Basic Architecture and Patentability Analysis of Artificial Intelligence Technology

Liu Xin

Abstract： Artificial intelligence， as one of the most advanced technology types， encompasses a wide range of tech- nical contents， including arithmetic designing， data integration， speech recognition， image processing， and so on. Its basic architecture could be classified into four levels： basic level， perception level， cognitive level， and appli- cation level. Different levels of technical characteristics and functional forms lead to different conclusions for pat- entability. For the purpose of realising full patent protection of artificial intelligence technology， a systematic analysis of the basic architecture of artificial intelligence technology is required， as well as the analysis of the pat- entability of all kinds of relevant artificial intelligence technologies based on the regulations of China’s patent law. The algorithm at the basic level is usually seen as natural law or abstract concept， which is excluded from the objects of patent protection under the current patent regulations. The technologies in perception and cognitive levels are created by combining algorithmic models and specific hardware or application scenarios. Only if the technologies are integrated with the algorithmic model and improved themselves， then the technologies could ful- fil the creative conditions of China’s patent law. The patentability of final productions at the application level is dependent on whether the nature and structure of the products comply with the basic requirements of patent law in China.

Keywords： artificial intelligence； technology architecture； patentability； algorithmic model； technical solution

CLC：D 923DC：AArticle ID：2096-9783（2022）01-0118-09

1 Introduction

Artificial intelligence is not only a tool that assists humans to make intellectual creations but also has the abili- ty to produce creations independently. For instance， IBM’s "Watson"， an artificial intelligence， can already create new recipes based on the diner’s preferences[1]. The "ROSS"， an artificial intelligence lawyer in a New York law firm， can read the law， gather evidence， make inferences， and provide a highly relevant answer based on evidence[2]. Dr. John Koza had obtained several patents in 2015， on the creations generated by Genetic Programming （a kind of artificial intelligence for making inventions） without direct human intervention[3]. Undoubtedly， the advancement of artificial intelligence is beyond people’s imagination， and it may make the scenarios depicted in science fiction films come true someday.

The rapid advancement of artificial intelligence technology has caused disruptive changes in human society’s lifestyles.Autonomy is a key attribute of artificial intelligence， making it different from traditional technologies[4]. Autonomy makes artificial intelligence have the ability to replace human beings in the completion of difficult work at the same time. Artificial intelligence is now found in all walks of life from agricultural production， industrial manufacturing to finance and insurance， advertising media， and medical diagnosis. Meanwhile， as artificial intelli- gence technology becomes more widely used， the patent regulations of artificial intelligence technology have emerged. Nevertheless， this issue has not been given due attention. Therefore， it is required to clarify the patent reg- ulations of artificial intelligence technology by investigating the patentability of corresponding technical solutions from the characteristics of artificial intelligence technology.

At present， the research on artificial intelligence-related patent issues is primarily concentrated on the legal subject qualification of artificial intelligence and the patent protection of creations generated by artificial intelli- gence， but the most fundamental issues of patent protection of artificial intelligence technology are rarely men- tioned. Some scholars respond positively regarding the legal subject qualification of artificial intelligence. Colin R. Davies believes that artificial intelligence should be given legal subject qualification through learning from histori- cal reasons why legal persons are endowed with legal personality； in particular， there already exist suggestions for giving "special legal status" or "electronic identity" to artificial intelligence in EU parliamentary proposals on artifi- cial intelligence[5]. Yuan Zen， a Chinese scholar， also holds an opinion that artificial intelligence could be endowed with legal subject qualification for its independent capacity， but that this legal subject qualification of artificial in- telligence should be imposed restrictions， because of its limited ability for bearing the consequences of its behav- iours[6]. Wu Handong also claims that the "intelligence" of artificial intelligence is comparable to that of humans and that artificial intelligence’s indecent willingness differs from that of a natural person and legal person who is a group of natural persons[7]. Xiong Qi was seriously opposed to artificial intelligence being given legal subject qualifi- cation based on the principle of private law that the subject of right and the object of right are not interchange- able[8]. Annemarie Bridy similarly believes that the author can only be a combination of natural persons and that ar- tificial intelligence has no possibility of becoming the right owner[9].

