人工智能生成物法律保护模式探索与比较研究

徐博

Exploration and Comparative Research on Legal Protection Mechanisms of Artificial Intelligence Machine Generated Products

Xu Bo

（JD Technology Research， Beijing 100101， China）

Abstract：In the present times， the law does not clearly stipulate the protection of Artificial Intelligence Machine Generated Products （hereinafter as "AIMGP"）. Due to the absence of legal protection mechanisms， economic values carried by AIMGP may cause arbitrary uses by all stakeholders involved， result in court disputes， and ultimately affect the healthy development and sound operations of the market. Therefore， finding reasonable protection paths for AIMGP as early as possible can not only achieve the effect of resolving issues and stopping needless disputes， but also encourage production and dissemination of AIMGP， as well as rescuing the growing cultural demands of people from aesthetic weariness of repetitive creations.

Key words： intellectual property rights; right owners; AIMGP; the Anti-unfair Competition Law; data acquisition; machine learning

CLC： D 923                 DC： A                    Article ID：2096-9783（2021）02-0117-11

1 Creative Text Analysis of Artificial Intelligence Machine Generated Products （AIMGP）

1.1 Preview of Legal Background of AIMGP by Machines and Agents

Driven by the advent of frontier technology including big data technology， cloud computing and machine learning， artificial intelligence technology continues to make breakthroughs on the margins in today's digital era. Without participation and involvement from human subjects， artificial intelligence technology can automatically generate texts for references， uses and dissemination.

In general， although machines have made creations of "products" in a more scalable manner， today's society needs to be more cautious of generated results， as machines strictly follow computer controlled programs such as coded algorithms， parameters and formulas. Machines do not possess necessary morality， ethics， and do not have concerns about compliance， as well as legal liability associated with such creations. One may argue that machines are unable to circumvent key legal "loopholes" as some smarter individuals do. However， the probability of "direct/indirect damages" brought forth by machines and agents is obviously a risk to the author. Such examples as unmanned self-driving vehicles have raised eyebrows for policymakers to cautiously move forward. As we all know， there are so many moving parts in a manufactured car such that should there be accidents during actual usage process， recognition of such damages would be very difficult[1]. Obviously， we place heavy blames on car manufacturers and operators， as they are major stakeholders in those occasions， including producers of upstream and downstream suppliers such as LiDAR （Light Laser Detection and Ranging） and map and data providers.

1.2 "Creating" Subject of Artificial Intelligence Products

Article 3 of "The Regulations for the Implementation of the Copyright Law of the People's Republic of China" approved by the State Council stipulates that the creation in the copyright law refers to intellectual activities that directly produce literary， artistic and scientific works. Organizing work for others to create， providing advisory opinions and material conditions， or performing other auxiliary work is not regarded as a "creation". To clarify the primary subject matter of "creations" of AIMGP， we need to understand what roles human beings and machines play in the output of AIMGP. From the perspective of the creating process of artificial intelligence products[2]， human beings have two main functions to exert impact： first， as initiators and acceptors， artificial intelligence "creations" require human operators to release instructions， since machines do not have the awareness and needs to actively initiate the "creating" process. Human beings invest a large lump sum of resources including labor， finances and hardware/software development to continuously optimize computing power and train algorithms of artificial intelligence， and "feed" big data into artificial intelligence models in order to analyze big data more efficiently and obtain the most effective information， thereby inspiring orders of magnitude of ways for artificial intelligence machines and agents to "create"[3]. However， in today's era of information explosion， only effective or novel information can attract the attention of interested audience and bring traffic to publishers. Therefore， it is particularly critical that AIMGP choose what specific contents to create and publish. Upon completion of the software program， the user then needs to examine the quality of AIMGP to decide on whether to publish and release those generated contents to the public audience. The final content presented to the public is the result of meticulous editing and selections; second， software developers， software engineers， and algorithm scientists conduct a series of preliminary model training processes on artificial intelligence technology. Artificial intelligence powered poetry composition or painting will gradually become indistinguishable from human works， and even the finest artist might not match the masterpieces. In the collaboration between the Central Academy of Fine Arts and the JDAI Research Institute， algorithm engineers collected large training data samples from different artistic styles and trained deep learning models （Convolutional Neural Network） to recreate masterpieces of great artist， stroke by stroke. These completed results are not able to be achieved overnight and such fine-tuning requires software developers to repeatedly debug artificial intelligence algorithm models for a considerable number of repetitions. The author also observes that large training samples of fine art masterpieces contributed by the CAFA also help a great deal in creating better models[4]. However， it should be duly noted that such training processed require many iterations to take place. Take another example， Microsoft's Little Ice has performed tens of thousands of iterative learning processes， driving changes from confusing sentences to beautiful language as repeated training by algorithm scientists. After algorithm scientists and software developers input big training data samples into artificial intelligence programs， artificial intelligence agents then apply deep learning technology to find patterns in data， imitate contents of input data and parameters， and generate expressions different from human works through natural language understanding and natural language generation technology. Artificial intelligence agents and machines are direct producers of AIMGP and final arithmetic expressions of contents.

