CopyrightAttributionandReconstructionofCopyrightabilityStandardsforAI-GeneratedWorksin theUSA

Abstract： The rapid development of the Artificial Intelligence （AI） industry has been matched by a rise in AIgeneratedworks. How to assign copyright ownership to AI-generated works is a dilemma that requires an urgent solution.These works can generally be divided into two categories： derivative works and non-derivative works. In the categoryof derivative works， if those works are the derivative works of an AI program， the programmer should own thecopyright. If the works are derivative works of some underlying work provided by the user， the user or the copyrightowner of the underlying work should own the copyright. In the category of non-derivative works， for those works generatedby AI non-independently， the user of the AI should own the copyright. For works generated by AI independently，these works are not protected by copyright under current US copyright law. But actually， these works meet the requirementof copyrightability， so placing them in the public domain is unreasonable. The copyright could be granted to theowner of the AI， the AI itself， or the corporation that owns the AI. In view of the negative impact of AI-generated workon the copyright market， copyright registration and an annual fee should be established for AI-generated works as theonly gate to getting them protected by copyright. In the process of such registration， the new criteria of review could beutilized.

Keywords： artificial intelligence; AI-generated works; copyright; public domain; fictional human author theory

CLC： D 923.41 DC： A Article： 2096?9783（2023）05?0138?11xxx 1 Introduction

Artificial intelligence （AI） is a branch of computer science that focuses on creating technology that simulates humanbehavior and makes decisions on their behalf. Now the question of who owns the copyrights to AI-generated worksis controversial in academia. Many scholars have researched the issue from a Chinese legal perspective， but there is nosystematic analysis or classification of the ownership of AI-generated works. It is difficult to develop a simple rule becauseof the varying types of AI-generated works and the many different approaches to dividing authorship. Moreover，with the development of AI， a series of new phenomena have emerged， such as Sofia， a robot who was the first to begranted citizenship， and the great number of works generated solely through AI， which will influence copyright law.These problems need to be solved urgently from a legal perspective; otherwise， human beings will be in chaos for a longtime[1]. The research will focus on analyzing the differences among certain AI-generated works and suggesting guidelinesfor a scientific system of assigning copyright ownership from the perspective of the law in the USA. In the last section，the author proposes legal designs to raise the requirement of copyrightability of AI-generated works.

2 Ownership of AI-Generated Works

2.1 Derivative Works

The generated work is a derivative work based on the program if it contains large blocks of expression that clearlycopy the expression contained in the program①. The Copyright Act defines a \"derivative work\" as \"a work based on pre-existing works， in which a work could be recast， transformed， or adapted②\". The 1976 Copyright Act clearly indicatesthat a second work is a derivative work only if it contains expressions from an underlying work. The secondarywork is derivative of the underlying work if there is a \"substantial similarity\" between the two works， such that \"an averagelay observer would recognize the alleged copy as having been appropriated\"③. So the AI-generated work is not aderivative of the program unless the generated work bears a recognizable similarity to expressions included in the underlyingprogram.

2.1.1 Derivative Works Based on the Program

A cut-and-paste program， such as Softkey's PC Paintbrush ClipArt Library， simply allows the user to \"cut-andpaste\"images contained in the program into the user's document. Another case is object-oriented programming. Perhapsthe most representative example of an objected-oriented program is the use of OOP methods when programming.According to this method， programmers create software subroutines， or \"objects\"， which other programmers use asbuilding blocks to create more complex programs[2].

The main difference between cut-and-paste and OOP is that the user of OOP may exert more creativity in choosingand composing the elements from the original program into the output. The user of an OOP is not just cutting andpasting pictures from the original work onto the output document. The OOP user must provide the functional design orflowchart to achieve an appropriate output program， and that is often the most difficult stage of the programming process.

2.1.2 Programmer Owns the Copyright in This Situation

For a cut-and-paste program， the programmer owns the copyright to the expressions that are contained in the program.

Works that are produced by the program and that recognizably incorporate these expressions are derivative worksbased on the program. So the user should not receive copyright for works generated by the original program because theprogrammer has the right to own the copyright of derivative works based on the programmer's original material.

2.1.3 Derivative Works Based on the Users' Providing Works

Programs that translate a user's document， such as Google Translate; programs that allow a user to edit picturesprovided by the user， such as Adobe Photoshop; or a program that allows a user to edit videos provided by the user，such as Corel Video Studio， would all create derivative works based on the user's input and not the program. In this situation，the original work contributes the most to the generated work， so the copyright does not go to the programmer. Instead，copyright ownership is awarded to either the user or the author of the original work （neither the user nor the programmer）.