In accordance with the patent protection of creations generated by artificial intelligence， scholars express their viewpoints one after another. In the opinion of Ben Hattenbach and Joshua Glucoft， mechanically generated claims could not only be published as prior art but also technically be granted patent rights under our current laws[10]. From the standpoint of policy and law， Erica Fraser gives the same answer[11].According to Zhu Xuezhong and Zhang Guangwei， the transfer doctrine on traditional subject matter rule in civil law will not be abandoned by pat- ent law under the current system[12].Ji Dongmei holds a prudent view. She argues that based on traditional patent grant conditions， the definition of patent scope should not be rigidly uniform. In order to achieve the integration and balance between the stability and flexibility of the patent law system， it should be adjusted with the develop- ment， contribution， popularity， and other specific conditions[13].Moreover， Yi Jiming appeals to take "Sweat of the Brow" as the primary doctrine and build an objective standard for checking the originality of the creations generat- ed by artificial intelligence[14].While， Liang Zhiwen believes that entering the public domain may be a proper ar- rangement for the ownership of creations made by artificial intelligence in the light of the ultimate goal of intellectu- al property law， which is to protect the public interest[15].

Artificial intelligence， as the most advanced technology kind at the moment， entails algorithm design， data inte- gration， speech recognition， image processing， and many other technical contents. In contrast to traditional technolo- gies， artificial intelligence is undoubtedly more complex in technical structure. Artificial intelligence is no longer a single technical solution in a specific technical field， but rather a comprehensive technical solution made up of a variety of technologies in different technical fields and forms. Thus， unlike traditional technologies， the patent pro- tection of artificial intelligence technology should not directly judge the correlation between technical characteris- tics and legal standards. Rather， we should first decompose the comprehensive artificial intelligence technology into single technical solutions， followed by the assessment of the patentability of these single technical solutions. This approach is based not only on the complexity of artificial intelligence technology itself but also on the choice of arti- ficial intelligence technology’s cooperative operation mode[16]. In this regard， for realising the full patent protection of artificial intelligence technology， we should thoroughly analyse the basic architecture of artificial intelligence technology， followed by the analysis of patentability of a single technical solution in various fields and forms that ar- tificial intelligence technology encompasses.

2 Basic Technical Framework of Artificial Intelligence and Its Patentability Challenges

The operation of artificial intelligence technology is based on algorithm driving and data guarantee. Through voice processing， image recognition， and other technologies， artificial intelligence completes reality perception and self-cognition in order to achieve high-level technical goals such as "automatic driving" and "intelligent manage- ment" as well as create corresponding artificial intelligence products. Currently， artificial intelligence technology is rapidly developing and being applied in areas such as automatic driving， natural language processing， image recog- nition， and intelligent robot[17]. In terms of technical architecture， artificial intelligence technology is not a single technical solution comprising only a specific technical field， but a comprehensive technical bundle on the basis of the technical algorithm. Artificial intelligence technology is driven by "big data" and "big computing"， and it is merged with a variety of single technical solutions in multiple technical fields and different functional dimensions. Thus， based on the functional forms of various single technical solutions in the operation of artificial intelligence technology， the basic architecture of artificial intelligence technology can be roughly classified into four levels： ba- sic level， perception level， cognitive level， and application level， just as it is presented in Figure 1 below：

The multi-level basic architecture of artificial intelligence technology is determined by its technical operation mode， and it is also a prominent manifestation of its technical complexity. Firstly， at the basic level， artificial intelli- gence technology demonstrates the basic operation mode of big computing driven and big data guarantee with the al- gorithm as the core. Secondly， artificial intelligence technology includes voice technology， image technology， video technology， AR/VR as the representative of augmented reality technology， and a variety of other technical fields at the perception level. Thirdly， at the cognitive level， artificial intelligence technology consists of natural language processing， knowledge mapping， user portraits， and other technology forms with machine self-learning as its core. Lastly， artificial intelligence technology is primarily reflected in the commercial operation mode of new products at the application level， such as automatic driving， intelligent software， and so on. The technology level division de- picts the infrastructure and operation mode of artificial intelligence technology relatively comprehensively and vivid- ly from the basic level， perception level， cognition level to product level.

Nevertheless， under the framework of the patent system， the patentability judgment of specific technology is frequently dependent on a specific technical solution， which could not be utilised to analyse the patentability of a multi-level comprehensive technical solution， for instance， artificial intelligence technology. Meanwhile， under the existing patent application mechanism， the principle of singleness is also strictly in compliance with the require- ments of "one invention one application" based on technological distinctions， limiting the specific form of patent technology at the form level[18]. This is undoubtedly a challenge for such a multi-level comprehensive technical solu- tion as artificial intelligence technology. We must classify the complex artificial intelligence technology in accor- dance with its basic technical architecture， file a patent application with a single technical solution， and judge the patentability. Therefore， for the purpose of completely investigating the patentability of artificial intelligence technol- ogy； it is critical to judge the patentability of specific technology solutions at the basic， perception， cognition， and application levels from the various manifestations of different technology levels.