Based on the above analysis， the "creating" process of artificial intelligence is completed by artificial intelligence alone. It is the main subject matter of "creations". Human beings do not participate in the "creating" process at all， and only do some auxiliary work in the process. At present， although such auxiliary work is essential to the final production stage and some middling stages of AIMGP， human involvement and participation are likely to be minimize as the latest technology becomes mature to make these participations obsolete.

2 Dilemma of Chinese Current Laws on AIMGP Protection

2.1 Dilemma of Copyright Law Protection under the Current Legal Framework

AIMGP share similar design traits with generic human produced works， and this particular phenomenon stimulates the scholarly work on copyright issues as to whether those products can be protected by copyright legislations. However， possessing the design similarity only shows that general copyright legal practices can be used as references and guidelines to protect artificial intelligence generated products， and it should not be inferred that such practices apply to the copyright legal protection framework under common perceptions. AIMGP are churned out by artificial intelligence agents and machines through function-based and parameter-based models， which do not reflect unique judgments and choices of software developers and algorithm scientists; they are not by nature original work directly created by developers and scientists， but rather set parameters and fixed data sets in trained models. Artificial intelligence does not have creative intentions， but passively accepts instructions and produces results. Such machine-led procedures do not take the lead of "author roles" in the traditional realm of copyright[5]. In such cases that artificial intelligence machines be treated as authors from the law technology perspective， civil subjects and corresponding authors are still under careful scrutiny. Therefore， AIMGP cannot be recognized as works and be protected by the copyright law. From the perspective of copyright legislations， the copyright law is established to protect creative and talented masterpieces by intellectual creators， inspire human beings on creativity and promote thought processes. Incorporating AIMGP into copyright law protections and framework not only fails to instigate the creativity， but also induces inertia and discourages creative activities by human involvements. Direct adoptions of "copinism" without any corresponding compatible adjustments will cause disruptions and distortions in the existing copyright systems， and in the meantime will cause breakdowns in the legislative framework[6].

Copyright law in the narrow sense ceases to provide protections for AIMGP， and some scholars advocate that it be incorporated in neighboring rights protections. Neighboring rights are established to protect interests of corresponding copyright owners and distributors. The original intention of such configurations is to boost and encourage disseminations of intellectual property works. Neighboring rights protections are shared openly such that every production entity can configure individualized neighboring right attributes based on technological developments and public policies. Although this viewpoint is mainstream for domestic scholars， the author remains skeptical on such claims. First， neighboring rights are also facing impacts of artificial intelligence. For example， performance right for virtual singer Luo Tianyi's stage presence is still under debate as to whether the singer herself or itself has ownership claims to "her" or its acts， and adjustments to neighboring rights are still pending reviews. Second， neighboring rights only protect dissemination of work， but not originations of work， yet artificial intelligence machine generated work reforms content creations from the production side. Third， neighboring rights protect products that lack sufficient originality， that is， originality requirements cannot be simply satisfied by merely focusing on appearance designs. However， appearance designs of AIMGP are no different from human produced works， but they at least need to satisfy the objective conditions of originality in expressions. Therefore， the author believes that it is unreasonable to incorporate AIMGP in realms of the neighboring rights protection.