2.2 Non-Derivative Works

A different question is presented where the output does not include any element of an underlying program. Programsthat write poems or compose music may generate works that are not derivative. In this situation， these workscould be divided into two categories： those generated by AI with human intervention and those generated solely by AI.

2.2.1 The Works Generated by AI with Human Intervention

Tremblay uses the term \"software agency\"， which is very fitting for these kinds of generative works. In this situation，the AI's actions and output can be traced back to a programmer's controls[3]. For this kind of work， where the userhas not contributed the minimum level of originality④， it seems fair to award the copyright to the programmer under currentlaw. However， the author argues it is more appropriate for the user to own the copyright to the output.

2.2.1.1 The Programmer Solely Owns the Copyright

The programmer has undoubtedly contributed most to the originality and creativity of the AI-generated work. Theprogrammer created the software program and generated the necessary coding that will instruct AI. Thus， one might claim that any AI-generated work is produced only through the wisdom of the programmer. So it would be reasonable toaward the copyright to the AI-generated work to the programmer. However， as the work was not finally fixed by the programmer，it is uncertain whether the programmer should be granted copyright ownership for any work beyond the originalsoftware[4].

The main impediment to the programmer's claim to copyright ownership is the programmer's failure to fix the AIgeneratedwork in a tangible medium. There is an argument favoring the award of copyright to the user， as the user isthe person who directly causes the work to be generated[5]. However， this argument was overthrown in the Stern Elec?tronics and Williams Electronic video game case. In this case， the defendant argued that the programmer did not fix theaudiovisual displays because the player's performance partly decided what display appeared on the screen. The courtclaimed \"many aspects of the display remain constant... regardless of how the player operates the controls\" and \"repetitivesequences of a substantial portion of the sights and sounds of the game\" meet the requirement of copyright protection⑤. In another case， the court also claimed the programmer fixed the audiovisual displays in the program⑥.

This reasoning applies to any program whose output is fairly repeatable. For instance， if the program composes thesame novel every time the user types in \"sky\" as a title， then it is the programmer， not the user， who fixes the work in atangible medium. Although it is a fact that the output would not be created without the user's typing， the limited functionof the user's role in fixation can be seen by the repeatability of the output. However， the argument fails in caseswhere the output is different each time the program inputs \"sky\". Therefore， the programmer can only satisfy the fixationrequirement where the output is repetitive or predictable.

In addition to the problem of fixation， there are other reasons for opposing the programmer as the copyright owner.Awarding ownership to the programmer has not been in accordance with the incentive theory for copyright[6]. The marketalready provides programmers with motivation to create programs because they can get sales benefits or licensingroyalties from users. If programmers enjoy both the copyright of software and the copyright of AI-generated works， itmay lead to a monopoly. This will weaken the motivation for users to utilize AI to generate new works. Another difficultywith granting ownership to the programmers is that they will often be unaware of the existence of such works[6]. Theuser may not notify the programmer and voluntarily pay royalties each time the user utilizes the program to produce anotherwork. In reality， the user is more likely to conceal the output[7]. It is difficult for the programmer to make sure thatthe user uses the software to generate works， copy AI-generated works， or even sell the works. Once the software is utilizedby the user， the programmer has to rely on the user's good faith. So the solution is not reasonable.

2.2.1.2 The User Solely Owns the Copyright

The requirement of originality is the main obstacle to the user's claim to copyright protection， even though the requirementof originality is low. To be copyrightable， a work only needs to be original to the author and possess someminimal degree of creativity⑦. To be sure， the requirement for creativity for the user is very low; even a slight amountwill satisfy. However， while the standard is low， it is not nonexistent. Kalin Hristov argues that users make a minimalcontribution to AI-generated works， and their claims for copyright ownership are the least convincing[8]. For example，when a user utilizes an AI program to compose a poem， the user will not meet the originality requirement by just typingin \"start\"， \"poem\" or clicking a button. Actually， the user makes no important contribution to the composition or arrangement.This limited contribution cannot meet the requirement for creativity.

Even though the user makes little contribution to the production， assigning the copyright to the user is more meaningfulfrom the perspective of social policy and economic incentives. The user is the one who decides whether AI-generatedwork is produced， so the interests of the users comply with the public's interests because the user decides thequantities and quality of articles that enter the market. The programmer has less aspiration to generate more AI-generatedworks as his or her program is already assigned copyright protection. If considering incentivizing the production ofmore excellent works， it is better to grant copyright protection to the user， who is more likely to take part in generatingworks than the programmer. In the production of news， a user usually plays a larger role in pushing AI-generated news to the public than an AI developer， as the user decides what news articles are worthy of publication. However， from theperspective of traditional copyright rules， it will be hard for a user to gain a copyright because he has contributed verylittle to the AI-generated work. The simple behavior of pressing a button will not meet the requirement of creativity asestablished in Feist[9].