Patentability judgments will likely have varied legal ramifications and draw various conclusions based on the diverse forms of artificial intelligence technology at different technical levels. At the basic level， the algorithm， as the bottom design of artificial intelligence technology and the operation guide of artificial intelligence technology， di- rectly identifies the attributes and functions of the final product. However， it will inevitably be recognised as "the rules and methods of intellectual activities" in the judgment of patentability， and then become the object type that is not protected by the patent system. Technologies in perception and cognitive levels as basic tools to cooperate with the algorithm to play its role， by combining the voice， image， and other sensing technology with the algorithm to form a new technical solution， which is reduced by the algorithm and serves the algorithm simultaneously. Al- though these technologies are not directly excluded from the scope of protection in patentability judgment， making an extra non-obviousness investigation on whether the technical solution is easily performed by the general techni- cal personnel is crucial. Following the further development and integration of perception level and cognitive level technology， the application-level technologies must be assessed regarding if they belong to the patent law protect object while judging their patentability. In the case of the patent object’s formal requirement being met， further judging whether the required technical features overlap with the corresponding technical solutions in the perception and cognitive levels would become necessary.

Hence， following the multi-level comprehensive technical framework of artificial intelligence， the patentability judgment of corresponding achievements should be conducted in varying levels； ranging from the basic level to the perception level， cognitive level， and then to the product level. As per the patentability judgment rules， the type of objects without patent rights have been excluded from being the essence judgment of the novelty， creativity， and practicability of technical solutions[19].The patentability of technological achievements in different technology levels must be assessed respectively in order to acquire a comprehensive understanding of the artificial intelligence tech- nology patentability.

3 Patentability of Related Algorithms in Basic Level of Artificial Intelligence Technology

The algorithm is taken as a primary content under the basic level of artificial intelligence technology， which not only forms the basis of the operation of artificial intelligence technology but also its essence. Nevertheless， re- garding patentability， algorithms generally fail to come under the object category protected per prevailing patent law in China. The patentable objects are limited under the patent law of every country. The patent authorisation of ab- stract concepts or rules of intellectual activities is excluded from these limitations. The algorithm itself is an ab- stract concept and theory along with being an extension of the rules and methods of human intellectual activities[20]. Thus， algorithms are often regarded as natural laws or abstract concepts and hence excluded from the protection of patent law[21].

However， analysing the patent system development in the past half-century， the restraints of the patent law on the object of protection are weakening to adapt to the industrial and technological changes. From gene technology patent to business method patent， the patentable object’s limitation under patent law has been gradually shaken. The industrial demand for patent protection of gene technology has been brought by the fourth scientific and techno- logical revolution since the 1970s. The emergence of gene patents has changed the strict restriction that natural dis- covery belongs to natural law and has no patentability. Starting from the case Diamond v. Chakrabarty in 1980， the patent protection on gene technology could be acquired as long as the gene sequence varies from the natural state. This even rendered the human gene as a part of the human body， which has been included in the subject of patent- ability[22]. Simultaneously， the rapid development of network technology and e-commerce has triggered a social ap- peal for patent protection of business methods. Following the case of State Street Bank & Trust Co. v. Signature Fi- nancial Group recognising the business methods patentability in 1998 led to the long-standing "business method exception" in the patent system losing its effect instantly. Although the Federal Circuit Court introduced the patent"machine or conversion" test standard for commercial methods in the Bilski case in 2008， making the patent au- thorisation standard of business method stricter， it fails to alter the conclusion that business methods are patent- able[23]. Considering this， it inevitably leads to the conjecture that the algorithm would be granted patent protection in the near future. Following the consistent enhancement of the commercialisation level of artificial intelligence technology， algorithms， as the basis and leading of artificial intelligence technology， it might also become a part of the protection scope of patent law based on the interests of related industries； breaking the conventional cognition that algorithms do not have patentability.

In the future， if it is absolute for the algorithm to be included in the patentable object category， we must pay heed to the differences shared by the algorithm as well as the object types such as the technical solutions protected by the current patent system in avoidance of the abuse of algorithm power. The algorithm varied from the technolo-gy， it is non-neutral， its final effect is largely constrained by the data quality， and it even dominates human choice through specific data processing and integration. The reason is that in terms of algorithm， "human" is just data， or a collection of electronic traces， after sorting out the data， all kinds of automatic differentiation， scoring， sorting， and decision-making could be formed[24]. Nevertheless， setting up a reviewing mechanism of artificial intelligence algo- rithm from the essential characteristics of the algorithm is necessary before patent protection[25]in order to eliminate the discipline of the Algorithm and make the algorithm protected by China’s patent law[26].