2.2 Obstacles to the Protection of Property Law

Protected objects of intellectual property protection are aggregations of human intelligence， and AIMGP are machine generated products that cannot constitute primary subjects. Therefore， there are propositions that perceive AIMGP as fructus and apply relevant regulations on fructus through the law of property protection[7]. As such propositions indicate， fructus can be treated as the derivative of original works produced and becomes separated from original works. From the perspective of relationships between original works and fructus， AIMGP are able to be disseminated independently， and become separated from artificial intelligence agents and machines， or in other words， the "creators". Products created by agents or machines can also be regarded as derivatives of the author， and this relationship cannot form the grounds for AIMGP to be used as fructus protection. There are the following institutional obstacles to using artificial intelligence products as fructus protection： first， AIMGP cannot be included in the existing fructus type. Technically speaking， China divides fructus into natural fructus and legal fructus. AIMGP are neither produced from the natural order， nor are they caused by legal provisions， but are results of artificial intelligence machine productions and cannot be included in the fructus protection. Second， the rights ownership system does not match. Fructus are economic benefits， generally collected by individuals who are in the possession of original work. Investors own the software portion of AIMGP， while users own the hardware portion; hence it is controversial as to which parties claim such benefits. Fructus is enjoyed by the owner or the usufructuary， and this article argues that rights of AIMGP should belong to users without stipulation. Third， the Chinese law dictates special regulations for objects in different forms; for example， property law stipulates protections on tangible objects， whereas the intellectual property law dictates protection on intangible objects. Due to the scarcity of tangible objects， possession of such objects is either by one or the other. Property law protects property owners through exclusive possession rights to gain possession interests of property owner protections. As for issues on intangible objects， once an intangible object is created， the scarcity ceases to exist afterwards， which means it can be widely disseminated and occupied by multiple subjects at the same time. For this particular reason， the intellectual property law sets exclusivity rights for rights holder with exclusive interests. AIMGP are in essence intangible objects. Their nature and characteristics should not be included in the scope of property law protection; otherwise it could lead to chaos in the legal system.

3 Comparative Research on Legal Protection Model of AIMGP

3.1 Protection Path of Overseas AIMGP

The protection of AIMGP has become a worldwide problem. The first nation to legislate on AIMGP was the United Kingdom. In the United Kingdom， the law of "Copyright， Designs and Patents Act" explicitly stipulates regulations on legal nature， ownership rights and protection rules of AIMGP. The law stipulates computer generated results as works in the realms of copyright works， in which case ownership rights of said author should belong to "the person by whom the arrangements necessary for creation of the works are undertaken"[8]. In terms of specific protection rules， the law excludes spiritual rights from owners on AIMGP， and the protection period is set to be 50 years upon completing the creations. Australia， New Zealand， South Africa and other countries have drawn on the experience from the British legislations. For example， copyright laws of South Africa and New Zealand stipulate that the right belongs to "the person who provides necessary preparations for creation of the work"1 ， and the Australian law grants it to the person who operates software programs and applications， and is legally responsible for creating the program， commonly known as software developers and algorithm scientists.

The United States has different legislative approaches from the United Kingdom to include artificial intelligence products in the scope of machine generated product protections. The United States law clearly stipulates in the "U.S. Copyright Office Implementation Plan" that copyright protected works be created by human beings， whereas completed works or products created by agents like animals or machines should not be filed or registered.