A problem exists because the party that can obtain the copyright is not the party that benefits from the policy ofcopyright. It may be unreasonable to decide that an individual's contribution leads to copyright protection for anotherperson. But this will be logical if the programmer has a transfer agreement regarding copyright ownership with the user.The programmer could give the copyright of AI-generated works to the user in the license agreement of such AI software.

2.2.1.3 Joint Authorship

The Copyright Act defines a \"joint work\" as \"a work completed by two or more authors with the consciousness thattheir contributions be merged into indivisible parts of an integrated whole⑧\". Parts of a unitary whole are \"interdependent\"when they have some meaning standing alone but achieve their primary significance due to their common effect⑨. There is a dispute about whether each joint author's contribution should be copyrightable or whether the individualcontributions do not need to be copyrightable⑩. Professor Nimmer claims that the contributions of each author neednot be copyrighted. However， the majority of courts support Professor Goldstein's opinion that each contributionshould rise to the level of copyright. The main reason that Nimmer's view has not been adopted by the courts is thatthe Goldstein model strikes a balance between copyright law and contract law by allowing any person to entrust anotherwith authorship status by contract and by not allowing parties to utilize contract law to transform uncopyrightable matterinto copyrightable matte.

This aspect means that to be considered joint authors， both the programmer and the user must satisfy the two requirementsof Section 102. For the first requirement， each individual's contribution should be independently copyrightable.But in fact， the user's contribution of button pressing or typing is too simple to be copyrightable， as pressing a buttonor typing a word is unlikely to meet the requirement of minimum creativity from the perspective of the law. Anotherimportant requirement of joint authorship is the intent that each joint author has when each joint author begins theircreation. On the second aspect， it is difficult to claim that the user collaborates with the programmer for the same intentionat the beginning， as the parties who will later use the program were not ensured when the program was created. Sothe requirement for joint work is unlikely to be satisfied. Even though assigning joint authorship might seem like agood solution， it is in fact an impossible approach.

2.2.1.4 AIs as Joint Authors

Because a joint author must intend to contribute to a unitary whole， an AI may only be a joint author if it has independentconsciousness， the ability to \"intend\". However AIs do not possess true intentionality in the current technicalsituation. Lawrence B. Solum claims that even if computers learn to simulate intentionality， they will never truly haveintentionality because they have no grasp of \"meaning\" to do things and are always programmed[10]. Perhaps it wouldchange their minds if AIs could possess free， independent will in the future. But now， it seems fair to say robots lackthe intentionality required to be classified as joint authors.

2.2.2 AI Creates Works Independently

As AI becomes more advanced， it can learn and create independently without human intervention. AI like GoogleMagenta uses a simulated \"human brain\" of neural networks to compose music without the aid of specific algorithms or human input[11]， and IBM's Watson was used to edit a movie trailer by analyzing a completed film for thematic highlightsand splicing them together to foreshadow the longer story[12].

If the person controlling the AI can reasonably expect the AI to create a certain product or if the AI is programmedto generate works， then the AI-generated works are produced with human intervention. However， if the AI'sactions are outside that range of expectation and are automatically generated， or if it goes rogue， just as one cannot predictthe random music created by wind chimes[13]， then the products are generated by AI independently.In fact， under current law， the work generated by AI independently cannot be protected by law. It seems that placingthe work in the public domain is the current solution. However the author argues that placing them in the public domainwill undermine the copyright industry， so it is not a suitable approach. Since humans do not contribute to theworks generated by AI independently， granting copyright ownership to humans under traditional legal theory is unreasonable.So assigning this kind of work is a problem; it is necessary for people to explore new theories and solutions tosolve the problem.

2.2.2.1 Place the Work into the Public Domain

A good solution would be for all these works to enter the public domain immediately. Some scholars hold thisview. Placing all AI-generated works in the public domain would enhance the public's interest in getting access tothose creative works without undermining incentives for content producers[14]. The public domain provides a path thathelps expand creativity and encourages human creators to create more works.

The author disagrees with the arguments above. The paper argues that giving copyright protection to AI-generatedworks is necessary when works are completed without human intervention. In fact， placing the works into the public domainwill undermine the copyright system and have a serious negative impact on the culture industry.