4 The Patentability of Technical Solutions in the Perception Level and Cogni- tive Level of Artificial Intelligence Technology

The technical solutions in the perception level and cognitive level of artificial intelligence technology is gener- ated through a combination of algorithm model and specific hardware or application scenarios. Particularly， voice technology， image technology， knowledge mapping technology， user portrait technology， and other technical solu- tions in the perception and cognitive levels of artificial intelligence technology could be split into two modes： one is"algorithm model + specific hardware"； the other is "algorithm model + application scenario". While， in the patent- ability judgment， irrespective of the mode used to generate the artificial intelligence technology solution， the assess- ment of it must be made step by step following the patent authorisation rules stipulated by the patent system. First- ly， the judgement must be made as to whether the relevant technical solutions belonging to the object types have been granted no patent rights， such as scientific discovery， rules， and methods of intellectual activities， diagnosis， and treatment methods of diseases. Secondly， the focus must be paid to measuring whether the relevant technical so- lutions cater to the requirement of patent authorisation such as novelty， creativity， and practicability. Following the perspective of the special generation modes of "algorithm model + specific hardware" and "algorithm model + appli- cation scenario" in the perception and cognitive levels of artificial intelligence， the technical solutions produced through the combination of the algorithm with specific hardware or application scenarios are no longer the algorithm itself， and usually fail to belong to the object category which China’s patent law has excluded. Simultaneously， the algorithm extension in specific hardware as well as application scenarios also distinguish them from the current technologies. They are applied to production and life， in order to fulfil the novelty and practicality requirements of patent authorisation. However， whether they could fulfil the creative requirements of patent authorisation， there is a need to conduct a further evaluation as per the degree of fusion between the algorithm and specific hardware or ap- plication scenarios.

Therefore， it can be said that the core of judging the patentability of technical solutions at the perception and cognitive levels of artificial intelligence is whether the necessary technical solutions meet the requirements of pat- ent licensing innovation. The main criterion of patentability judgement in the current patent system is creativity， which is based on the assumption of the rational person of general technical personnel in a specific technical field. In other words， the creativity of the technical solution is measured by the reference standard of whether the general technical personnel experienced with a particular technical field can easily complete the technical solution[27].Hence， the creativity of artificial intelligence perception level and cognitive level technologies must be judged on their ability to be completed by regular technicians. On the basis of the special generation modes of "algorithm model + specific hardware" and "algorithm model + application scenario" of relevant technical solutions， the judg- ment of creativity should consider the general technical personnel’s cognition of the algorithm model and the con- trol of specific hardware or application scenarios. Consequently， it is critical to analyse the creativity of relevant technical solutions respectively using the mode division of "algorithm model + specific hardware" and "algorithm model + application scenario" in the perception and cognitive levels of artificial intelligence， for further refining the patentability standard of technical solutions in artificial intelligence perception and cognitive levels.

Firstly， the specific hardware programs are just carriers of the technologies in the "algorithm model + specific hardware" technology generation mode and have no substantial impact on the patentability of the related technolo- gies. To put it another way， the creativity of the algorithm model is the determinant that determines whether a tech- nology fulfil the requirements of patent licensing. However， If only the well-known old algorithm was realised by hardware program without any creative improvement， this technical solution could be readily performed by ordinary technicians， and would not fulfil the creative conditions of patent authorisation stipulated by China’s patent law[28].Therefore， only by creatively improving the old algorithm to form a new algorithm， along with the hardware imple- mentation of the new algorithm， can we truly develop a technical solution that could not be easily performed by or- dinary technicians and fulfil the creative requirements of patent authorisation.

Secondly， in the "algorithm model + application scenario" technology generation mode， the use of the new algo- rithm could likewise yield technical solutions with patentability. However， in the case of the combination of "old al- gorithm model" and "new application scenario"， the patentability of artificial intelligence generation technology should be thoroughly examined in light of the actual circumstances. Undoubtedly， applying the old algorithm in the new scenario would generate the desired effect， but it is critical to focus on that the effectiveness of the implementa- tion effect is not the same as the creativity of the technical solutions， and the two should not be confused. Thus， when determining the patentability of certain technical solutions， we should consider if a "chemical reaction" oc- curs whether the old algorithm and the new scenario are combined. Also， whether the artificial intelligence generat- ed technical solutions could be easily performed by regular technicians[29]. We can only assume that technological solutions are patentable if the algorithm on which they are based is effectively integrated with the new scenario， and technical solutions are difficult to be easily completed by ordinary technicians. On the contrary， they could not be the objects protected by China’s patent law.

