited to the human
intellect. Computers make possible "alien intelligence" based
on artificial knowledge." This is knowledge of a sort that the
human brain can not obtain. 2 Recognizing the immense prac-
tical importance of this application of AI, one researcher has
remarked that replacing "a human chess player by a machine
may have scientific and philosophical interest but it has little
immediate practical effect. In contrast, development of capa-

7. For a brief account of Alan Turing's contributions to computer science, see P.

MCCORDUCK, MACHINES WHO THINK 50-64 (1979).
8. Milde, Can a Computer be 'An Author' or an 'Inventor?,51 J. PAT. OFF. Soc'y.
378 (1969) (citing Turing, Can a Machine Think?, 4 WORLD OF MATH. 2099, 2193 (1956)).
9. "Branching" in computer jargon denotes a logical selection or decision to per-
form a given task when supplied with a given set of data. For example, if a user is
asked to input "yes or no" as the answer and he inputs "zzz" instead, the computer
might "branch" to an error-handling subroutine to query the user about his answer.
. 10. See generally Berliner, A Chronology of Computer Chess and its Literature, 10
ARTIFICIAL INTELLIGENCE 201 (1978).
11. Bremmerman, supra note 3, at v35.
12. "Knowledge" in this context means organizing and manipulating facts and in-
formation. See WEBSTER'S NEW COLLEGIATE DICTIONARY 633 (1981). Large masses of
information with complex inter-relationships are not comprehensible and therefore
not "obtainable" by the human brain. In this sense, such a body of organized informa-
tion ("knowledge") is "alien" to humans. See generally Bremmerman, supra note 3.
 COMM/ENT [Vol. 4
1 3 Sim-
bilities that no human being possesses conveys power.'
ulation models are a common example of the "power" of such
applications. By simultaneously solving complex equations
with numerous variables and vast amounts of data, simulation
models can generate solutions necessary in corporate plan-
ning, economic analysis, operations research, defense prepara-
tion and agricultural production.
However, it is unimportant whether one views artificial intel-
ligence as limited to an emulation of human intellect or in a
broader sense to include "alien," non-human intelligence.
Both views recognize the possibility of machine intelligence,
and both assume that "when in a computer knowledge base
and deductive capabilities are combined, then artificial intelli-
gence results."'1 4 Moreover, each view incorporates aspects of
behaviorism, mathematics and cybernetics. 5 These major dis-
ciplines of AI have focused research and development on lan-
guage understanding, problem solving, sensory perception,
modeling, learning and adaptive systems, robots and games. 6
As a result of such research, automated data retrieval and in-
quiry systems, industrial robots, medical diagnosticians,
weather 7and economic models, and teaching aids have been
created.
Although AI functions are carried out by electromechanical
devices such as printers, robot arms, relays, etc., the underly-
ing control of the machine is provided by the computer pro-
gram loaded into the resident computer. Whether the goal is to
duplicate human behavior or to simulate a complex environ-
ment, the element of a "computer program" is common to
both.' 8 Thus, the development and implementation of sophisti-
cated computer programs is the principal means investigators
use in attempts to create artificial intelligence. This note fo-
cuses upon two types of AI programs which mimic human
thinking and intelligent activities in two specific ways: (1) AI

13. Bremmerman, supra note 3, at v38.

14. Id. at ii2.
15. Behaviorists seek to duplicate goal-seeking behavior. Mathematicians con-
struct devices capable of proving logical and geometric theorems with great speed and
accuracy. Cyberneticists try to duplicate and extend human sensory and locomotive
functions. Id.
16. Firschein, supra note 6, at 106.
17. Id. at 112.
18. See note 56, irtfra.
No. 41 COPYRIGHT AND ARTIFICIAL INTELLIGENCE 711

programs functioning as software code generators 19 and as "au-

tomatic" programmers;2 0 and (2) AI programs producing tradi-
tional literary works comparable to those of a human author.
Although both of these functions are accomplished by an AI
program, one outputs a computer program [hereinafter Type
I], and the other a story [hereinafter Type II]. The reader
must at all times strive to distinguish the AI program from its
product. The product of Type I AI programs is a computer pro-
gram or a section of program code.2 The product of Type II AI
programs is a story which appears to have been authored by a
human. The copyrightability of the products of Type I and
Type II AI programs is the subject of this note.

A. Code Generators and Automatic Programming

The development and maintenance of applications software
is usually the most costly element of a computer system, be-
cause software production is labor intensive and requires a
skilled labor force.22 As the use of computers has spread
throughout our society, demand has increased for quality pro-
grammers and software.23 Computer users have consequently
sought ways to reduce the costs of programming and maintain-
ing their systems. It is not surprising that one of the solutions
to the escalating cost of software development has been to let
the computer "write the program" itself (Type I AI programs).
Money and time can be saved because uniformly written pro-
grams result, and man-hours are spent more efficiently. This

19. A code generator is a computer program which translates a program design

into an accurate, detailed representation in a language the computer can interpret and
execute.
20. "Automatic programming" is a technique using the computer to help plan and
code a problem into a programming language. See C. MEEK, GLOSSARY OF COMPUTING
TERMINOLOGY 14 (1972).
21. "Program code" is the set of instructions for a given computer, each of which
when decoded by the computer will cause a predictable change in the state of the
machine. Id. at 45.
22. Two-year and four-year degree programs in computer science are offered by
community colleges and universities respectively. Most employers require a mini-
mum of two years college level education in addition to some outside work experience
in a computer-related environment. For insight into the labor-intensive nature of com-
puter programming, see generally T. KIDDER, THE SOUL OF A NEW MACHINE (1981).
23. "Software" is a general term denoting programs which can be run on a given
computer, whether stored temporarily or permanently within the machine itself or on
some portable media like magnetic tape. Though software stored permanently in a
computer is usually termed "firmware" because of its relatively immutable nature,
both software and firmware are means of storing computer programs. Computer pro-
grams, not their medium of storage and expression, are the focus of this note.
 COMM/ENT [Vol. 4

trend towards automating the programming process is evident

in the development of programming language "code" genera-
tors and in AI research into problem-oriented and knowledge-
based programming languages.
1. Code Generators
A code generator is a computer program which assists the
programmer in writing a computer program. 24 At present, code
generators are sometimes used to write data entry sections of
programs, output report procedures, documentation and tuto-
rial sections, and associated video screen display formats.25
The programmer interactively answers questions posed by the
computer, and the computer automatically generates sections
of the program code.26 In answering questions presented by
the computer, the user supplies parameters, variables and la-
bels relevant to the application. The computer supplies the
code and tailors the procedure algorithm according to the an-
swers input. The computer then generates and stores the
coded instruction. Finally, the programmer modifies the list-
ing27 to suit his needs and the program is ready for use.
As code generators become more sophisticated, the program-
mer will function as a mere problem-specifier. His job will be
to accurately inform the computer of the problem, wait for the
computer to generate the program and then modify it to suit
the particular application. The programmer will not concern
himself with formulation of the algorithm coded by the com-
puter. He will only have to understand the factors relevant to a
concise statement of the problem and to be fluent in the pro-
gramming language used in the program output by the com-
puter if minor debugging 28 or tailoring is necessary for proper
implementation. The economic pressures behind the develop-
ment of code generators ensure that the computer will become
increasingly efficient and effective at this function.29
24. See generally Jacobs, GeneratingProgramsAutomatically-Let Your Apple II
Do the Programming, BYTE, Dec. 1981, at 352, and Louden, The Last One, a program
generatorfrom D.J. 'AI', INFOWORLD, Jan. 18, 1982, at 18.
25. Id.
26. Id.
27. A "listing" is a printed list of the program as written by the programmer in a
given language (i.e., FORTRAN, COBOL).
28. "Debugging" is the process of systematically locating and correcting problems
in program logic or typographical errors which prevent the program from functioning
properly.
29. See note 49, infra.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 713

2. Problem-OrientedLanguages
The task of problem specification when using code genera-
tors is simplified by using problem-oriented languages. "A
problem-oriented language allows one to state and to solve a
whole family of problems .... The user just has to state his
problem in a descriptive manner ... .30 ALICE was the first
language of this kind. ALICE has been applied to real-life
problems involving operations research and has been more
successful than traditional programs. 31 When using ALICE, a
purely descriptive statement of the problem is entered using
mathematical sets and symbols. No algorithm 32 is supplied by33
the user. The computer systematically introduces heuristics
to derive implications and make appropriate choices.3 4 The
problem solution is obtained, and the coded algorithm used is
available for future use.

3. Knowledge-Based Systems
Code generators and problem-oriented languages like AL-
ICE evidence only two approaches towards the AI goal of auto-
mating the programming process (Type I AI programs). A
third major attempt has focused on the knowledge-based ap-
proach to automatic programming.3 5 One such project was PE-
COS, where the researchers summed up their method and
rationale as follows:
[H]uman programmers seem to know a lot about program-
ming. This suggests a way to try to build automatic program-
ming systems: encode this knowledge in some machine
readable form. In order to test the validity of this approach,
knowledge about elementary symbolic programming has been
codified into a set of about four hundred detailed rules, and a
system, called PECOS, has been built for applying 36
these rules
to the task of implementing abstract algorithms.

30. Lauriere, A Language and a Programfor Stating and Solving Combinatorial

Problems, 10 ARTIFICIAL INTELLIGENCE 29, 123 (1978).
31. Id. at 29.
32. An "algorithm" is a fixed step-by-step procedure for accomplishing a given re-
sult. See C. MEEK, supra note 20, at 7.
33. An "heuristic" is a "rule of thumb" for solving a problem. It is a description of
a procedure which may solve a given problem but one that cannot be guaranteed to
always lead to a solution. Id. at 103.
34. Lauriere, supra note 30, at 29.
35. Barstow, An Experiment in Knowledge-Based Automatic Programming,12 AR-
TIFICIAL INTELLIGENCE 73, 116 (1979).
36. Id. at 73.
 COMM/ENT [Vol. 4

Using this approach, PECOS successfully coded algorithms

for sorting data, graph theory and simple number theory.3"
Significantly, "in each case, PECOS' knowledge of different
techniques enabled the construction of several alternative im-
plementations.""8 In other words, the program algorithm and
resulting code was a product of the data entered, the problem
format and the knowledge base present. For any given set of
these elements, PECOS could generate one or more alterna-
tive coding structures because the problem-specifier had no
control over the choice of algorithm. Guaranteed duplication
of the given result program could be made impossible by sim-
ple changes in the rule-selection procedures embodied in the
knowledge base. Consequently, the inputs would not strictly
determine the output program contents. The problem-specifier
could no longer guarantee the identical program result by in-
puting the same data on different occasions.
The AI methods used in code generators, ALICE and PE-
COS offer welcome means to reduce the expensive human ele-
ment of the program development process. In areas where
programs can be formally stated in a descriptive, problem-ori-
ented language, the computer formulates the algorithm, cre-
ates the code and solves the problem specified by the
''programmer." One might question who or what should be
deemed the "author" or "originator" of the resultant program.
The need to make this distinction becomes more pressing as
the role of the human programmer diminishes and that of the
computer increases. PECOS, ALICE and code generators all
point in the same direction: a monkey trained to input a prob-
lem (e.g., "Get me the banana") into the computer could cause
the computer to create a program, execute it and then solve his
problem.3 9 Has the monkey authored the program? Perhaps.
Just as the programmer will become less involved in the pro-
cess of creating a solution algorithm and coding it into a pro-
gramming language and more concerned with the result (a
working program or its derivative product), the monkey has lit-
tle involvement in the process, its only concern is to obtain the

37. Id. at 75.

38. Id. at 73.
39. One commentator describes the "famous" monkey-banana problem: "A mon-
key sits in a cage. At the top of the cage are bananas, too high for the monkey to reach.
How does the monkey get the bananas? He pushes the box so it comes to rest under
the bananas. He climbs the box and grasps the bananas. Result: Monkey has the
bananas." Bremmerman, supra note 3, at ii4.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 715

result (bananas). If the product of an AI program (Type I) ini-

tiated by a monkey is a computer program, have the author-
ship and originality requirements of federal copyright law
been met? As Type I AI programs become more powerful, the
character and quantity of human input necessary to satisfy
traditional copyright notions of authorship and originality will
have to be addressed.