Similar to the current legislative framework of the United States， Japan also does not recognize the copyright protection of AIMGP. In the "Intellectual Property Strategic Program 2016" approved by the Cabinet Office， policymakers and law enforcers proposed to draw from the legislative experience of trademark laws， establish new legislations to register， and protect AIMGP， or modify anti-unfair competition laws to prevent the abuse of software programs. In terms of specific rule setting， objects seeking protection should bear a certain amount of economic values， and those that can be easily acquired by simple operations would not be able to become protected. Rights should be assigned to software developers or investors， and right holders may request damages for unauthorized uses by other potential customers or users. In addition， Japan legislations also provide exemptions for AIMGP to use other people's works. Since AIMGP are based on large amounts of data analysis and training， data acquisition is particularly critical to the successful development of artificial intelligence software programs. Therefore， reducing difficulty of acquiring large training data samples can substantially abate general costs of AIMGP， which in turn can improve the production efficiency and the quality of such products created.

It could be conclude that there are two major generic modes of overseas protections for AIMGP： one mode is to transfer such rights from products to human subjects， so as to incorporate them in the protection scope of the copyright law， such as the United Kingdom; the other mode is to deny the ability to apply for copyright of artificial intelligence products， seek new ways of protection outside of the generic copyright scope， and establish special legislations， such as Japan. When drawing from the overseas legislative experience， it is suggests that China should combine protection models with legislative rules that are most suitable for the current development of Chinese cultural market under the current political and economic environment.

With the UK as a forerunner example， in the latest consultation outcome from "artificial intelligence call for views： copyright and related rights2"， the Intellectual Property Office of the UK stipulates that government's objectives for copyright and AI should resolve three key issues： the use of copyright works and data by AI systems， whether copyright exists in works created by AI， and who it belongs to; copyright protection for AI software. The simplest form of AI copyright is the use of license， which provides foundational work to the IP legal framework. There is also a specific right to the database that protects data from obtaining the contents within. Similar to China， if the ultimate purpose is for non-commercial research， the law will become more lenient.

What's in store for Chinese policymakers？ UK's law emphasizes the establishments of clear divide between commercial purposes and non-commercial research， and subsequently lashing out exemptions and exceptions following the regulations. Interestingly enough， British policymakers have made exemptions for neural network storing permanent data inside the system. Such cuts may have provided guidelines in one form for related Chinese counterpart laws.

As for AIMGP， it is interesting that as opposed to other countries， UK laws provide protections for such products which do not have human as creators or authors who are aforementioned "the person by whom the arrangements necessary for the creation of the work are undertaken".

The idea of protecting works generated by artificial intelligence is challenged and controversial as to whether generated products should all be protected. From a technical perspective， artificial intelligence agents have no consciousness or mindset during the creation process， therefore the fundamental concept of originality is questioned， given that those machines store permanent data and contents generated by others. As the new institution and foundation is challenged， the concept of originality has also transitioned， from the old "Copyright， Designs and Patents Act"3， to "the author's own intellectual creation"4. Hence， it would be challenging to recognize the personalities associated with machine productions.

Here， we can conclude from the UK example that license and database copyright is fundamental to the copyright law and should serve as basis Chinese policymakers can rely on.

3.2 Protection Route Selection of Chinese AIMGP

The Anti-unfair Competition Law aims to maintain and regulate the market competition fairness， and protect the legitimate interest of business owners. The Intellectual Property Law aims to encourage product innovations， protect the intellectual work of rights holders， and promote the prosperity and development of cultural undertakings. Intellectual property law and the Anti-unfair Competition Law have deep and profound relationships. Originated as the supplement to Intellectual Property and Trademark laws， the anti-unfair Competition Law adds extra layers of protections. The German Act against Unfair Competition （Gesetz gegen den unlauteren Wettbewerb， UWG） stipulates that "for results that cannot be specially protected by the copyright law or the industrial property law， opportunities to be protected by anti-unfair competition laws should exist". International treaties such as the "Paris Convention for the Protection of Industrial Property Rights" and the "The Agreement on Trade-Related Aspects of Intellectual Property Rights" （TRIPS Agreement） also stipulate laws that regulate the anti-unfair competition. The Anti-unfair Competition Law gives right holders rights to prohibit abuses by other potential interested software developers. Their shared functions suggest that they also share the same protection goals.