Firstly， without a period of protection for independently AI-generated works， there is no incentive for users to generatemore work by using AI， which will weaken their desire to purchase AI because these works will bring them no value.If users choose not to purchase this software， profits from AI developers will greatly drop， which may decrease a developer'sdesire and financial power to continue creating and improving their AIs. On the one hand， programmers andcompanies have invested large amounts of time and money into the creation of AI; on the other hand， few users chooseto purchase the AI. This phenomenon could finally curb innovation by discouraging developers and companies from investingin the AI industry. As a result of the declining AI industry， there would be less material available for use inteaching， scholarship， and research under the Copyright Act's fair-use doctrine. The arts， education， literature， andtechnology， among others， could suffer significantly. Now many AI machines are utilized by the developer to serve businesses，such as Dreamwriter[15]， an AI that is developed by the company Tencent in China and could produce a largenumber of articles of news in a short period of time. Tencent itself is also the user; the interest of Tencent would be seriouslydamaged if these articles could not be protected by copyright， which would inhibit Tencent's enthusiasm for investingin the AI industry， so protecting works generated by AI independently is necessary.

Secondly， putting the works in the public domain may undermine the culture industry and copyright system in thefuture. A considerable part of the creation of works in literature and art may be monopolized by AI because it is capableof mass production. If these works are not given copyright protection， there will be a great many \"works\" that are notcopyrighted and are not substantially different from the general works created by humans. At this time， for any potentialusers， if there are enough artificial intelligence creations in the public domain， there is no need to pay for the use ofcopyrighted works. Thus， the copyright value of all other human works may tend toward an infinitely zero value. Mosthuman authors will exit the market， with the exception of those who have a strong desire to express themselves or donot expect compensation. The copyright industry will gradually dry up. Copyright licensing and transaction volume willplummet. When human works lose market appeal and mobility， the economic incentives of creation will also disappear.Ultimately， the interests of neighboring rights holders will be seriously damaged， such as publishers， audio and videoproducers， broadcasters， and so on.

Thirdly， placing the works in the public domain has difficulty in execution. If copyright protection is given toworks that are generated by AI independently， people must distinguish these works from AI-generated works that have a human contribution. In reality， it is difficult to distinguish these two. It demands a detailed investigation into the essenceof the interaction between the user and the AI and an accurate understanding of the functioning of the AI program.For example， in the case of Penguin Books U. S. A.， Inc. v. New Christian Church of Full Endeavor， Ltd.， thecourts were required to determine whether the works actually do originate from a spirit voice other than the human intermediary-a decision that courts have struggled to avoid in these cases.

2.2.2.2 AI as Copyright Owner

Whether an AI should be seen as the author of a copyrightable work is one of the most difficult problems in copyrightlaw today[16]. Since only the authors of creative works may enjoy legal protection， some scholars have argued thatthe definition of \"authorship\" should be redefined to include both human and non-human authors. Professor Ryan Abbottis one of the scholars who claim legal rights for non-human authors. In his paper， he argues that awarding copyrightownership to non-humans is an effective way to encourage further development in AI[17]. In theory， this could preventAI-generated works from falling into the public domain. This theoretical solution， however， is controversial andmay give rise to an uncertain future full of legal challenges. An AI is not a natural person and cannot be held legally responsiblein court. So they may not be considered authors according to guidelines set by the U.S. Copyright Office.

Moreover， when considering whether to award copyright ownership to an AI， the Copyright Office and the courtsmust consider if such ownership will not encourage the AI to generate more works because AI has no independent consciousnessto generate more works in current technical conditions. Moreover， AI has no awareness of how to enforce itscopyright. Redefining copyright ownership to include non-human authors would undermine the current legal system inthe USA， creating further uncertainty about legal problems such as AI's tort liability. As a result， the scientific solutionwould require both stability in the legal system and incentives for AI developers.

2.2.2.3 Work Made for Hire

According to the doctrine， \"if a work is made for hire， an employer is considered the author even if an employeeactually created the work.\" The employer can be a firm， an organization， or an individual. These principles on grantingcopyright ownership to a party that did not directly produce a copyrightable work are suitable for solving the problemspresented by AI. The employee-employer relationship in the \"made for hire\" doctrine may be applied to AI programsand their developers if the terms \"employer\" and \"employee\" are reinterpreted within the confines of the doctrine.

From the perspective of current law， the traditional definition of an employee may be restrictive as a person usuallybelow the executive level who is hired by another to perform a service， especially for wages or salary， and is underthe other's control. After the case of Goldstein v. California， the terms have been expanded. Just as the term \"author\"may be interpreted as different entities （an individual， a firm， or an organization）， the term \"writings\" could meanbooks， sound recordings， films， images， and even computer code. So the meaning of employer and employee should beexpanded in order to satisfy such newly arising requirements and reflect contemporary social changes. A more flexibledefinition could also be utilized to accommodate the current legal limitations of AI-generated works.