B. Simulation of Intelligence
Although devising and writing computer programs is also
traditionally a human endeavor now sometimes accomplished
wholly or partially by a computer, RACTER is a computer pro-
gram that simulates the written output of a human author
(Type II AI program).4 Using a vocabulary stored in its mem-
ory, RACTER applies grammatical rules to construct semi-co-
herent stories in English. The author of the program stated
that "once the story-writing program is set into motion the out-
put is not only novel and a priori unknowable but also cohe-
sive and apparently thoughtful. It is crazy thinking, I grant
you, but thinking conducted in perfect English."' 4 1 RACTER
picks nouns and verbs, adjectives and adverbs, at random from
lists that [the author] supplies. Then it tinkers with them and
strings them together according to the rules of grammar.
RACTER conjugates verbs, keeps track of singular and plural
nouns and of male 42 and female characters, and chooses verbs
and pronouns to fit.
The source program 43 and the vocabulary are supplied by the
original author of RACTER. The products of the machine,
simulating the output of a human author, are those of the com-
puter and its software. As programs like RACTER become
more sophisticated, the style and coherency of the resulting
stories should improve. The human will merely be required to
push a button to execute the program. After the human starts
the program, the results are not foreseeable or predetermined;
the story has possibly not been "conceived" in the sense of
original authorship required by the copyright laws.' Under
40. RACTER, supra note 1, at 97-98.
41. Id.
42. Id. at 98.
43. A source program is the coded instruction set usually written in a high level
procedural language (FORTRAN, COBOL, etc.) which has been input by the user and
represents the human-readable version of the compiled object module.
44. See notes 127, 136 and accompanying text, infra.
 COMM/ENT [Vol. 4

what circumstances should "authorship" be determined by

who pushes the button? Before we answer this question, we
must understand a bit more about the workings of the digital
computer and computer programs.

III
The Modern Digital Computer and Computer
Programs
A. Historical Development
The history of the digital computer has been characterized
by a decrease in physical size, a reduction in cost per calcula-
tion, an increase in commercial applications and the creation
of a mature computer industry and associated support technol-
ogies. Although Charles Babbage first formulated the concept
of an "Analytical Engine" capable of automatically performing
numerous mathematical functions in 1833, 45 the first electronic
digital computer with a stored program capability was not op-
erational until over a century later when the EDSAC (Elec-
tronic Delay-Storage Automatic Computer) was constructed at
Cambridge University in 1949.46 Since the 1940's, computer
technology has passed through several generations. These
generations are distinguishable by the technology used, the
programming languages employed and by type of function or
application sought or achieved."
45. S. MANDELL, COMPUTERS AND DATA PROCESSING CONCEPTS AND APPLICATIONS
22 (1979).
46. Id. at 27.
47. The first generation, (1951-1959), was characterized by vacuum tube compo-
nents. Data was input by paper tape or punched cards. The languages available to the
programmer were restricted to actual binary code and assembly-level symbolic lan-
guages. Mainly used for military and scientific applications, these computers were
also sometimes employed in the business environment to expedite payroll and billing
tasks. Though much faster than the earlier electromechanical computers, these ma-
chines generated large amounts of heat and experienced reliability problems because
of vacuum tube failure.
During the second generation of computer technology, (1959-64), vacuum tubes were
replaced by transistors, magnetic drums replaced magnetic cores and assembly and
machine level languages were supplemented by higher-level, procedural languages
like FORTRAN. Data was input from or stored on magnetic tapes, paper cards,
punched cards or from the keyboard. These machines were less expensive, smaller
and more reliable than the previous generation. This generation of computers was
primarily built for general business use.
The third generation, (1965-1977), was characterized by use of integrated circuits,
sophisticated operating systems, magnetic disk storage and the introduction of a pleth-
ora of general and specific programming languages. Though "old" from a technological
No. 41 COPYRIGHT AND ARTIFICIAL INTELLIGENCE 717

In a brief span of 30 years, computer technology has com-

pleted nearly four generations of development. 48 Conse-
quently, machine hardware capabilities have far outstripped
the availability of user software, creating excess demand for
applications programs used in business and entertainment.4 9
In an attempt to fill this vacuum, the software industry has

viewpoint, many of these machines are still in use. They are very fast in calculation,
low in cost, small in size and extremely reliable. This generation has formed the data
processing infrastructure which currently supports U.S. business, education, govern-
ment and defense. See S. MANDELL, supra note 45, at 19-39. See generally W. RODGERS,
THINK: A BIOGRAPHY OF THE WATSONS AND IBM (1974).
The fourth generation extends from the late 1970's into the present. This generation
of computer technology is characterized by very large scale integration (VLSI), in-
creased data storage capacity, a marked reduction in price and an increase in reliabil-
ity and applications possibilities. This technology is used in very powerful and fast
computers like the Cray I, which is capable of close to 100 million calculations per
second. The fourth generation has also spawned the "personal" computer. Based on
microprocessor technology, these machines are small and inexpensive, and are prolif-
erating rapidly into the world of the small businessman, professional, scientist, educa-
tor and hobbyist. See Toong and Gupta, PersonalComputers, 247 Sci. AM. 87 (Dec.,
1982), and Matisoo, The Superconducting Computer, Sci. Am. May 1980, at 50, 65. See
generally Sc. AM., MICROELECTONICS (1977) (collection of 11 articles published in Sci-
entific American covering development, technology and applications of microelectron-
ics) and Hogan, From Zero to a Billion in Five Years, INFOWORLD, Aug. 31, 1981, at 6.
The Japanese feel we're well on our way towards a fifth generation of computer tech-
nology. "The previous four generations of computers ... were categorized according
to their elements: the first used the vacuum tube, the second the transistor, the third
the integrated circuit, or semiconductor, and the fourth the very large scale integrated
circuit, or VLSI.... [The Japanese] expect the fifth generation to use, in addition to
VLSIs, circuits that work on different physical principles from today's semiconductors.
These powerful new circuits, however, are far from the only way in which fifth genera-
tion computers will be revolutionary." Lehner, Japan Starting 10-Year Effort to Create
Exotic Computer, Wall St. J., Sept. 25, 1981, at 29, col. 1. See also Uttal, Here Comes
Computer Inc., FORTUNE, Oct. 4, 1982, at 82.
48. See note 47, supra.
49. See Toong and Gupta, supra note 47, at 100-104. The 42 members of the Com-
puter and Business Equipment Manufacturer's Association in 1976 predicted an ex-
penditure of 17 billion dollars on software alone during 1976-79. Gemgnani, Legal
Protectionfor Computer Software: The View from '79, 7 RUTGERS J. COMPUTERS TECH.
& L. 269, 274 (1979). An estimated seven million Americans spent their work days in
front of video display terminals in 1981. Markoff &Freiberger, VDT's Can Cause Stress
and Other Health Hazards in the Office, INFOWORLD, Oct. 26, 1981, at 21. "When elec-
tronic computers became commercial, just after World War II, forecasters predicted
that no more than a dozen machines ever would be needed. IBM even decided not to
sell computers because the market seemed so small. Yet this year, for the first time,
computers on the planet will outnumber people. According to market researchers
Dataquest, Inc. and International Data Corp., by the end of 1982, more than five billion
computers of all sizes, from microprocessors to mainframes, will be in use." Shaffer,
Computing Industry is Finding That It's Vulnerable to Slump, Wall St. J., April 16,
1982, at 27, col. 1. "So many tiny computers are working under the hood of today's
automobile to increase mileage and lower pollution that a single car maker, General
Motors, actually manufactures more computers every year, albeit smaller ones, than
 COMM/ENT [Vol. 4

evolved rapidly. Though a relatively esoteric profession in the

late 1960's, over 300,000 programmers presently ply their trade
within a large and complex computer manufacturing, data
processing, "information" industry.5"
The rapid growth rate and economic significance of the com-
puter industry are important when discussing legal issues in-
volving a "leading edge" area of research and development like
artificial intelligence because what is presently considered fu-
turistic may become commonplace as the next generation of
computer hardware and software evolves and disseminates.
Equally important is at least a cursory understanding of the
concept of a computer "program."

B. Computer Programs
A modern digital computer functions as a vast array of elec-
trical switches, whose values ("on" or "off") can be manipu-
lated to represent and modify information. 5 This
manipulation of information is controlled within the machine
by a set of instructions known as a "program. "52
The "program" most people are familiar with is the actual
"coded" instruction set entered at the computer keyboard (or
on punched cards or paper tape, etc.) .53 The program is a rep-4
5
resentation of an algorithm prepared initially as a flow chart
or step-wise procedure 5 written in standard English
5 6
to solve a
particular problem or to perform a specific task.

International Business Machines Corp., Digital Equipment Corp., Burroughs Corp.

and all other computer-system manufacturers combined." Id.
50. See Gemgnani, supra note 49, at 273-276.
51. In large scale integration, a single chip can hold approximately 64,000 bytes. A
byte of information in most microprocessors consists of eight bits capable of either "1"
or "0" values. For background information concerning the components of modern digi-
tal computers, see generally S. MANDELL, supra note 45 and H. KATZAN, OPERATING
SYSTEMS, A PRAGMATIC APPROACH 12-64 (1973).
52. See note 56, supra.
53. Two kinds of programs should be distinguished. First, system programs such
as the operating system, input-output routines and user-interface programs allow the
user to communicate with the machine while executing a program. Second, user pro-
grams are entered into the computer by the user to perform a given user-defined task.
This note focuses on user programs.
54. A "flow-chart" is a graphic representation of the operation and flow of informa-
tion through a computer program. Special symbols are used to represent specific oper-
ations (input, output, memory storage, etc.). See C. MEEK, supra note 20, at 94.
55. A "step-wise" procedure is the English-language version of an algorithm to be
later translated into a specific programming language.
56. 17 U.S.C. app. § 101 (1976) states that "A 'computer program' is a set of state-
ments or instructions to be used directly or indirectly in a computer in order to bring
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 719

From initial formulation of an algorithm to final execution,

the program changes dramatically in form and structure. On
the outermost level of computer programming and operation,
the program is entered into the computer via the keyboard or
some other input device. Here, the program is represented
within the computer as a direct translation of the characters
entered.57
Next, the program must be compiled or
translated into
machine readable form. This is done by the "compiler." The
compiler is a sophisticated set of programs which checks the
5 8
user program for logical, syntactical and typographical errors.
These errors are flagged on an output listing for the user to cor-
rect. A compiler performs a rigid mapping function; a given
element of code is translated into a corresponding element of
machine language.59 A programmer writes the coded instruc-
lions; a compiler translates them. Thus, in the normal pro-
gramming environment, the programmer can exercise a large
degree of originality and creativity in formulation of the al-
gorithm and in the expression of this idea in a coded program
instruction set.