A complete and organized intellectual property system is conducive to orderly operations of the Chinese cultural market. Since the copyright law carries significant economic values， it will in turn trigger and push forward the relevant market competition. While the copyright law protects rights of authors， it also maintains fair competition orders in the cultural market. From the perspective of legal applicable cases， these two legislations are the relationship between special laws and general laws. The Intellectual Property Law is a specialized law， and the Anti-unfair Competition Law provides the supplementary protection for the IP law[9]. These two laws will also at times contradict， and subject of rights can freely choose which laws to apply for protection of their rights. The second chapter of the Anti-unfair Competition Law enumerates specific acts of the unfair competition. Regarding unstipulated unfair competition acts， the Anti-unfair Competition Law sets up Article 25 as general clauses to regulate those acts. When we apply general terms， three requirements should generally be satisfied： first， the current law framework does not provide stipulations for the act; second， such acts bear harms to legitimate rights and interests of other business owners; third， the act violates principles of good faiths and business ethics. Since principles of good faiths and business ethics are highly abstract and ethical， their evaluation criteria is to be entirely based on reasonable standards within the reasonable scope. The Internet economy has given birth to many emerging types of competitive behaviors， and business protocols in those emerging areas have not yet reached consensus， which poses challenges to the enforcement and application of Article 2.

In the case of unfair competition disputes between Beijing Baidu Netcom Science & Technology Co.， Ltd. and Shanghai Hantao Information Consulting Co.， Ltd.6， the solicitor believed that violations of business ethics should be considered from the following aspects： first， such behavior should exert positive effects. The market economy encourages the competition on efficiency， rather than improving the competitive strength by hindering others from competing or， in the worst case scenario， pushing competitors out of the market. After a series of initiatives in the Fourth Industrial Revolution， China's State Administration for Market Regulation （"SAMR"） issued the draft of the Guidelines for Anti-monopoly in the Platform Economy （"Guidelines"） in order to prevent the monopolistic behavior in the platform economy， guide business owners and operators to engage in compliance with laws and regulations， and promote the sustainable and healthy development of the online economy. As the name indicates， this guideline is directed towards internet moguls which benefit from population dividends， grow in massive scales as a result and at present extend their boundaries outside the regular scope of business. In the case in which business owners （especially for platform-based companies） completely steal others' business results and provide homogeneous services， such a predicament will not help encourage business innovation and promote market competition， and is not in line with business ethics. Such theft acts have become increasingly rampant from the previous model-driven economy to the technology-driven economy.

Some scholars argue that business models can be drawn upon more easily than technology advances; however， the "stealing" act of business models is not easily quantifiable or detected versus stealing computer programs. In most cases in the model-driven economy， stealing acts can be dubbed as "drawing experience" and merely signify violations of business ethics in most extreme cases. However， in today's technology-driven economy， stealing acts indicate violations of software or hardware program infringement and oftentimes will lead directly to the criminal offense. Second， such unfair competitive behavior exceeds the necessary ethical limit. Business owners should follow the principle of minimization and necessity when taking advantages of others' business results; third， information beyond the necessary limit exerts impacts on market orders. This case provides ideas and guidelines for whether AIMGP can be protected by the Anti-unfair Competition Law. One exception which should be encouraged and promoted by relevant bureaus is the use of open-sourced software APIs and programs. The author recognizes collaborative efforts by BAAI， Chinese Academy of Sciences and many leading internet and technology companies as nationwide efforts to establish the artificial intelligence and big data infrastructure， thereby reducing possible disputes or arbitrations， since open-sourced software programs can be fully utilized without the possible infringement risk.