Furthermore， if a relative interpretation is used， an AI machine may be thought of as an employee， as its servicesare employed by its programmer or owner. This new interpretation of two terms （employer and employee） in the \"madefor hire\" doctrine could be vital for the development of AI in the future by providing the incentive of copyright protectionto AI developers. Annemarie Bridy argues that the \"work made for hire\" doctrine is a more suitable solution. Thesolution solves the problem of assigning authorship because it is an effective mechanism for directly vesting copyrightownership in a legal person who is not the author-in-fact of the work[18]. Russ Pearlman also claims that the \"made for hire\" doctrine is a practical method， as it can avoid the confusion of treating an AI as a legal or natural person underthe law[19].

As Kalin Hristov argues， the theory of work for hire has some advantages. The most distinct merit of the theory isthat copyrights are attributed to a legal or natural person instead of a non-human， eliminating a puzzling debate overthe legality of non-human authorship， which will not shake the traditional civil law system[20]. But there is a fatal disadvantageto the theory. It is hard to give a definition of a legal， contractual， or agency relationship between a human anda machine. But the human employee or agent has legal rights and duties by signing an agreement with his or her employer.The AI \"employee\" cannot be in accordance with a real agreement for employment， nor can the AI claim its legalrights or undertake duties. Also， this solution is an abuse of the original legal meaning of \"work made for hire\".From the perspective of the law， the author of the \"work made for hire\" is a human. The requirement is affirmed inCONTU's rejection to even \"speculate\" upon the computer-authored written works. In the broad sense， if peoplebroaden the law to cover man-AI \"work for hire\"， this distortion would also mean recognizing other man-machine andman-animal agreements. The concept of a person \"hiring\" his parrot to \"author\" a soundtrack distorts the sphere ofcopyright protection far beyond its original， legally enforceable boundaries. So this theory still does not comply withexisting legal requirements.

2.2.2.4 The Fictional Human Author Theory

One alternative solution for protecting the copyright of works generated solely by AIs is to set a natural person orlegal person that has a close relationship with AIs and award the copyright to the person. The representative solution isthe Fictional Human Author Theory. Under this theory， when a court finds that a work of AI is \"generated\" solely bythe AI rather than a human， the court should presume the existence of a fictional human author and assign the copyrightto the owner of the AI[20]. Under this theory， owners of AIs should be awarded copyrights to works that are generatedsolely by AI， even if they are not the \"creators\" of the work. In reality， the Copyright Office's registration for The Po?liceman's Beard is Half Constructed appears to follow the Fictional Human Author Theory. In the registration， the CopyrightOffice recognized an AI named Racter as the fictional human author in fact but awarded ownership of the copyrightto a human， William Chamberlain， the owner of the AI.

The theory has many advantages. Firstly， by presuming the existence of a human author， a court could ensure thatcopyright ownership is awarded to a human. It maintains the traditional law system that copyright ownership should notbe assigned to non-humans. As Timothy L. Butler argues， courts would not have a debate over philosophical questionsabout the concept of \"thinking machines\" and their legal rights and duties[20]. Secondly， contractual problems associatedwith work made-for-hire theory are avoided because the party assigning rights is \"human\" and another partyis also human. Thirdly， copyright protection would be available under this solution and would thus be in accordancewith the goals of copyright. Furthermore， economic incentives would be available under the theory. Awarding the rightsto the owners will give them more incentive to use the AI program and produce new works. It will also encourage theowner to purchase more AI programs from the programmer or the company， which will stimulate the development of theAI. Evan H. Farr argues that the fictional human author theory fails to mention who will initiate the related litigationand enforce the copyright[21]. The author claims the owner of the AI is suitable for initiating litigation and enforcing therights.

2.2.2.5 AI Who Has the Citizenship

With the rapid development of AI technology， AI with the qualification of citizenship emerged. For example， theRobot \"Sophia\" was granted citizenship in Saudi Arabia. The author argues that more and more AIs may also have nationalityand citizenship rights in the future.

2.2.2.5.1 The Dilemma of Granting Copyright to AI Who Has Citizenship

From the historical process of the scope of civil subjects， the system of civil subjects is constantly evolving andequipped with inclusiveness. The process of obtaining slavery rights in human history， the legal person system createdby legislation along with economic and social development， and the protection of animal rights with the view that personhoodand dignity should be granted to non-human animals in recent years[22] have proved that the civil subject isnot a closed system， that conforms to social development. In the same way， when the creative action of AI has a great influenceon human society， it will invariably be endowed with such status in law. In addition， the EU intends to affirmthe legal status of intelligent AI by means of a \"fictional electronic person\"， with relevant legislative proposals havingbeen formally put forward[23].

If an AI has nationality and citizenship， it seems that the AI has the qualifications to own the copyright. However，the aim of copyright law is to \"promote the progress of science and useful arts\". Even if the AI has nationality and citizenship，the Copyright Office and the courts must consider whether such a copyright will encourage the AI to generateworks for society's benefit. Now， awarding the copyright to AI fails to realize the goal of copyright law because AI hasno independent awareness and does not know how to enforce its rights.