about a certain result." See Gemgnani, supra note 49, at 280-281, for further discussion
of definitions of "computer programs." See also Nycum, Legal Protectionfor Computer
Programs, I COMPUTER/LAw J. 1, 11-12 (1978).
57. See generally S. MANDELL, supra note 45, at 53-54.
58. "Compilers are necessarily dependent upon the language being processed;
however, the following steps are usually involved:
1. The source program is read and analyzed on a statement by statement
basis.
2. A lexical analysis routine scans each source statement and identifies re-
served words, variables, operator symbols, constants, etc.
3. A syntactical analysis routine identifies the type of statements and veri-
fies that the structure of that statement is admissible.
4. Tables and lists of symbols, expressions, and statements are maintained so
that an inter-statement analysis can be made.
5. Analysis is made of logical flow of the source program and a global error
analysis is made.
6. Machine language instructions, in an intermediate symbolic form internal
to the compiler are generated and optimization is performed, as required.
7. An object module is generated from the intermediate language and a pro-
gram listing is produced.
H. KATZAN, supra note 51, at 22-23. After the compilation state, the output module
contains not only the translated code, it also contains various system routines, or in-
structions invoking these routines to assist in execution and uses of system resources.
Some compilers "optimize" the module by modifying it in even more drastic fashion to
allow for more efficient execution. See id. at 24-36 for an introduction to loading, link-
ing and editing.
59. But this correspondence is not necessarily in a 1:1 ratio.
 COMM/ENT [Vol. 4

However, AI developments like code generators, 60 ALICE 6

and PECOS62 threaten to alter this normal method of program
creation. Code generators remove the programmer from the
authorship process to the extent the machine self-generates
the high level source code, and in some cases supplies the solu-
tion algorithm 63 as well. Programs like ALICE and PECOS re-
quire only a problem statement. The computer then develops
a solution algorithm, creates a code listing, compiles it and exe-
cutes the program. The element of human input into the pro-
gram-creating process is minimal. Furthermore, if the product
of the program, as in RACTER, is designed to simulate a
human-authored manuscript, and the output is a priori un-
knowable to the operator who executes the program, then ar-
guably neither the programmer nor the operator is the author
of the finished product.

IV
Artificial Intelligence and Federal
Copyright Law: 1982
The computer industry has sought to protect its ideas, tech-
niques, aesigns and products from misappropriation, infringe-
ment and unauthorized use. The industry has traditionally
relied upon the use of trademarks, copyrights, trade secrets,
patents and licensing agreements to protect its property inter-
ests. 64 However, the recent growth of the market in video-
games, home or "personal" computers and small business sys-
tems has created problems in protecting and enforcing legal in-
terests in computer software.65 Low cost, ease of copying,
60. See note 24, supra.
61. See note 30, supra.
62. See note 35, supra.
63. See note 32, supra.
64. See generally Nycum, supra note 56, at 1-81; Gemgnani, supra note 49, at 269-
313; Comment, Copyright Protectionfor Computer Programs, 47 TENN. L. REV. 787
(1980); Schmidt, Legal ProprietaryInterests in ComputerPrograms:The American Ex-
perience, 21 JURIMETRICS J. 345 (1981).
65. 'The case of [MICROCHESS] shows how severely amateur copying can dam-
age software sales. Before the International PET Users' Group published a method of
copying [MICROCHESS], the game program had sold more than 100,000 copies. After
publication of the copy method, sales dried up. By contrast, the semiprofessional pro-
gram [WORDCRA-T] enjoyed a dramatic increase in sales when the protection rou-
tine known as the 'Dongle' was incorporated." Hayman, Software Protection in the
United Kingdom, BYTE, Oct. 1981, at 132. See generally Freiberger, Software Piracy,
INFOWORLD, Mar. 22, 1982, at 31.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 721

portability and often immense commercial value and popular-

ity all serve to motivate the illicit copying, use and sale of com-
puter software. Searching for a viable means of protection,
software suppliers have now turned to licensing agreements,6 6
sale of "read only" non-copyable programs 67 and copyright re-
gistration.6 8 Licensing agreements usually rely on self-submit-
ted forms included in a software package at the point of sale.6 9
To the would-be pirate and the hobbyist, such agreements are
seen as ridiculously unenforceable.7 0 "Copy-protected" pro-
grams have caused such concern among users in need of back-
up copies that a thriving market has developed for programs
specifically designed to copy any program stored on any given
disk.7 1 Because of such problems with copy-protection and li-
censing, and because of the large amounts of money involved,
companies have increasingly sought protection under the fed-
eral copyright laws.72

66. See generally D. REMER, LEGAL CARE FOR YOUR SoFrWARE (1982).
67. See, e.g., Teiser, Locksmith Version 4.0from Omega Micro-Ware, INFOWORLD,
Feb. 1, 1982, at 22 (review of a popular program used to defeat copy-protection
schemes). One commentator highlighted the problems of copy-protection:
"[A] mateurs confront software publishers with a dilemma: if publishers take no steps
to protect their programs, making a copy becomes the easiest thing in the world. On
the other hand, if publishers use protection routines, making a copy is for many ama-
teurs the most enjoyable thing in the world. Unlike semiprofessional users of
software, amateurs have both the time and the enthusiasm needed to defeat protective
measures." Hayman, supra note 65, at 132. Peter Laurie, editor of PracticalComput-
ing, confirmed this view by saying "Any intelligent tpenager will make it [overcoming
copy-protection measures] his first task of the day." Id. See generally Freiberger,
Software Piracy, INFOWORLD, Mar. 22, 1982, at 31.
68. See text section IV and accompanying notes, infra.
69. A program is usually sold imbedded on a "floppy disk" (a magnetic recording
medium about the size of a 45 rpm record) enclosed within some form of packaging.
The licensing agreement, warranty, etc. is often sealed within the package and is un-
available for review prior to purchase of the program. The user is supposed to fill out
the enclosed forms, mail them to the manufacturer and then abide by the terms stated
therein. See generally D. REMER, supra note 66.
70. Id.
71. A single-disk user is in great need of a backup copy, especially where his data
and program are still on the same disk. Some users take this privilege too far, causing
problems for all users. See generally Freiberger, Software Piracy, INFOWORLD, Mar. 22,
1982, at 31. One commentator has said, "amateur piracy will have five consequences
for the average software buyer. It will reduce the range of software available, raise
prices, and make companies reluctant to invest in software development. ...
[Pliracy also leads to lack of support and maintenance, and discourages development
of software by cottage industries which cannot afford to go to court to protect their
interests." Hayman, supra note 65, at 133.
72. "[Djramatic change in the law and the growing trend toward mass-marketed
programs mean that copyright is likely to be increasingly important in protecting com-
 COMM/ENT (Vol. 4

Statutory federal copyright protection for computer pro-

grams is relatively new. In its final report issued July 31, 1978,
the National Commission on New Technological Uses of Copy-
righted Works (CONTU) recognized that "[clomputer pro-
grams are the product of great intellectual effort and their
utility is unquestionable. '73 To afford protection "necessary to
encourage the creation and broad distribution of computer pro-
grams in a competitive market, ' 74 CONTU recommended that
the Copyright Act of 197671 [hereinafter the Act] be amended:
(1) to make it explicit that computer programs, to the extent
that they embody an author's original creation, are proper sub-
ject matter of copyright; (2) to apply to all computer uses of
copyrighted programs by the deletion of the present section
117; and (3) to ensure that rightful possessors of copies of com-
puter programs may use or adapt these copies for their use.7 6
These recommendations were incorporated in an amend-
ment to the Act enacted under the Carter administration on
July 12, 1980.7 The amended Act expressly placed computer
programs within the realm of literature and written works. It
extended protection to programs that are the original expres-
sion of an idea, by an author, fixed in a tangible medium of
expression capable of being directly or indirectly viewed with
or without the aid of a device.7 8 A systematic enumeration of
the provisions of the Act, relevant case law and their impact
upon AI software developments like code generators, ALICE,
PECOS and RACTER delimits the present boundaries of fed-

puter programs, particularly those of small entrepreneurs who create their works for
individual consumers and who can neither afford nor properly use other forms of pro-
tection." FINAL REPORT OF THE NATIONAL COMMISSION ON NEW TECHNOLOGICAL USES
OF COPYRIGHTED WORKS, July 31, 1978, at 15. [hereinafter CONTU]. See also Remer,
Legal expert on software theft: the piranahasversus true pirates, INFOWORLD, Mar. 22,
1982, at 40; Lawlor, A Proposalfor Strong Protectionfor Computer ProgramsUnder the
Copyright Law, 20 JURIMETRICS J. 18 (1979) (broad support within business commu-
nity for copyright protection for computer software); Freiberger, Sony Case Scares
Micro Makers, INFOWORLD, Nov. 16, 1981, at 1 (discusses Universal Studios v. Sony
Corp. of America, 659 F.2d 963 (9th Cir. 1981) as it affects possible copyright infringe-
ment by makers of anti-copy protection devices specifically designed and marketed to
defeat copyright owner's attempts to protect source code from unauthorized copying).
73. CONTU, supra note 72, at 11.
74. Id.
75. 17 U.S.C. app. §§ 101-810 (1976 & Supp. IV 1980).
76. CONTU, supra note 72, at 1.
77. Boorstyn & Fliesler, Copyrights, Computers, and Confusion, 56 CAL. ST. B.J.,
148 (1981).
78. Id. See also Wehringer, Copyright in Brief, PRAC. LAw., July 15, 1979, at 77
(brief discussion of filing procedures for copyright protection).
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 723

eral copyright laws applicable to the products of Type I and

Type II AI software.

A. Requirements for Copyright Protection

1. Computer Programsare "Literature,Written Works"
Under the Constitution, Congress is empowered to "promote
the Progress of Science and useful Arts, by securing for Times,
to Authors ... the exclusive Right to their respective Writings
.. The term "writings" in the Constitution has been
broadly interpreted by Congress and the courts to protect new
forms of expression created by technological advances. In
1884, the Supreme Court recognized photographs as writings.8 0
Motion pictures were recognized as writings in 1911,81 sound
recordings in 1973.82 Although the U.S. Copyright Office began
to register computer programs in 1964,83 the express subject
matter provision under the "writings" interpretation was not
provided until the 1980 amendment to the Act.84
By placing computer programs and data bases under the
term "literary works," the Act does not "connote any criterion
of literary merit or qualitative value; it includes catalogs, direc-
tories, and similar factual, reference, or instructional works
and compilations of data. It also includes computer data bases
and computer programs .... Thus, within the provisions of
the Act, products of Type I AI software are clearly copyright-
able subject matter as computer programs or data bases under
the rubric of "literature or written works." Products of Type II
AI programs, like stories written by RACTER, also fit within
the "literature" subject matter category.