AIMGPs are essentially information-based products in market distributions， signifying a faster distribution rate than tangible objects and greater economic values than business models in the creating process. Under the current legal framework， legal nature and attribution of rights have not been stipulated， and there are no special laws to protect them[10]. This will inevitably cause market entities to compete for interests of AIMGP， thereby affecting regular market orders and falling into regulated scope of the Anti-unfair Competition Law. AIMGP can meet people's needs for information use， and can bring user traffic and reputation. For the acquisition of AIMGP， all parties have devoted efforts and contributions to the creating process. Software developers devote significant amounts of efforts to writing software programs and repeatedly training artificial intelligence deep learning models. Technology dedicated investors invest a considerable amount of R&D expenses and associated labor costs into developing software and hardware programs， whereas users pay corresponding considerations， are involved in production progress of creating AIMGP and by nature bear the possible infringement risk. For commercial model driven economies， if market subjects arbitrarily abuse the use of AIMGP on their own accounts without permissions of the rights owner， they will compete with the rights owner for fans and internet traffic， thereby reducing the viewership and alleviating the influence of rights owner， diverting traffic of the target groups to other parties， and making the aforementioned labor uncompensated so as to incur losses， especially for those platform-based companies with massive amount of traffic.

At this current stage， leveraging the Anti-unfair Competition Law to protect AIMGP has the following advantages： First， the development of artificial intelligence is currently in an emerging stage. As for how this ongoing development process will unfold in the future， the process is contingent on real world practices to regulate recurring issues. Legislations must not supersede the role that society plays. Therefore， current applications of the Anti-unfair Competition Laws to protect AIMGP not only save costs to establish relevant laws， but also ease the social conflict of all stakeholders involved. Second， to entrust users with appropriate rights can prevent the unfair competition and protect investments contributed by all parties. Third， compared with the absolute protection rule， the Anti-unfair Competition Law provides protections in the broad and weak scale that can promote information free-flow to the greatest extent possible. However， the Anti-unfair Competition Law， as supplementary and miscellaneous provisions to the special law， also has its disadvantages. The Anti-unfair Competition Law is the legalization of moral principles， expressed in generic terms and on the one-verdict-per-case basis. It can easily lead to inconsistent judgments， and may need to be revised and perfected. Moreover， not all the parties disseminating AIMGP are market participants and they cannot cover all bases for protections. In addition， the Anti-unfair Competition Law does not clarify issues such as identification and ownership of AIMGP. Therefore， the law can only be used as temporary protective measures， but not as a long-term solution beyond the transitional stage of future legislations. A more definitive solution needs to be explored.

4 Incorporate AIMGP into Protection of Information Property Rights Theory

Based on the above analysis， the Anti-unfair Competition Law is unable to completely resolve the problem. The timeframe for legislation is still in the maturing stages， and once it matures， it should be included in the protection of information property rights， and analyze the feasibility and rationality of this protection path.

Information property rights， also known as data property rights by academic scholars， were proposed by the Australian scholar Professor Michael Pendleton of University of Canterbury in "Industry and Intellectual Property Rights in Hong Kong" in 1984. In 1987， the well-known Chinese scholar Chengsi Zheng made comprehensive discussions on information property rights in the "Legal Protection of Computers， Software， and Databases" and regarded it as the generic concept of intellectual property rights. In the property rights system， information property rights are new types of property rights， independent of property rights and intellectual property rights. Their theoretical values lie in bringing objects that are not protected by intellectual property rights， and such arrangements should be incorporated into legal system for protection. This should provide complete regulations for their use and ownership. Basic institutional framework of information property rights theory mainly includes the following contents： object of rights is information or data， served as intangibles， and strictly speaking is independent of original personal information in the legal form， with specific functions or values. This information is characterized by being readable， and does not require originality and other conditions. However， the information must satisfy legitimacy of content and access. Subject of rights is the information collector， and any subject can obtain corresponding property rights based on labor or investment. Content of rights mainly includes powers and functional work such as control， use， profit， and dispose. In terms of rights restrictions， compared with intellectual property rights， information property rights is a rights system with lower thresholds. From the general legal point of view， the degree of protection is inferior to that of the intellectual property system.