2.2.2.5.2 The Guardianship Rights of the Owner of AI

This author believes that guardianship rights in civil law can be used to treat artificial intelligence as citizenswhose intelligence is not yet fully mature. AI could be guarded by the personal owner. Guardianship is the guardian'sright to supervise and protect the personal rights of protected persons， such as minors， people with mental illness， orpeople with developmental disabilities. The guardian has the responsibility to make major decisions about the healthand well-being of the ward. The guardian could make financial decisions utilizing the assets and estate of the protectedperson. The machine seal named Paro obtained Japanese household registration as a pet in 2010， and the inventor'sidentity on the household registration card is \"father\" [24]. This is an inspiration for establishing a robot-oriented guardianshipsystem.

In the author's view， in the future， AI will be granted citizenship and be bestowed civil rights around the world. Atthis time， the relationship between AI and individual owners can be drafted as a guardianship relationship. That is， theAI owns copyright， but property rights in copyright are guarded by the owner. In order to best protect the AI's legalrights， the law may stipulate that the guardian shall bear a duty of care. If the guardian fails to perform his duties ofguardianship or infringes on the legal rights of the AI， the guardian shall bear responsibility; if the guardian causes anyloss of property for the AI， the guardian shall compensate for the loss. Guardianship， involving copyright litigation， maybe enforced by the owner. The guardian cannot dispose of the property of the AI at will. When an AI causes damage toothers and needs to bear the liability for compensation， the guardian can use the AI property to pay compensation. Thispaper argues that a liability insurance system for guarded AI should be established. Through the compulsory insurancesystem， the guardian will insure the AI that is being supervised， thus helping to solve the possible infringement problemof AI[10].

2.2.2.5.3 The Corporation Owns the Copyright of AI's Output

As discussed above， personal owners can serve as guardians. But for those corporations that use many AIs， it isnot suitable for corporations to serve as guardians because the cost of too many AIs may be too high for the company.The author claims the theory of works for hire can be used in this situation. Under the theory， corporations that own anAI possess the copyright to works created solely by an AI citizen. Why is the theory of work for hire appropriate in thiscase？ Because AI already has citizenship and has the same legal qualifications as humans， it is reasonable to regard itas an employee. The contractual employment or agency relationship between a human and an AI complies with the law.The author also argues that there should be a liability insurance system for the AI employed by corporations in this situation，just like employers insure their human employees.

3 Reconstruction of Copyrightability Standards of AI-Generated Works

From the perspective of the Copyright Law， it mainly realizes its economic function by prospering the cultural market.Granting copyright is a critical motivation for authors to create more works.

However， if AI with \"high-yield characteristics\" is not regulated， its creations will undoubtedly have a huge impacton traditional copyrights from the supply side， which will lead to confusion in the originally limited marketplace.The works of AI will inevitably reduce the market pricing power of human authors， resulting from their low operatingcosts and abundant creative resources. The price of AI-generated works may be very cheap. This may bring huge competitivepressure to human copyright owners， which may ultimately lead to a shrinking of the whole copyright industry. The Intellectual Property Promotion Plan 2016 made by the Japanese government has performed a good analysis of theproblem. It points out that the creation of AI may lead to a sharp increase in the number of copyrightable works. Also，giving intellectual property protection to works made by AIs could lead to too much protection， so it is also important togive intellectual property protection to AI creations with a certain market value to stop people from \"free-riding\"[25].Thus， it is necessary to set new standards for the copyrightability of AI-generated works.

3.1 Review about Value Orientation

According to the principle that intellectual property rights should be given to resources with rare attributes， the authorbelieves that the standard of copyrighted works for AI should be higher than the standard of human creation byconsidering the characteristics of \"high yield\" for AI-generated works. So the minimum requirement as stated in thecopyright law cannot be used as a criterion for the copyrightability of AI-generated works. AI-generated works of lowquality cannot be protected by copyright. The AI-generated works of high quality can compete with the works createdby humans， which is conducive to the formation of a \"defense embankment\" that eliminates the negative impact of AIon the human works.

Now AI can compile news by collecting and judging the latest event information. But it turns out that most AI payslittle attention to the technical characteristics of news writing. For example， the AI program named Wordsmith canmass-produce financial and economic reports[26]. Although it can generate different versions of articles with diversestyles， the content is simple and boring. It must be pointed out clearly that articles should be in pursuit of social value.It means that articles are not only a true description of facts but also emphasize the factor of social value orientation.For instance， reports on natural disasters should focus more on heroic deeds， disaster prevention， and response whileensuring the accuracy of the damage suffered by the disaster areas. Such reviews about social value put forward higherrequirements for AI.