2. The ProgramMust Be Fixed in a Tangible Medium of

Expression
Although computer programs are copyrightable subject mat-
ter, they are not copyrightable until they are "fixed." The Act
as amended in 1980 provides that copyright protection is avail-
able to original works of authorship "fixed in any tangible me-

79. U.S. CONST. art. I, § 8, cl. 8.

80. Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884).
81. Kalem Co. v. Harper Bros., 222 U.S. 55 (1911).
82. Goldstein v. California, 412 U.S. 546 (1973).
83. Boorstyn, supra note 77, at 148.
84. N. BOORSTYN, COPYRIGHT LAW. 68 (1981).
85. CONTU, supra note 72, at 16.
 COMM/ENT [Vol. 4

dium of expression, now known or later developed, from which

they can be perceived, reproduced, or otherwise communi-
cated, either directly or with the aid of a machine or device. '"86
Fixation in a tangible medium of expression occurs when
"its embodiment in a copy or phonorecord, by or under the au-
thority of the author, is sufficiently permanent or stable to per-
mit it to be perceived, reproduced, or otherwise communicated
for a period of more than transitory duration."8 7 In ruling for
the copyrightability of a program stored in Read Only Memory
[hereinafter ROM], the courts have found that the "Act con-
tains no restrictions on the type of material objects suitable for
fixation."8 8 Thus, computer "chips"8 9 have been considered
copies of copyrighted programs.9 0 Programs stored on other
media, capable of being perceived, reproduced, or otherwise
communicated by or from the copy directly or indirectly are
also "fixed" within the meaning of the Act. 91
Technology employed in the fixation of AI software is no dif-
ferent from that used for any other computer program.2 AI-
produced software can be stored in RAM,9 3 ROM, "chips,"
tapes, cards, etc. It can be "perceived, reproduced, or other-
wise communicated" in the same way as ordinary software. 94
Thus, programs produced by Type I AI programs meet this re-
quirement to the same extent as would a program produced in
a normal manner.95 Computer written stories, (products of

86. 17 U.S.C. app. § 102(a) (1976).

87. 17 U.S.C. app. § 101 (1976 & Supp. IV 1982).
88. Midway Mfg. Co. v. Dirkschneider, 214 U.S.P.Q. (BNA) 417, 427 (D. Neb. 1981).
In microprocessor computers, the primary memory is usually "hardwired" into the
system. In these computers, the Read Only Memory (ROM) cannot be modified by the
user and generally contains the main system programs. The Random Access Memory
(RAM) portion is available for use by the programmer and is where the user program
is loaded and executed. The contents of ROM are permanently hardwired into the
computer in much the same way as an electronic circuit. Thus, ROM contents are not
destroyed when electrical power to the machine is shut off. The RAM is volitile and is
only stable while power is supplied; the contents are destroyed when electrical power
ceases.
89. See generally Sc. Am., MICROELECTRmOCS, supra note 47.
90. Tandy Corp. v. Personal Micro Computers, Inc., 524 F. Supp. 171 (N.D. Cal.
1981).
91. Id.
92. Like any other computer software, Al code is amenable to fixation and storage
in all normal media (e.g., tape, disk, cards, RAM, ROM).
93. See note 88, supra.
94. 17 U.S.C. app. § 102(a) (1976).
95. Midway Mfg. Co. v. Dirkschneider,214 U.S.P.Q. (B.N.A.) 417,427 (D. Neb. 1981),
held that audiovisual displays are entitled to copyright protection even though the un-
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 725

Type II AI programs), would also be deemed "fixed" either

within the memory of the computer (stored as a data base), in
video-display form or in tangible paper printout.

3. Registration and Notice Requirements

Under the Act as amended in 1980, the "concept of eye-reada-
bility, for the purpose of copyright eligibility, has been aban-
doned. '9 6 The Act only requires registration of program copies
which can be visually perceived directly or indirectly with or
without the aid of a device.9 7 AI-produced software (products
of Type I AI programs) would meet this requirement to the
same extent as ordinary software because its manner of fixa-
tion, storage and display are not unique and can be accom-
plished "with or without" the aid of a device. Type II AI
products would meet the notice requirements in the same
manner as ordinary literary works stored, displayed or printed
by a computer.
CONTU has stated that if the program is copyrighted, then
the notice should be displayed prominently on any source list-
ing. "Those programs that may be read only with the aid of a
machine or device should contain notice in the medium of fixa-

derlying computer program was not copyrighted. But see In re Certain Coin-Operated
Audio-Visual Games and Components Thereof, 537 PAT. TRADEMARK & COPYRIGHT J.
(BNA) No. 337-TA-87, at A-5 (July 16, 1981) where the International Trade Commission
distinguished between the "attract mode" and the "play mode" when assessing
copyrightability of commercial audiovisual displays. Midway had asked the ITC to ex-
clude importation of video games appearing to infringe on copyrights it held on its
Galaxian video game. The Commission held the attract mode was copyrightable be-
cause it always replayed the same identical sequence. Similarly, it held the first few
seconds of the play mode also copyrightable, because this sequence is also repeated
identically in every game. However, the Commission refused to decide whether per-
formance of the game could possibly infringe upon a valid copyright of the play mode:
First, for statistical reasons, it is virtually impossible for a performance of
Galaxian ever to duplicate that performance fixed in the video tape. If we
were to hold that such performances could infringe a copyright in the play
mode, we might be protecting the game itself or its mode of play, items which
are specifically not subject to copyright protection. Second, each performance
of the Galaxian play mode depends, in part, on the player. It is therefore pos-
sible that the player may be considered a "coauthor" of each performance of
the play mode. Our research has indicated no legislative history or case law
on whether coauthored works of this sort are subject of copyright, and we de-
cline to rule on this issue. Third, in view of the remedy we are granting in this
investigation, a ruling on either copyrightability or infringement of the play
mode is unnecessary.
Id. at A-6.
96. Boorstyn, supra note 77, at 148.
97. 17 U.S.C. app. § 102(a) (1976).
 COMM/ENT [Vol. 4

tion so that the contents of the program cannot be listed with-

out reproducing the notice in the position just described."98
Indeed, in Midway Manufacturing v. Dirkschneider,a federal
district court applied this concept to audiovisual displays pro-
duced by a computer program, even though the underlying
program was not copyrighted.99 Finding the audiovisual output
was itself copyrightable subject matter, the court discussed no-
tice requirements of the Act, stating that:
[t]he [Aict apparently requires fixation on the copies of the
works, the proposed regulations for this section provide that
the copyright notice is satisfied, for works embodied in printed
circuit boards, by including the notice within the visual display
of the work or by affixing the notice to the terminal where the
work is displayed. 00
Thus, copies of computer programs or their audiovisual prod-
ucts must contain notice of copyright protection so that if the
copy or product becomes "visually perceived," then the notice
becomes conspicuous. Like other computer programs, prod-
ucts of AI software can easily be tailored to meet this need.

4. The Expression, Not the Idea Itself, Is Protected

Under the provisions of the Act, the expression of the idea,
not the idea itself, is the object of protection.' 0 ' This rule is
consistent with case law. In 1879, the Supreme Court in Baker
v. Selden drew a "clear distinction" between Mr. Selden's book
and the "art which it is intended to illustrate."'' 0 2 Selden had
sued Baker for alleged copyright infringement on a book of
blank accounting forms with accompanying text explaining a
particular system of bookkeeping. 10 3 The court held that copy-
right would extend to the particular explanation or expression
of the bookkeeping method, but would not grant "to the author
an exclusive right to the methods ... he propounds, or to the
diagrams which he employs to explain them."'1 4 Similarly, in
Mazer v. Stein the Supreme Court found "protection is0 5given
only to the expression of the idea-not the idea itself.'

98. CONTU, supra note 72, at 14.

99. Midway Mfg. Co. v..Dirkschneider, 214 U.S.P.Q. (BNA) at 428.
100. Id. at 428.
101. 17 U.S.C. app. § 102(b) (1976).
102. Baker v. Selden, 101 U.S. 99, 102 (1879).
103. Id. at 100.
104. Id. at 103.
105. Mazer v. Stein, 347 U.S. 201, 217 (1954).
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE

However, when the number of ways to express an idea is lim-

ited, courts generally will not allow copyright of the expres-
sion, because this would lead to a virtual monopoly on the
underlying idea. Although this rationale has been employed in
a recent case to disallow copyright protection to data input for-
mats for structural analysis software, 10 6 once courts become
aware of the vast choice available to the programmer in coding
a given algorithm, this argument will no longer be persuasive.
As one commentator stated, "[o] nce the ability to vary the la-
beling of variables and the sequence of commands is widely
recognized, computer programs should not fall prey to the doc-
trine that infringement does not occur when an idea can be ex-
pressed in only a limited fashion."'1 7
The Act clearly intends to continue the historical division be-
tween an idea and its expression. CONTU manifests this in-
tention in its interpretation of section 102(b) of the Act.
CONTU stated that section 102(b) "is intended, among other
things, to make clear that the expression adopted by the
programmer is the copyrightable element in a computer pro-
gram, and that the actual processes or methods embodied in
the program are not within the scope of the copyright law." 108
Satisfying this idea/expression dichotomy required under
the Act presents a major obstacle to copyright protection for
products of both Type I and Type H AI software. In copyright,
the "expression" to be protected necessarily presumes the
existence of an underlying idea to be expressed. 10 9 As previ-
ously noted, the trend in code generation, automatic program-
ming and automatic storywriting is towards a minimization of
human input into the production of an apparently creative re-
sult. Although the resulting program may embody an intricate,
unique algorithm or the story a seemingly coherent plot, these
"expressions" produced by AI software cannot be said to be
expressions of ideas "adopted by the programmer" as intended
by the Act." 0 The computer software, not the programmer, se-
lectively "adopts" the particular form of expression evidenced

106. Synercom Technology, Inc. v. University Computing Co., 199 U.S.P.Q. (BNA)
537 (N.D. Tex. 1978).
107. Comment, Copyright Protectionfor Computer Programs, 47 TENN. L. REV. 787,
788 (1980).
108. CONTU, supra note 72, at 19 (emphasis omitted).
109. Id. See also notes 114, 115 and accompanying text, infra.
110. CONTU, supra note 72, at 19.
 COMM/ENT [Vol. 4

in its product."1 To the extent the programmer has less and

less control over the expression ultimately "adopted," the
2
programmer loses possible copyright protection. 1 If the ex-
pression is no longer the programmer's, is it the computer
113
software's?
As courts attempt to investigate the copyright potential of AI
products, the goal of affording protection to the expression of
ideas, as set forth in the case law and in the 1976 Act, will be
very difficult to achieve because it will be a herculean task to
determine if the expression is one of an underlying idea (and
thus of human origin and therefore possibly copyrightable) or
is only "apparently" an expression of an idea (the existence of
which is anthropomorphically read into the expression by the
viewer) and thus not within the contemplation of copyright
law." 4' If courts strictly maintain the idea/expression dichot-
omy as a central pillar of copyright law, AI produced "expres-
sions" will lead into a metaphysical quagmire as things get
"curiouser and curiouser" per poor Alice."'