The protection of AIMGP is consistent with information property rights theory. From the perspective of nature of artificial intelligence products， AIMGP are restricted by the subject and creative factors， which should not be included in copyright protection. However， they are newly emerging intangible objects that can be traded and spread in the market， and have potential commercial values. Information property rights are in fact legal embodiments of information product commercialization. Artificial intelligence machine generated product is essentially an information product， which contains investments contributed by all parties and conforms to object characteristics of information property rights. From the perspective of rights subjects， collectors of AIMGP are users， and granting users property rights can encourage them to produce and collect more valuable AIMGP， and promote the effective use and dissemination of information. In terms of content rights， various capabilities of information property rights can provide the adequate protection for AIMGP. Therefore， the information property rights theory is just applicable to the protection of AIMGP.

There is much commentary in the academic community about the protection model of AIMGP. Incorporation of AIMGP into protective framework of information property rights have unparalleled advantages over other protection paths. First， information property rights have already been enforced in international legislations， which provides Chinese law enforcers with legislative experience. For example， the United States has established rights for electronic information to be acquired， used， and accessed by computers in the "United States Uniform Computer Information Transaction Act" （UCITA）. In the "The Legislation of the Russian Federation on Information， Information Technologies and the Protection of Information"， Russia regards information as new types of civil rights objects， independent of intellectual property and objects， and claims those information rights as ownership. Second， from the perspective of Chinese legislative trends， it leaves room for legislation on the establishment of information property rights. Although China has not clearly established information property rights， the law enforcement bureau has recognized the status of information as subject of civil law and indirectly indicated that it can be protected by another law. Due to the slow maturing timeframe for legislation at this stage， the legal basis of information property rights still has certain controversies. Therefore， the legislation adopts indirect provisions on the institutional basis for the establishment of information property rights in the future. Third， information property rights are more in line with social development trends. Driven by increasingly common use cases of technology and information technology， more types of information products will be produced and be protected legally. Information property rights have certain inclusivity. Focusing on the information itself， but not on the source or on the formation process of the information， we cannot only resolve legal protection issues of AIMGP， but also provide protection for other new types of information products and boost the information flow.

In addition， in terms of the rights restriction， the protection of AIMGP is inferior to that of intellectual property. Since finished works are consolidated masterpieces of human wisdom and embodiments of human dignity and value， we should offer strong protection to them in order to encourage creative geniuses or technology prodigies to fully leverage their creativity and create more masterpieces to be disseminated to the public audience， as manifested in the art community. The author believes that computer programmers can in a way be equivalent to their artist counterparts， and such comparisons have grown more alike in the digital ages， which requires more creativity to avoid homogeneous creations as limited by the human creativity. AIMGP are general information products or commodities produced on assembly lines， which have no spiritual symbols and cannot be compared with intellectual achievements full of human creativity. However， in terms of special configurations of specific rights， we can learn from the protection rule of the copyright law， make full use of the copyright legislation experience， and reduce the incompatibility of new laws in real practices. Since use cases of artificial intelligence products are still in their infancy phase， future directions need to be tested and observed in practices. If regulations are established in an overly stringent manners， the law will take precedence over the society， which may lead to the wrong guidance; hence this article will not be too specific in terms of rules setting. The rules herein are set so as not to become obstacles to future evolutions of AIMGP， but to merely propose ideas on systems from the macro standpoint.