3.2 Review about Promoting the Progress of Science and Art

The research states the second requirement for the copyrightability of AI-generated works is enhancing the progressof science and art， which is also prescribed in the copyright law. Now， works created by human beings do not requirea minimum artistic aesthetic but merely minimal creativity. For example， if a student in junior high writes a poemabout his mother's life， it is protected by the Copyright Law despite its lack of literary value， as the creation of works ismainly a factual behavior and does not require too much for the value of the art. But if AI writes poems， the work mustcontain a certain artistic or literary value. If the poems written by AI are a simple combination of words or even includemany linguistic errors， then from the perspective of artistic value， these poems should not be awarded copyright. Therefore，the author claims that the copyrighted property of AI-generated works must be identified with the goal of promotingthe progress of science and art， as with current copyright law.

As discussed above， the research proposes that it is necessary to improve the access standard for AI-generatedworks protected by the Copyright Law. The standards could be whether the value orientation of AI-generated works isbeneficial to society and whether the work promotes the progress of science and artistic development.

3.3 Registration and Paying an Annual Fee

In order to be reviewed， the author argues that AI-generated works must be registered in order to obtain copyright. The author should also pay an annual fee for registration to maintain the copyright. There are three advantages. Firstly，the copyright registration system can clarify the rights of holders of AI-generated works. The copyright registration systemcan make the attribution of copyright more distinct， especially in the digital environment when vast numbers ofpieces flood the market. Secondly， registration and paying an annual fee can help to review the works， which may effectivelylimit the copyrights of many low-quality products so that the total number of works created by humans and AIgeneratedworks is relatively balanced. This measure can avoid a monopoly in the copyright market due to the quantitativeadvantage an AI possesses.

Thirdly， a registration policy could filter out those works with limited commercial value. Under traditional copyrightlaw， a work can be granted copyright almost automatically without the requirement for an extra act. But in somecircumstances， authors may object to granting copyright to their works in the digital age. For example， many authors inpost-bar may publish their short articles online to express their opinions， with no intent to gain a copyright license fee.They just want to distribute their works online freely and do not need the motivation of copyright to continue. In fact， inthe digital market， many websites earn their primary income through the web traffic generated by their work. So formany authors， free circulation of their works is more important than strict copyright protection. Paying a copyright feefor their works may decrease network traffic to their works， which is not an outcome that authors wish to see.

For registration， there would be direct costs， such as the payment of an annual fee. There may also be indirectcosts， such as attorney fees. An author will analyze the costs that must be invested and the profits that may be gained asa result of registration. From a rational perspective， if the profits outweigh the costs， then the author will register thework. This is a cost-benefit analysis and will encourage the author to consider how to commercialize works to gain thegreatest benefit after registration， which is in accordance with the aim of copyright law because copyright law gives acommercial motivation to create. If no economic benefit was related to creating the work， there is no meaning behindthe granting of copyright， which could exclude a considerable portion of AI-generated works from copyright protection.

4 Conclusion

Over the last few decades， AI has become more advanced. The emergence of a large number of AI-generatedworks and AI with citizenship has a new impact on copyright law. There are many kinds of AI-generated works， so assigningcopyright ownership is a complex thing that requires categorical discussion. To eliminate the negative impacton the human copyright market， the author proposes setting the copyright registration and paying the annual fee as thenecessary approach for AI-generated works to gain copyright protection. The research also argues that new criteria forreview should be established in registration to filter out works of low quality. The new criteria determine whether thevalue orientation of AI-generated works is beneficial to society and whether the works promote the progress of scienceand artistic development.

References：

[1] KAPLAN J. Humans need not apply. a guide to wealth and work in the age of artificial intelligence[M]. New Haven：Yale University Press， 2015： 3.

[2] BARKAN D M. Software litigation in the year 2000： the effect of object-oriented design methodologieson traditionalsoftware jurisprudence[J]. High Technology Law Journal， 1992（7）： 315-368.

[3] OSTERGARD R L. Intellectual property： a universal human right？[J]. Human Rights Quarterly， 1999（1）： 156-178.

[4] YU R. The machine author： what level of copyright protection is appropriate for fully independent computergeneratedworks[J]. Pennsylvania Law Review， 2017（165）： 1245-1269.

[5] SAMULSON P. Allocating ownership rights in computer-generated works[J]. University of Pittsburgh Law Review，1986（47）： 1185-1228.

[6] DENICOLA R C， MACHINA E. Copyright protection for computer generated works[J]. Rutgers University Law Review，2016（69）： 251-288.

[7] WU A J. From video games to artificial intelligence： assigning copyright ownership to works generated by increasinglysophisticated computer programs[J]. AIPLA Quarterly Journal， 1997（25）： 131-180.