5. The Program Must Be an Author's Original Work

The "author" of a work seeking copyright protection is "he to

111. The programmer who uses the software has no control over the exact code or
story that is produced by the Al software. The programmer who created the AI
software itself might have an idea of what the program will do when faced with a given
set of data, but, as in RACTER, if the program randomly selects nouns and verbs then
places them in grammatically correct relationships consistent with the rules given it
by the original programmer, even he would be unable to foresee or influence the con-
tent of the story output.
112. The federal copyright laws require a work to be that of an author and that it
evidence a non-trivial amount of "intellectual labor." See notes 116-118 and accompa-
nying text, infra.
113. See generally Milde, supra note 8.
114. See notes 116-118, 127 and accompanying text, infra. See also Apple Computer,
Inc. v. Franklin Computer Corp., 545 F. Supp. 812 (E.D. Penn. 1982) (slip opinion, July
30, 1982) (in denying copyrightability of a series of programs stored in ROM on a chip,
the court discussed the authorship requirements of copyright law). But see Williams
Electronics, Inc. v. Artic International, Inc., 685 F.2d 870 (3rd Cir. 1982) (slip opinion,
August 2, 1982) (on similar facts, court held programs stored in ROM to be works of
authorship, fixed in tangible medium of expression, etc., meeting all requirements of
copyright law). See also Midway Mfg. Co. v. Artic International, - F. Supp. - (E.D. Ill.
1982) (slip opinion, March 10, 1982) (programs stored in ROM copyrightable).
115. Apple Computer, Inc. v. Franklin Computer Corp., 545 F. Supp. at 825. "If the
concept of 'language' means anything, it means an ability to create human interaction.
It is the fixed expression of this that the copyright law protects, and only this. To go
beyond the bounds of this protection would be ultimately to provide copyright protec-
tion to the programs created by a computer to run other computers. With that, we step
into the world of Gulliver. .. "
No. 41 COPYRIGHT AND ARTIFICIAL INTELLIGENCE 729

whom anything owes its origin; originator; maker; one who

completes a work of science or literature."1'16 The art of author-
ship requires some independent reproduction or independent
contrivance. 117 Thus, the "author" of a program is the one who
originates it.
The work must be original. The test of copyright originality
is "one with low threshold, in that [a]ll that is needed ... is
that 'author' contributed something more than 'merely trivial'
variation, something recognizably 'his own.' "1 8 Thus, the non-
trivial alteration of a prior work will "constitute sufficient origi-
nality to support a copyright if such variation ... is the prod-
uct of the author's independent efforts .... ."119 Though the
requirement of "substantial as opposed to trivial variations"
from the original is met with a low threshold, courts continue
to interpret copyright protection
120
as requiring some degree of
"originality" in authorship.
While investigating the status of computer technology in
forming its recommendations, CONTU found the state of artifi-
cial intelligence did not yet pose problems relating to copyright
requirements of original authorship.' 2' The report noted that:
[t]his discussion may have stemmed from a concern that com-
puters either had or were likely to soon achieve powers that
would enable them independently to create works that, al-
though similar to other copyrightable works, would not or
should not be copyrightable because they had no human au-
thor. The development of this capacity for "artificial intelli-
gence" has not yet come to pass, and, indeed, it has been
suggested that ... 1 22such development is too speculative to con-
sider at this time.
What the commission found "too speculative" is rapidly be-
coming commonplace only five years later. 23 Code generators

116. Burrow-Giles Lithographic Co. v. Sarony, 11 U.S. 53, 57-58 (1884).

117. Alfred Bell & Co. v. Catalda Fine Arts, Inc., 191 F.2d 99, 103 (2d Cir. 1951).
118. L. Batlin & Sons, Inc. v. Snyder, 536 F.2d 486, 487 (2d Cir. 1976) (quoting Alfred
Bell & Co. v. Catalda Fine Arts, Inc., 191 F.2d 99, 103 (2d Cir. 1951)).
119. Id. at 492 (Meskill, J., dissenting) (quoting 1 M. NIMMER, NIMMER ON COPYRIGHT
§ 10.1).
120. Id. at 490.
121. CONTU, supra note 72, at 44.
122. Id. at 42.
123. CONTU's apparent lack of insight regarding the direction and pace of com-
puter industry development can be understood when one notes that: (1) CONTU re-
lied upon testimony of experts in computer science as the basis for its conclusions that
AI was too primitive to present copyright issues; (2) most of the data used by the com-
mission was from the mid-1970's, thus not focusing on the revolutionary impact of
 COMM/ENT [Vol. 4

are being mass-marketed for home computers, 2 4 industry is

clamoring for devices using automatic programming tech-
niques125 and a computer-generated story has been published
in a national magazine.'26 , As AI software and its products be-
come increasingly sophisticated, courts will soon be forced to
grapple with the unstated assumption underlying the copy-
right concepts of authorship and originality: an author is a
human. 27
In the vast majority of programming situations the legal re-
quirements of human authorship can be easily satisfied. The
programmer expresses his solution algorithm in the form of a
coded instruction set or a high-level language. This program is
then compiled a'ad executed, causing the computer to manipu-
late data to produce the results desired. The results can all be
found to originate from, or be copies or translations of, the
work of a human author within the meaning of the Act as
amended in 1980.128
In primitive code generation systems (Type I AI programs)
the level of participation of the programmers is still quite
high.'29 By interactively answering questions posed by the
microprocessor technology; and (3) it was probably in the best interest of the commis-
sion to follow the conservative lead of its experts to avoid political repercussions and
needless controversy. See generally CONTU, supra note 72.
124. See Jacobs, supra note 24, at 352. The 1981 and 1982 issues of BYTE, INFOWORLD,
PERSONAL COMPUTING and other computer-oriented periodicals all carry advertise-
ments for code generators to ease the task of program development.
125. Industrial robots, Computer Alded Design and Manufacturing (CAD/CAM)
and medical diagnostic programs all use Al principles.
126. RACTER, supra note 1, at 96.
127. Under the U.S. CONST. art. I, § 8, cl. 8, Congress is authorized to legislate for the
protection of authors whose writings are the result of intellectual labor. The copyright
is intended to protect original writing founded in the "creative powers of the mind. . .
fruits of intellectual labor .... ", the Trade-mark Cases, 103 U.S. 82, 94 (1879). The
original writing must evidence the skill, labor and judgment of the author. Dorsey v.
Old Surety Life Ins. Co., 98 F.2d 872, 873 (10th Cir. 1938). However, even though the
author need only contribute "something more than a 'merely trivial variation' some-
thing recognizably his own," L. Batlin & Son, Inc. v. Snyder, 536 F.2d 486, 487 (2d Cir.
1976) (quoting Alfred Bell, 191 F.2d at 103), this low threshold of originality "does not
prevent the court from concluding ... that works may lack even the modicum of origi-
nality required for a valid copyright. . . ." Gardenia Flowers, Inc. v. Joseph Marko-
vits, Inc., 280 F. Supp. 776, 782 (S.D.N.Y. 1968). Thus, Congress has allowed copyright
protection for writing resulting from intellectual labor. The courts have provided an
interpretation requiring a minimal amount of original contribution by the author. By
implication, the use of "him" and "his," as well as the concepts of "intellectual labor"
and "creative powers of the mind" all rest on the assumption that the "author" of the
work at issue is a human. See Apple Computer, Inc., 543 F. Supp. at 824-825.
128. 17 U.S.C. app. §§ 101-810 (1976 & Supp. IV 1982).
129. See note 24, supra.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE

machine, the user formulates and expresses a vague algorithm

and specifically decides output format. Although not involved
13 1
130 and functions
in the specific coding of the parameters
needed to express this vague algorithm or the desired output
form, the variable names,'1 32 field labels 33 and array dimen-
34
sions1 provided by the programmer will tailor the resultant
coding into a unique, copyrightable program. But, as code gen-
erators become more specialized in given applications, the
amount of user input will probably decrease. 35 At some point,
courts will be confronted by extremely sophisticated programs
produced by code generators aided by very minimal human in-
put, forcing them to determine if the human effort involved is
"non-trivial" or "substantial" in the copyright sense.
Problem-oriented language systems like ALICE present
even more difficult human/machine threshold questions be-
cause the user merely states the problem in a descriptive form
for the computer to solve. The user does not formulate the al-
gorithm or influence its expression in coded form. He is con-
cerned with the results, not the process developed by the
computer to obtain them. Although the problem-specifier
would probably be able to copyright the problem statement, he
would have a difficult time convincing a court of the
copyrightability of the code created by the computer and its
software.1 36 The author originates the expression of the idea;

130. A parameter is a piece of information such as date or time, which functions as

a constant, boundary or reference value for internal use within the program.
131. E.g., mathematical equations and logical relationships (less than, greater than,
etc.).
132. A variable name is a symbol which represents a certain value which can be
operated upon by the program (e.g., in the equation a = 4, "a" is a variable).
133. A field label is a name given part of a record of information. For example, in an
"address record" field labels might be assigned for those portions which comprise the
street number, street name, city, state and zipcode.
134. An array can have one or more dimensions. A one-dimensional array can be
expressed as a linear arrangement of data. A two-dimensional array has an x-axis and
a y-axis. A three-dimensional array has an x-axis, y-axis and z-axis. If one were to
write out a multiplication table through the 10's, then he would have a 10xl0 two-di-
mension array with the values 10,10 representing its maximum size.
135. This conclusion follows logically from an assumption that commercial dissemi-
nation of code generators reflects labor saving economies created by their use.
136. Ascertaining the "flow" of originality or idea expression and creation through
the process of problem specification, algorithm development, program coding, compila-
tion and execution is best illustrated by a series of examples of increasing technical
complexity. Consider first the author who uses a computer as a word-processor. The
author formulates the plot, characters and dialogue while at the terminal or types in a
previously written rough draft of his manuscript. The characters input at the keyboard
 COMM/ENT [Vol. 4

this expression, not the idea itself, is afforded copyright protec-

tion.137 If one grants authorship and copyright protection to
programs written by programs like ALICE, which arguably fall
below the "non-trivial" test of copyright authorship and origi-
nality, then one might conclude either: (1) computers can
have ideas, because a computer program is generally recog-
nized as an expression of an idea-algorithm or solution
method; or (2) "expressions" can be copyrighted
1 38
even though
they do not embody an underlying idea.
Finally, Type II AI programs accurately simulating human
behavior, like story-writing programs (e.g., RACTER), also
present problems regarding authorship and originality require-
ments of the Act. As noted earlier, the program outputs a story
outwardly meeting all criteria needed for copyright, yet the

are represented within the computer as their electronic equivalents. The author then
executes certain word-processor editing programs to manipulate this data as desired
and then directs the computer to print the final product. Here, the ideas concerning
the story have originated in the mind of the author. This set of ideas is expressed
through the display on the video screen, within the computer memory and in the final
output version of the story. The word-processor has functioned merely as a tool to
help the author express his ideas. Next, consider a participant in a sophisticated inter-
active fantasy game played on a computer modified to print out a narrative summary of
the game as it is being played. Here, the player/author initiates the game by starting
the program. His idea is to play the game. He may have ideas regarding strategy and
select the level of complexity at which he desires to play, but he will not be able to
determine the specifics of his "plot" with any degree of certainty. As the player selects
the general traits of his characters (certain powers, skills, etc.), the computer program
fills in the other character attributes using preprogrammed (random) values unknown
to the player. The actual course of events and the outcome of the game are determined
jointly through the interaction between the values input at the keyboard by the user
and those supplied by the computer as a programmed, sometimes random, response.
Although the software-generated portion of the game dialogue is apparently idea ex-
pressive, the ideas expressed by the software are not those of the human player. The
flow and amount of the author's originality has been intermixed with the product of
the machine. Finally, if one were to replace the human game participant with a com-
puter program which randomly selected allowable answers to the game software que-
ries, the amount of human input attributable to the game player would be reduced to
nil because the human would merely initiate the game dialogue-producing program,
yet would not contribute at all to its idea-expressive content. See generally Word
Processing,INFOWORLD, Jan. 11, 1982, at 17 (special section on word-processing technol-
ogy) and Supergames, INFOWORLD, Apr. 12, 1982, at 14 (special section on computer
video-games).
137. Mazer v. Stein, 347 U.S. at 217.
138. Presently, machines are not deemed able to formulate ideas capable of being
expressed in a copyrightable manner and cannot be said to be authors of original ex-
pressions within the meaning of the Act. Allowing copyright protection for expres-
sions not evidencing an underlying human-originated idea would raise significant
problems. See Apple Computer, Inc. v. Franklin Computer Corp. 545 F. Supp. at 824-
825.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 733

"author" might be deemed the software and the computer. Ex-

cept for pushing the button on the console of the computer, no
human creative process is involved in writing the story. 139 Al-
though the person initiating the program execution may desire
or have the idea of producing a story, the program supplies the
specific words, plot, characters and apparent "expression" of
an idea. 140 The story has no human "author" within the
bounds of common usage of the term and, within the meaning
of the Act,"' is possibly not copyrightable material under pres-
ent law. But, as programs like RACTER inevitably become
more sophisticated in mimicking human literary output, 142 the
increasing economic value of the works produced will heighten
the need for copyright protection.