5 Conclusion

With the rapid development of science and technology， the digitization and intelligentization of artificial intelligence technology is bound to greatly improve， and once such technology matures it will reach or even surpass achievements by human beings. Therefore， early discussions on the protection model of AIMGP will not only help the development of artificial intelligence technology industry， but also meet people's needs for information use， promote the prosperity of cultural market and secure the stable operation of the economy. Without disclosing the source of the subject， AIMGP cannot be distinguished from human works， but there is a quantifiable difference between the creating processes of the two. The work is the intellectual achievement of human beings' full use of creativity， reflecting the author's individual judgment and analysis， building new products from scratch， and demonstrating human dignity and spiritual beliefs. However， AIMGP are structured and patterned （mechanized） under artificially run operations of data models and algorithmic rules. Their content is merely the arrangement and combination of existing data without qualitative changes， such as mass produced goods on assembly lines， and generated products are machines controlled by humans. Therefore， it is not original intellectual achievement and cannot be protected by the copyright law. However， artificial intelligence products bear a certain amount of economic values， and they are manifestations of intellectual labor from software developers and algorithm scientists， committed investments from investors， and dedications from software program users in labor， materials， and sweats.

In light of the recent artificial intelligence development， commercial values of the technology have not yet reached their peak points. Corresponding technology capabilities （by the names of software packages or sets of models/APIs） should not be attributed to public domains， but should be attributed to specific subjects. Since users have intentions of applying artificial intelligence technology to generate， use and disseminate AIMGP， and acting as enforcers to directly control and supervise over AIMGP， they bear the infringement risk， so it is most reasonable to attribute the rights to users. Without any protective legal clauses， rights will cease to exist. Therefore， reasonable and systematic guarantees are essential for the development of AIMGP. As the technology continuously develops and creates more commercial values， law enforcements will gradually establish the robust legal system to protect rights of contributors.

Reference：

[1] See Executive Office of the President National Science and Technology Council National Science and Technology Council Committee on Technology， Preparing for the Future of Artificial Intelligence， October 12， 2016， Washington， D.C. 20502， p.8.

[2] James G. There's No Such Thing as a Computer-authored Work - And It's a Good Thing， Too， 39 Colum. J.L. & Arts， 2016， 403.

[3] Annemarie B. Coding Creativity： Copyright and the Artificially Intelligent Author[J]. Stan. Tech. L. Rev. 5， 2012.

[4] Robin B. The 7 Myths of AI[DB/OL].[ 2017-5-14]. https：//www.crowdflower.com/7-myths-ai/.

[5] Thomas K. European Union Policy-Making on Robotics and Artificial Intelligence： Selected Issues[J]. Croatian Yearbook of European Law and Policy， 2017， 13.

[6] Annemarie B. Coding Creativity： Copyright and the Artificially Intelligent Author[J]. Stanford Technology Law Review， 2012， 20.

[7] Burkhard S.， Editorial： The Future of IP Law in an Age of Artificial Intelligence[J]. A Journal of Law， Technology and Society， 2016， 13.

[8] Gonenc G.， Ilay Y.， Gunes H. Stifling Artificial Intelligence： Human perils[J]. Computer Law & Security Review， 2016， 32.

[9] Andres G. Artificial Intelligence and Copyright[J]. World Intellectual Property Organization Magazine， 2017， 4.

[10] Julio R.， Sacha W-V.， Andrew K. Breakthrough Technologies – Robotics and IP[J]. World Intellectual Property Organization Magazine， 2016， 6.

摘    要：目前，法律對人工智能生成物的保护并无明确规定。由于缺少法律保护机制，人工智能生成物承载的经济价值可能会引发各方主体无序使用，产生纠纷，最终影响文化市场的健康发展和良好运行。因此及早为人工智能生成物找出合理的保护路径，不仅可以起到定分止争的效果，还可以促进人工智能生成物的生产和传播，满足人们日益增长的文化需求。

关键词：知识产权;权利所有人;人工智能生成物;反不正当竞争法;数据获取;机器学习