[8] HRISTOV K. Artificial Intelligence and the copyright dilemma[J]. IDEA， 2017（57）： 431-454.

[9] SAMUELSON P. Allocating ownership rights in computer-generated works[J]. University of Pittsburgh Law Re‐view， 1986（47）： 1185-1228.

[10] SOLUM L B. Legal personhood for artificial intelligences[J]. North Carolina Law Review， 1992（70）： 1231-1288.

[11] Hutson M. How Google is making music with artificial intelligence[EB/OL]. （2017-08-08）[2022-09-27]. https： //www. science. org/content/article/how-google-making-music-artificial-intelligence.

[12] IBM. Watson creates the first AI-made film trailer-and it's incredibly creepy[EB/OL]. （2016-09-02）[2022-09-27]. https： //www. wired. co. uk/article/ibm-watson-ai-film-trailer.

[13] KHOURY A H. Intellectual property rights for hubots： on the legal implications of human-like robots as innovatorsand creators[J]. Cardozo Arts amp; Entertainment Law Journal， 2017（35）： 635-668.

[14] PERRY M， MARGONI T. From music tracks to google maps： who owns computer generated works？[J]. ComputerLaw amp; Security Review， 2010（26）： 621-629.

[15] Media Summit： AI shaping modern news reporting[EB/OL]. （2017-11-17）[2022-09-27]. https： //news. cgtn. com/news/3d45544d78554464776c6d636a4e6e62684a4856/share_p. html.

[16] WU A J. From video games to artificial intelligence： assigning copyright ownership to works generated by increasinglysophisticated computer programs[J]. AIPLA Quarterly Journal， 1997（25）： 131-178.

[17] ABBOTT R. I think， therefore iinvent： creative computers and the future of patent law[J]. Boston College Law Review，2016（57）： 1079-1126.

[18] BRIDY A. Coding creativity： copyright and the artificially intelligent author[J]. Stanford Technology Law Review，（2012）： 1-28.

[19] PEARLMAN R. Recognizing artificial intelligence （AI） as authors and investors under u. s. intellectual propertylaw[J]. Richmond Journal of Law amp; Technology， 2018（24）： i-37.

[20] BUTLER T L. Can a computer be an author -copyright aspects of artificial intelligence[J]. Comm/Ent L. S， 1 981（4）： 707-748.

[21] FARR E H. Copyrightability of computer-created works[J]. Rutgers Computer amp; Technology Law Journal， 1989（15）： 63-80.

[22] BILCHITZ D. Moving beyond arbitrariness： the legal personhood and dignity of non-human animals[J]. South AfricanJournal on Human Rights， 2009（25）： 38-72.

[23] BECK S. The problem of ascribing legal responsibility in the case of robotics[J]. Ai amp; Society， 2016（4）： 473—481.

[24] CNN. Meet Mr. Robin， Grandma's robot buddy[EB/OL]. （2014-05-16）[2022-09-27]. https： //edition. cnn. com/2014/05/16/tech/innovation/meet-mr-robin-grandmas-robot/index. html.

[25] Intellectual Property Strategic Program 2016[EB/OL]. （2016-05-09）[2022-09-27]. https： //www. kantei. go. jp/jp/singi/titeki2/kettei/chizaikeikaku20160509_e. pdf.

[26] Wordsmith's 'robot journalist' has been unleashed[EB/OL]. （2015-10-20）[2022-09-27]. https： //www. wired. co.uk/article/wordsmith-robot-journalist-download.

美國人工智能生成物的版權歸屬和可版權性標準的重構

余祥 聶建強

摘 要：人工智能 （AI） 行業的快速發展與人工智能生成物的興起相匹配。如何為人工智能生成物分配版權所有權是一個亟待解決的難題。這些作品一般可以分為兩類：演繹作品和非演繹作品。在演繹作品類別中，如果生成物屬于人工智能程序的演繹作品，程序員應擁有版權。如果生成物屬于用戶提供的一些基礎作品的演繹作品，用戶或基礎作品的版權所有者應擁有版權。在非演繹作品類別中，對于人工智能非獨立生成物，其版權歸人工智能使用者所有。對于人工智能獨立生成物，根據現行美國版權法，這些作品不受版權保護。但實際上這些作品符合版權的要求，將作品置于公有領域是不合理的。版權可以授予人工智能的所有者、人工智能本身或擁有人工智能的公司。鑒于人工智能生成物對版權市場的負面影響，應為人工智能生成物設立版權登記和收取年費，作為其獲得版權保護的唯一途徑。在這種登記過程中，可以使用新的審查標準。

關鍵詞：人工智能；人工智能生成物；版權；公有領域；虛擬人類作者理論