V
Authorship, Originality and Artificial Intelligence:
Coping with the Man/Machine Threshold
We have seen that developments in AI challenge the applica-
tion of present copyright laws to computer generated software
and literature. Both the "original expression by an author" re-
quirement and the goal of affording copyright protection to ex-
pressions of ideas rest on a fundamental assumption that
"authors" are human beings. 43 As AI progresses, this assump-
tion will be tested in a legal sense as valuable and apparently

139. One might argue the ideas "expressed" in the story or program produced by an
Al machine are "derived" from the ideas of the original human programmer or the AI
software. This approach misconstrues the definition of a "derived work" as used in the
Act. See note 173 and accompanying text, infra. Also, one might question how far
removed the human can be from the process of expression and still maintain a claim
for authorship. For example, if a programmer designs an AI machine which produces
other AI machines which produce other AI machines, etc.-each of which randomly
produces stories, can the ideas of the original software programmer be effectively
linked or traced to the nth story written? What about a program which summarizes
news stories as they appear on the UPI wire service? See Waltz, Sci. Am. Oct. 1982, at
118, 132.
140. See note 136, supra.
141. An author is "one that originates or gives existence," WEBSTER'S NEW COL-
LEGIATE DICTIONARY 75 (1981). This colloquial meaning is given more definite meaning
under copyright law. See note 127, supra. Also, CoNTu's refusal to even speculate
upon the copyright issues presented by computer-authored literary works seems to
indicate an explicit presumption that the Act is meant to deal with human (or at least
non-computer) authors. See CoNTU, supra note 72, at 44.
142. Language understanding and synthesis is an area of great interest to Al re-
searchers. See Waltz, Sci. Am. Oct. 1982, at 118, 132.
143. See note 138, supra.
 COMM/ENT [Vol. 4

idea-expressive computer generated software and literature

are produced. If not found the product of a human author,
these works will not be afforded copyright protection under the
Act.'"
To prevent a gap in legal protection offered by copyright,
courts have several alternatives at their disposal. Specifically,
when a court determines that a given product of AI software is
"authored" by machine rather than a person, it may choose to:
(1) disallow copyright completely; (2) give authorship and
copyright to the computer and its software or find authorship
"shared" between the AI software and a human; (3) settle
copyright upon the owner of the underlying AI software or the
machine owner; or (4) create a fictional human author and as-
sign its copyright to the AI software copyright owner, the prob-
lem-specifier or the computer owner either individually, jointly
or in part. The following is an examination of the advantages
and disadvantages of these alternatives. Number four, propos-
ing a fictional human author and implicitly assigning its eco-
nomic rights, offers the best protection within a slightly
modified traditional framework of federal copyright law.

1. Disallow Copyright Protection Completely

Consistent with the traditionally implied assumption that

authors contemplated by the Act are necessarily human,
courts might choose to deny copyright protection for AI pro-
duced materials whenever they determine the human input
into the creation of such products not legally sufficient. Be-
sides having the advantage of consistency with the traditional,
anthropocentric approach to legal matters, this alternative
would spare courts from possibly having to implicitly credit in-
animate objects or machines with a capability of formulating
and expressing ideas. 45 The test of the quantum of human in-
put, if failed, would immediately preclude all associated copy-
right issues." It could be argued that any harm resulting from
denial of copyright protection would be adequately offset by
use of other available means of protection like trade secrets,

144. See notes 126, 127 and accompanying text, supra.

145. See note 136 and accompanying text, supra.
146. All requirements for copyright must be met for copyrightability to be found. 17
U.S.C. app. §§ 101-810 (1976 & Supp. IV 1982). See generally N. BOORSTYN, supra note
84, at 22-29.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 735
14 7
licensing agreements or patent law.
However, the benefits of eliminating copyright protection for
machine authored works are significantly outweighed by the
advantages gained by allowing some form of copyright protec-
tion. Though consistent with the traditional concept of an au-
thor as a person, denying protection is inconsistent with the
historically flexible interpretation and application of copyright
laws as technology has developed. 14 AI products are the result
of technological development and should be deemed 149
within the
historically flexible scope of copyright laws.
Also, one must recognize the social policy behind providing
copyright protection of human authored expressions is equally
applicable to works produced by AI software. Traditional
copyright policy aims at maintaining an economic incentive for
expression of valuable ideas, 150 promoting scientific and liter-
ary development' 5 ' and preventing monopolization of the mar-
ket for ideas and their derivative products. 5 2 Expressions
produced by AI software fit neatly into the same policy struc-
ture. Copyright availability for AI generated works would pro-
vide an incentive for their development and dissemination
because the proprietary interests of investors and inventors
would be guaranteed a degree of legal security not otherwise
available. 53 Scientists, artists, programmers and educators
would be free to produce and use AI materials just as they
would any other copyrighted written work. A competitive mar-
ket for AI products would also be encduraged by availability of
copyright protection." These important policy considerations
are not served well by the elimination of copyright protection
for AI products.

147. See text and accompanying notes, infra.

148. See generally, L. PATrERSON, COPYRIGHT IN HISTORICAL PERSPECTIVE (1968).
149. At the very least, products of AI software should not be denied consideration
under copyright laws because of their technological "newness" or novelty. See notes
79-84 and accompanying text, supra.
150. See L. PATTERSON, supra note 148, at 190.
151. Id. at 183, 191.
152. Id. at 224-225.
153. As will be seen, infra, copyright protection offers an inexpensive and workable
means of affording protection relative to the alternatives of patent registration, trade
secrecy and licensing.
154. Once AI programmers are assured of copyright protection for the products of
the AI programs they produce, economic incentive will induce production of Al pro-
grams just as copyright availability has become important to producers of normal
software. See generally CONTU, supra note 72.
 COMM/ENT [Vol. 4

Furthermore, the available alternatives of patent registra-

tion, licensing agreements and trade secrets do not promote
the same policy goals as copyright, are more costly and more
time consuming to create and enforce and are also victims of
the same fundamental assumption of a necessarily human ele-
ment in the creation of ideas and their expression.

(a) Patent Registration

Patent registration focuses on protection of the idea itself,
not its expression. 155 In a sense it creates a limited monopoly
on a given idea to provide incentive for the development of
such ideas and to allow society to benefit from their commer-
cial dissemination. 1 56 Copyright focuses on the expression, not
the idea, and seeks a much less restricted, more competitive
market.'5 7 Patent registration procedures, verification and en-
forcement are much more costly in dollars and in time than for
copyright. 1 58 Furthermore, patent protection of AI generated
ideas would create difficulties because courts would have to
credit machines with a capacity for thought. Protection of the
idea is the central element of patent law; to hold for patentabil-
ity, the court would have to find that the idea at issue satisfied
criteria equally applicable to ideas formed by persons. 159 Thus,
if the court found the "idea" produced by the AI machine pat-
entable, a strong implication would be made that, at least for
purposes of patentability of AI products, machines can think.
Another problem tending to preclude the patentability of AI
products is that computer programs and stories are not of

155. 35 U.S.C. § 101 (1970) provides: "Whoever invents or discovers any new and
useful process, machine, manufacture, or composition of matter, or any new and useful
improvement thereof, may obtain a patent therefore .. " The word "process" means
"process, art or method, and includes a new use of a known process, machine, manu-
facture, composition of matter, or material." 35 U.S.C. § 100(b) (1975). See also Nycum,
supra note 56, at 4-5 and Schmidt, supra note 64, at 355-57.
156. A patent is valid for a period of seventeen years from date of issuance. The
holder has the right to "exclude others from making, using, or selling the invention
throughout the United States .... " 35 U.S.C. § 154 (1970).
157. For instance, in copyright multiple expressions of a similar idea are permitted,
but in patent multiple use of the idea itself is prohibited without authorization from
the patent holder.
158. The costs of litigation in patent enforcement can be prohibitive. In 1967, a
study found costs of litigation for patent enforcement/infringement to range from
$25,000 to over $1,000,000. See Harris & Chuppe, Cost of Enforcement ofIndustrialProp-
erty Rights, 14 IDEA 77, 81 (conf. issue, 1970). See also Nycum, supra note 56, at 72-73.
159. To do otherwise would distort the meaning of "idea" under patent law. See
note 160, infra.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 737

themselves patentable.16 ° To make a computer program (Type

I AI product), or a story (Type II AI product), patentable
standing alone, the courts would have to extend patent to cover
an expression, not an idea or process.16 ' Stretching the scope
of patent law to such an extreme would open the proverbial
"floodgates" of litigation as persons sought to patent their liter-
ary works. 162 As a result, traditional policy goals of patent and
copyright law would be hindered as the "marketplace of ideas"
became enmeshed in a complicated legal environment. One
must conclude the availability of patent protection for AI
software produced algorithms is uncertain at best and should
not help form the basis for denying copyright protection.
(b) Licensing and Trade Secrecy
Licensing and trade secret contracts present less costly'63
and more expedient protection methods than patent registra-
160. A mathematical formula as such is not afforded protection under the patent
laws. Gottschalk v. Benson, 409 U.S. 63 (1972). The court must look to see if the mathe-
matical formula at issue is itself the primary object of patent protection or, if it is im-
plemented or applied in "a structure or process which, when considered as a whole, is
performing a function which the patent laws were designed to protect [e.g. transform-
ing or reducing an article to a different state or thing] .... " Diamond v. Diehr, 450
U.S. 175, 192 (1981). Thus, if a computer program is used in a process for curing rubber
which, considered as a whole, is "performing a function which the patent laws were
designed to protect," then the process containing the program and its formula is pat-
entable subject matter. Id. at 191-193. However, a mathematical formula or computer
program not sufficiently part of a process is not of itself patentable subject matter
(standing alone). Id. at 186-88. See also Gottschalk v. Benson and Parker v. Flook, 437
U.S. 584 (1978). See generally Nycum, supra note 56, at 16-28 and Schmidt, supra note
64, at 355-58.
161. "Excluded from ... patent protection are laws of nature, natural phenomena,
and abstract ideas." Diamond v. Diehr, 450 U.S. at 185. The abstract ideas contained
within the algorithm or the story are not patentable subject matter. "An idea of itself
is not patentable .. " Rubber-Tip Pencil Co. v. Howard, 68 U.S. (1 Wall.) 498, 507
(1874). See also Funk Bros. Seed Co. v. Kalo Inoculant Co., 333 U.S. 127, 130 (1940).
162. A guiding consideration in drawing the line between the "idea" not protected
by copyright and the "expression" which is protected is the preservation of the balance
between competition and protection. Herbert Rosenthal Jewelry Corp. v. Kalpakian,
446 F.2d 738 (9th Cir. 1971). By allowing protection for literary works under the patent
laws, this balance between protection and the promotion of competition would be de-
stroyed. Although the litigation and time "costs" associated with patent far exceed
those associated with copyright, patent protection can be used to suppress dissemina-
tion and commercial development of an idea, whereas copyright is unlikely to result in
such suppression because only the expression, not the idea itself, is protected. See
note 158 and accompanying text, supra. See also Nycum, supra note 56. (American
Telephone and Telegraph, among other large corporations, has been accused of ob-
taining patents on technical innovations and then "sitting" on them until the corpora-
tion recoups capital invested in older technologies).
163. The cost of printing a standard non-disclosure or licensing contract is minimal.
 COMM/ENT [Vol. 4

tion, 164 but their applicability to AI produced expressions is

made uncertain by fundamental assumptions of contract law
regarding the subject matter of the contract itself, mutual as-
sent and the identity and nature of the parties involved. In the
software industry, licensing and trade secret contracts are
made to limit access to a particular program algorithm, source
code or its use while at the same time allowing for a profit by
the supplier. 165 Although the contract itself may be valid be-
cause few limitations exist concerning the subject matter of
contracts, 66 the implication of contracting to protect legal
property interests a party claims to have in the apparently
thoughtful expressions formed by a machine is rather thought
provoking, to say the least. However, whenever the subject
matter of the contract is disputed by the parties, the courts
might not be able to avoid considering such issues. Also, fed-
eral copyright protection would afford a more uniform, consist-
ently applied body of law for protection of expressions
produced by AI machines and would foster a less restrictive
environment for the production and dissemination of such ex-
pressions than would licensing or trade secret agreements.
The policy goals of trade secrecy and, to a lesser extent, of li-
censing laws are not identical to those of copyright, and the
laws themselves are not uniform throughout the different
states.16' For these reasons, contractual licensing and trade se-
cret agreements are not good substitutes for copyright
protection.

164. Patent registration and approval are time-consuming, costly mechanisms when
compared with the immediate contract arrived at through use of trade secret or licens-
ing agreements made with employees or co-authors.
165. A trade secret is "any formula, pattern, device or compilation which is used in
one's business, and which gives him an opportunity to obtain an advantage over com-
petitors who do not know or use it." RESTATEMENT OF TORTS § 757 comment b (1939).
See generally Raysman, Protectionof ProprietarySoftware in the Computer Industry:
Trade Secrets as an Effective Method, 18 JUXIMETRICS J. 335 (1978) and Schmidt, supra
note 64, at 386-99.
166. Exceptions to subject matter not valid as objects of services contracted for in-
clude "illegal bargains." See generally 1 A. CORBIN, CORBIN ON CONTRACTS §§ 1373-78
(17th printing 1975).
167. 17 U.S.C. app. § 301 effectively preempts state and local governments from ex-
ercising jurisdiction over copyright in works "fixed" after January, 1978. At present,
states may exercise authority over works fixed prior to Jan. 1978, "unfixed" works and
certain other narrow areas not yet covered by federal statute. See generally N. BOOR-
STYN, supra note 84, at 10-20. At present, there is no uniform, national statutory basis
for enforcing rights granted under trade secrecy and licensing agreements.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 739

2. Computer or Man-Machine Hybrid as the "Author"

At the other extreme from denial of copyright to AI software

products is an alternative endowing the AI software creating
the materials in need of protection either exclusive copyright
ownership or one shared in some manner with one or more
humans. The disadvantages inherent in granting exclusive
copyright ownership to the computer and its software are man-
ifest. First, by doing away with the traditionally implied requi-
site of human authorship, this alternative concedes that a
machine can formulate ideas and express them.16 8 Although
the time may come when truly intelligent, sentient machines
are created, thereby causing a revolutionary change in the
corpus of traditional law, it presently seems unnecessary and
extreme. Second, even if courts did recognize machines as au-
thors, enforcement, ownership and assignment of rights would
not be easy. How could a machine be a real party in interest in
a lawsuit? Would a human be allowed to "own" a thinking
machine? Would the human act as trustee for revenues re-
ceived for products created and sold by the AI device? Who is
the master? Who is the servant? Third, although legal interests
have been recognized in trusts and corporations, these inter-
ests are based on the essentially human nature of the endeavor
(evidenced by shareholder participation).169 Public policy con-
170
siderations support the existence of trusts and corporations,
but the problems caused by endowing machines with copy-

168. If one grants copyright to "expressions" not purported to be the result of un-
derlying intellectual labor, the concept of authorship becomes meaningless. Natural
phenomena as well as animals would thereby become likely candidates for authorship
if, by accident, an "expression" is produced. (e.g., a cat walks on an ink pad and then
onto a piece of paper, leaving an "expressive pattern" in his trail. One cannot distin-
guish this expression from one intentionally done by a human for some unknown rea-
son. If one doesn't look to the expression of an underlying idea or some evidence of
intellectual effort having been expended as the basis for copyright, then the accidental
markings of the cat would be equated with the works of famous artists: both would be
copyrightable or neither would be. Critics of modern art might cheer this interpreta-
tion but the damage done to the goals and the availability of copyright is significant).
See Apple Computer, Inc. v. Franklin Computer Corp., 545 F. Supp. 812 (E.D. Penn.
1982) and Alfred Bell & Co., 191 F.2d 99 (2d Cir. 1976) (inadvertent expressions can be
"adopted" by a human and thus meet authorship requirements under copyright).
169. In other words, the "legal person" status of a corporation reflects a need for
society to recognize and facilitate a method for efficiently organizing human physical
and intellectual effort to achieve maximum return on monetary and human capital in-
vestment. See P. DRUCKER, CONCEPT OF THE CORPORATION 30-45 (rev. ed. 9th printing
1975).
170. Id. at 174-215.
 COMM/ENT [Vol. 4

right ownership outweigh any advantages of this alternative by

7
a wide margin.1 '
Alternatives creating a legal relationship between the man
and the AI software, granting copyright to man and machine as
joint authors, as works made for hire or through assignment of
copyrights held by the machine to the human, are also not via-
ble options. In these alternatives, the human would be granted
copyright protection with or through his "co-author," "em-
ployee" or "assignor": the AI machine. Less drastic than total
ownership of copyright by the machine, this approach still has
several fatal drawbacks. Importantly, to the extent these alter-
natives grant authorship to the AI machine, the implication
1 72
can be safely drawn that it can form and express ideas.
In normal joint authorship situations, each co-author has le-
gally enforceable rights and duties. 173 Economic gains can be
split according to prior agreement; liabilities can be assessed
and shared. In the man-machine joint author situation, the
concept of an "agreement" has little meaning and the notion of
a machine having legal rights and duties is absurd. 174 Allowing
the machine "authorship" rights would appear as a mere ges-
ture as the human exercised complete control over "their"
joint work. Also, the normal joint authorship concept as ex-
pressed in the case law, in the Act and by CONTU assumes the
parties involved in the agreement are humans. 75 The law does

171. One of thb fundamental goals of copyright protection is to give the author in-
centive to express his ideas. However, if the "author" is an AI machine,,the availability
of copyright protection would be of no consequence because the AI machine, being
non-sentient, would be unable to perceive the "incentive" or reap the economic bene-
fits of such protection. Thus, one must turn towards finding authorship in humans-
who are capable of recognizing and responding to such incentives-t]1ereby promoting
broad policy goals of economic development and dissemination of valuable ideas.
172. See note 168, supra.
173. 17 U.S.C. app. § 101 defines a "joint work" as work prepared by two or more
authors "with the intention that their contributions be merged into inseparable or in-
terdependent parts of a unitary whole." It is doubtful that a computer can have "inten-
tions" in a legal sense. See generally A. CORBIN, supra note 166, at §§ 2, 3, 9, 15. But,
one commentator argues for this approach when he states "[P resumably both the
original programmer for the machine and the person whose problem the machine is
designed to handle are both contributing to the formulation of the final problem. But,
since neither of these human beings need have full knowledge of the steps the com-
puter will perform to create a program, can it not be said that the machine is also at
least co-author of the program?" Milde, supra note 8, at 395.
174. See generally A. CORBIN, supra note 166, at §§ 3, 9, 622-26, 923.
175. See notes 72, 127, supra.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 741

not contemplate a man/machine "joint author."176 Neither a

typewriter, nor a trained animal has ever been accorded joint
177
authorship status under the copyright laws.
Under the provisions of the Act, an employer can obtain
ownership of the copyright to a work authored by an employee
while acting in his job function or by an agent contracted for
hire. 178 One could argue that the computer and its software is
the "agent" or "employee" of its owner or user, and that any
works produced by its software is thus a "work made for hire"
under the Act and is therefore copyrightable. But, it is difficult
to define a legal, contractual employment or agency relation-
ship between a human and a machine. Furthermore, the
human employee or agent has legal rights and duties by agree-
ment with his or her employer. The computer "employee" can-
not be said to enter an agreement for employment, nor can the
computer assert its legal rights or be cognizant of its legal du-
ties. Also, this approach is a misapplication of the statutory
meaning of "work made for hire." The Act clearly contem-
plates that the author of the "work made for hire" is a
human. 179 This factor is emphasized in CONTU's refusal to
even "speculate" upon the possibilities of computer-authored
written works. 180
In the broad view, if one were to extend the Act to cover
man-AI "work for hire" situations, this distortion would also
imply recognition of other man/machine and man/animal
agreements. The idea of a person "hiring" his trained parrot to
"author" a soundtrack or literary work distorts the scope of
copyright protection far beyond its meaningful, legally-en-
forceable boundaries.' 8 ' On the other hand, if one takes a nar-
row view and limits man/machine "work for hire" agreements
to cases involving AI software, we are still faced with the prob-

176. See In re Certain Coin-Operated Audio-Visual Games and Components

Thereof, supra note 95, and CONTU, supra note 72, at 44.
177. Although the products of such items or creatures may be copyrighted when the
expression results from human intellectual effort, copyright "authorship" has never
been extended to inanimate "tools" of the human.
178. 17 U.S.C. app. § 201(b) (1976) points out that a work prepared by an employee
within the scope of employment is a "work made for hire."
179. One does not speak of "scope of employment" when discussing machines used
in production or manufacture of a product at issue in copyright. It is questionable
whether a computer is an employee or a programmer/owner is an employer of a com-
puter within the meaning of the Act. Id.
180. See CONTU, supra note 72, at 42.
181. See generally A. CORBIN, supra note 166, at §§ 8, 9, 99, 100, 624.
 COMM/ENT [Vol. 4

lem of according contractual rights to a machine.182

Another alternative is the possibility of allowing the AI
machine to obtain the copyright and then "assigning" the eco-
nomic rights attending this copyright to the owner of either the
AI software or the computer.'83 But, before a property can be
assigned, it must be owned or controlled by the assignor. How
can a machine "own" something?"8 4 Assignment is a contrac-
tual concept which entails all elements necessary in a normal
contract: mutuality, consideration, performance, etc. It is hard
to see how these elements could be satisfied by the AI machine
without doing severe damage to the traditional ambit of con-
tract law. 8 5 Also, for an assignable copyright to be found in
the first place, the question of authorship would have to be al-
ready settled.8 6 Thus, a machine would be found the author of
expressions worthy of copyright protection, implying such ex-
pressions reflect underlying ideas. Once again, courts would
be faced with the problem of machines having ideas and ex-
pressing them or, in the alternative, of expressions which do
not "express" anything being given copyright protection.

3. Copyright to the Owner of the Underlying Al Software

or to the Computer Owner
At first blhsh, these alternatives present logical answers to
the problehs of authorship and originality posed to copyright
law by AI produced works. The copyright owner of the under-
lying AI software used to produce the expressions may assert
he is the author because the expressions are a derivative work
product of his copyrighted program. 187 Similarly, when the
182. See notes 127, 138, 178, 179 and accompanying text, supra.
183. Provisions for copyright protection for derivative works include "works con-
sisting of editorial revisions, annotations, elaborations, or other modifications which,
as a whole, represent an original work of authorship." 17 U.S.C. app. § 101 (1976). How-
ever, it is apparent that the "author" contemplated by the Act is a person, not a
machine. See CONTU, supra note 72, at 44.
184. "Ownership" of a property implied rights and duties which can be ascertained
and enforced in a court of law. To date machines have not been recognized as a "real
party in interest" and, as such, would not be able to protect "their" ownership rights.
Furthermore, ownership is sometimes spoken of in terms of dominion and control over
property. It is difficult to see how an AI machine could exercise control over its
product.
185. See generally A. CORBIN, supra note 166.
186. Before a copyright interest can be assigned, it must be awarded to the original
assignor/owner by the copyright office. Therefore, the question of authorship must
already have been settled.
187. See note 183, supra.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 743

computer owner also owns the copyright to the AI software,

one might consider allowing him to copyright the expressions
produced by it. Under these alternatives, products of AI
software would be deemed "derived" from the original, copy-
righted source. 88 Thus, the holder of the original AI software
copyright or the computer owner could be allowed to assign,
license, sell or divide his legal interest in any literary products
produced.
However, these approaches misconstrue the meaning of "de-
rivative" used by the Act and interpreted in the case law. By
its very nature, a derivative work is wholly or significantly
based upon the work of another. 189 The amount of the second
person's creative input is found to be insufficient to adequately
distinguish the second work from the first. But, in AI produced
works, the resulting literature or program is not "based upon"
or derived from the underlying, copyrighted AI software: it is
"created" by it. The similarity between the expression created
by a program and that of the actual program itself is usually
nonexistent except when the sole function of the program is to
produce a listing of its code. 190 In AI software applications, the
work product bears little or no resemblance to the underlying
program code which created it.' 9 ' Thus, it is not "derivative" in
a copyright sense. Al software performs a creative function,
lacking as a matter of fact when a work is found to be deriva-
tive from another. The owner of the copyright in the original

188. The products of the AI software would not be in existence but for the prior
original expression of a human programmer. This expression is embodied in the AI
software. Because the original programmer can own the copyright to such software if
he meets the requirements of the Act, one could argue that ownership of the copyright
to the products of this software should be traced back to this original programmer
because the products are "derived" from his copyrighted work. See note 183 and ac-
companying text, supra.
189. WEBSTER'S NEW COLLEGIATE DIcTIoNARY 303 (1981) speaks of "source" or "ori-
gin" when attempting to define the meaning of "derived." The Act defines a derivative
work as a "work based upon one or more pre-existing works, such as a translation,
musical arrangement, dramatization, fictionalization, motion picture version, sound re-
cording, art reproduction, abridgement, condensation, or any other form in which a
work may be recast, transformed, or adapted." 17 U.S.C. app. § 101 (1976).
190. For instance, in Type I Al programs, the code generator source program re-
mains unaffected by the creation of the product program. Similarly, in Type II situa-
tions, the output story bears no resemblance as a whole to the source code of RACTER
or to the contents of the various vocabulary data files used by RACTER to write the
story. Thus, in a copyright sense, the product of Type I and Type II AI program are not
"derived" from the underlying AI program because they bear no resemblance to it.
The AI program creates a product distinct from its own expression format.
191. See note 190, supra.
 COMM/ENT [Vol. 4

AI software or the owner of the computer can enforce their

rights against others who choose to derive programs from
theirs, 192 but they cannot claim derivative copyright protection
over the product of the AI software because this is not within
the contemplated meaning of "derivative" used in the Act. 193
AI software functions as a tool. Although legal ownership of
this tool may be held by a person other than the one who uses
it to create a product of value, for this product to be deemed a
"derivative work" of the original in a copyright sense, it must
significantly resemble it. 94 The creative function of AI
software virtually precludes this possibility and effectively
blocks application of the copyright concept of "derivative
work." Also, if the computer owner or the owner of the under-
lying AI software rights claims he is directly involved in the
authorship of the resulting expressions, he must be able to
prove his input in the creation of the finished work was sub-
stantial. Ownership per se of the right to the AI software or the
computer has little to do with the originality or the derivative
nature of the expression produced by the machine. When the
human contribution is found insufficient, the court cannot
grant copyright in the traditional sense. Without completely
eliminating the requirement of human originality and author-
ship, the AI machine owner or the owner of the underlying AI
software rights would rarely be able to obtain copyright protec-
tion for works substantially created by the machine.

4. Conservation of Ideas: The FictionalHuman Author

The alternatives discussed thus far either misconstrue the
basic tenents of copyright law, do not adequately serve tradi-
tional policy goals or require drastic revision of basic copyright
assumptions to achieve limited availability of copyright protec-
tion for products of AI software. Careful thought reveals an
alternative which avoids these pitfalls. When a court finds a
given product of AI software is "authored" by machine rather
than a person, the court should presume the existence of a
fictional human author and assign the appropriate fractions of
the copyright rights to the owner of the AI software copyrights,
the problem-specifier or the computer owner, either individu-

192. Id.
193. See notes 189 and 190, supra.
194. Id.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 745

ally, jointly or in part.'95

Founded on the assumption that expressions "express"
ideas and that ideas must spring from a human mind, this ap-
proach has several advantages. First, by presuming the exist-
ence of a human author, a court makes no implication that
ideas can be formed or expressed by non-humans. It main-
tains the traditional approach of law viewing creation as an
uniquely human endeavor and that authorship, originality and
creativity are human qualities not to be usurped by ma-
chines.'9 6 Second, contractual problems associated with sev-
eral of the other alternatives are avoided because the party
assigning rights is "human," and all other possible copyright
contenders are also human. 9 7 Third, this approach would ex-
hibit the traditional flexibility of copyright law as it expands
and changes in the face of new technologies.' 9 8 Fourth, copy-
right protection would be available under this alternative and
would thus serve the important policy goals cited earlier.'9 9
Economic incentives would be provided for creation of AI
software, scientific and literary interests would be furthered as
products of AI software were commercially disseminated, and
monopolization of the expressions produced by AI programs
would be limited to a degree consistent with traditional copy-
right policy. Finally, courts would not have to cope with philo-
sophical questions revolving around the idea/expression
dichotomy and the concept of "thinking machines" and their
legal rights and duties. The main structure of traditional copy-
right law would remain untouched, its basic requirements
merely extended to yet another form of technological
development.

195. Courts often have to determine which contending party owns how much, if
any, of a given copyright. The question is usually one for the fact-finder to determine
the quantity and quality of the contribution of each "author" to the given work. The
court will also look at all contracts, agreements, customs and other facts and circum-
stances when determining the distribution of "ownership" of the copyright. Under this
proposed solution, the "authorship" requirement under the Act would be presumed
met in Type I and Type II AI product situations whenever human contribution is found
to be trivial. The court would then look to the other requirements under the Act to
allow or disallow copyright in a given case. If copyright were found to exist, the court
would be able to "assign" the rights of ownership to the humans who, given all the
facts, are most deserving.
196. See notes 114, 115, 127, 136 and accompanying text, supra.
197. Id.
198. See generally L. PATTERSON, supra note 148.
199. Id.
 COMM/ENT [Vol. 4

VI
Solution: A Proposed Interpretation of Section
102 of the Copyright Act
To assist the courts in creating a presumption of a fictional
human author whenever a product of Type I or Type II AI pro-
grams evidences insufficient human contribution to satisfy
copyright requirements of "originality" and "authorship," sec-
tion 102 of the Act 2°° should be interpreted by the court as if it
incorporates the following passage:
In determining the copyrightability of expressions wholly or
partly produced by computer software and which are appar-
ently thoughtful or indistinguishable from those produced by a
human author, human authorship will be presumed and "au-
thorship" and "originality" requirements of this Act will be
deemed satisfied.2 ° '
This interpretation would enable courts to presume authorship
in AI product situations and allow time for analyzing the other
requirements of the Act to determine copyright availability in
a given situation. If all copyright requirements were met, then
the policy goals of copyright would adequately be served be-
cause protection would be granted. The court would have to
determine the correct apportionment of copyright ownership
rights among the owner of the underlying AI software, the
owner of the machine and the problem-specifier.2 °2 Assessing
the economic interests of human parties, the court would be on
familiar ground. The traditional integrity of the idea/expres-
sion dichotomy would be maintained in copyright law, and the
human focus of jurisprudence would remain inviolate.

200. See, e.g., discussion in Apple Computer, Inc. v. Franklin ComputerCorp., supra
note 114 (the court developed a "Seldon-Taylor" doctrine of copyright wherein expres-
sions to be given protection must have an underlying explanatory purpose or be at-
tempting to communicate in some way. Finding the expressions fixed in ROM to be, at
most, communication from a man to a machine, the court found programs stored in
ROM to be non-copyrightable expressions). See also Alfred Bell & Co. v. Catalda Fine
Arts, Inc., 191 F.2d 99 (2d Cir. 1976) (concept of inadvertent expressions (i.e. natural
effects of lightning strike caught in nature photograph) which are not planned for nor
created by the human, yet are copyrightable if the human "adopts" such expressions
as his own. Note, however, that in adopting the expression as his own work, the
human puts forth the minimal degree of "intellectual effort" needed to satisfy the Act).
201. See notes 195 and 200, supra. This "interpretation" could also be statutorily
enacted.
202. See note 195, supra.
No. 4] COPYRIGHT AND ARTIFICIAL INTELLIGENCE 747

VII
Conclusion
Throughout its history, federal copyright law has flexed
under the pressures of technological advancement. Develop-
ments in artificial intelligence will soon place even more ten-
sion on the scope and application of its traditional
requirements. Code generators, problem solving languages
and story-writing programs all act as harbingers of more pow-
erful, creative and economically important innovations soon to
come. Although dealing with many issues germane to the ad-
vent of computer technology, CONTU did not adequately ad-
dress the looming onslaught of AI based products and
creations. Finding that extension of copyright protection to AI
software work products is advantageous for the same reasons
copyright is extended to normal, human produced literary
works, this note has analyzed the threat AI developments pose
to copyright concepts of authorship and originality and has dis-
cussed several alternative solutions to afford copyright protec-
tion. The solution least destructive to traditional copyright
concepts presumes a fictional human author and then "as-
signs" copyright to the creator of the AI software, the problem-
specifier or the computer owner, as the court sees fit.
It should be emphasized that the solution proposed in this
note is a stop-gap measure. As AI develops more fully, the
ability to sidestep the tremendous impact such advances will
have on the legal system will diminish significantly. Man will
confront "thinking machines," and the impact will be unprece-
dented. The technical feasibility of artificial intelligence has
been shown at a relatively primitive level, but "[ o ] nce artificial
intelligences start getting smart, they're going to be very smart
very fast. What's taken humans and their society tens of
thousands of years is going to be a matter of hours with artifi-
cial intelligences. ' 203 We must prepare for "the possibility that
we may have to undergo still another redefinition of ourselves
as a species, another Copernican revolution that will move us
further yet from the center of the universe. "2°4

203. P. MCCORDUCK, MACHINES WHo THINK 351 (1979).

204. Id. at 328.