5 The Patentability of Final Products in the Application Level of Artificial In- telligence Technology

Automatic driving， intelligent software， and other artificial intelligence products are all examples of artificial in- telligence application-level technologies. The type and structure of the products determine the patentability of the final products at the application level. If the final product is a new technology solution resulting from the improve- ment or upgrading of the perception level and cognitive level technologies， its patentability should be further exam- ined in accordance with the patent authorisation requirements. It is not required to investigate the patentability of the end product if it is a functional combination of many perception level and cognitive level technical solutions， we must explore patentability of perception level and cognitive level technical solutions. Currently， the majority of artificial intelligence products on the market are created by improving or upgrading the perception and cognitive level technologies， and the operation type of the final products is still the technical solutions generated using "old algorithm model + new application scenario" mode. Therefore， the fundamental issue in determining the patentabili- ty of application-level technologies is whether ordinary technicians could easily complete these technical solutions.

As previously mentioned， one of the most important factors to examine in judging the patentability of these technical solutions generated by integrating the old algorithm model with the new application scenario is if these technical solutions could be easily evaluated by ordinary technicians. While the creativity of these technical solu- tions developed by integrating the old algorithm model with the new application scenario is dependent on the combi- nation degree of the algorithm and the scenario. That is to say， only if the old algorithm and the new scenario are fully integrated， could these newly generated technical solutions be easily completed by ordinary technicians and fulfil the requirements of patent licensing creativity. There would be no patentable technical solution if it were merely a simple superposition of the old algorithm and the new scenario. More than that， we should pay greater at- tention to the impact of whether the algorithm model and application scenarios are fully disclosed on the final re- sults in the process of judging the patentability of these technical solutions generated by combining the old algo- rithm model with the new application scenario. In general， full disclosure of technologies is a prerequisite for patent authorisation. Meanwhile， patent disclosure causes technical spillover， which is a crucial way for the patent system to share technological knowledge and encourage innovation[30].However， as can be illustrated in Figure 2 above， the working process of the artificial intelligence algorithm model is similar to a "black box" that has not been dis- closed. The new application scenario is completed by debugging the experimental environment， and the specific op- eration mode is usually unknown. Even though these technical solutions created by integrating the old algorithm model with the new application scenario have not caused the issue of disclosure in the current patentability judge practice， the complete disclosure of technical information will be required in the future to achieve patentability.

6 Conclusions

In different historical times， the patent system has presented diverse value orientations and legal structures due to distinct technology operation modes[31]. The patentability standards for technical solutions should keep up with the times in the multi-level complex technical framework of artificial intelligence. The general patentability judgement approach can no longer be used to judge the patentability of comprehensive artificial intelligence， which consists of several technologies in different technical fields and technical forms. Instead， we should analyse and ex- amine the patentability of single technical solutions in different fields involved in all levels of artificial intelligence one by one based on the basic technical architecture of artificial intelligence.

Due to differences in technical features and functional forms， patentability judgments vary from the algorithm model at the basic level to the technical solutions in the perception and cognitive levels， following the final product at the application level of artificial intelligence. As demonstrated in Table 1 below， the algorithm model at the ba- sic level is generally deemed not patentable as it is based on the natural law or abstract concept. The technical so- lutions in perception and cognitive levels are dependent on the combination of algorithm models and specific hard- ware or application scenarios， so the patentability of corresponding technical solutions should be judged in accor- dance with the combination. The nature and structure of the products determine the patentability of final products at the application level.

Furthermore， the various forms of artificial intelligence in the four technical levels of basic level， perception level， cognitive level， and application level， not only implies different patentability judgment conclusions， but also has an influence on the effect of industrialisation， commercialisation， and capitalisation in the patent operation of ar- tificial intelligence technology[32]. Artificial intelligence technology’s complex multi-level infrastructure will almost certainly lead to decentralisation of rights ownership as well as aggravate the dependence and restriction between upstream and downstream industries. The different levels of technologies involved in a product are frequently devel- oped by inventors in different countries and dominated by enterprises in various countries， particularly in the cur- rent global value chain with the increasingly refined social division of labour and constantly upgraded industrial co- operation[33].Artificial intelligence technology’s precise technical solutions at the basic level， perception level， cogni- tive level， and application-level are usually in the control of enterprises or inventors of different countries， involv- ing a variety of technical fields and experience ranges. Moreover， driven by economic benefits， the majority of enter- prises would make a large-scale patent layout for one or several core technologies of artificial intelligence for en- hancing market competitiveness and seizing monopoly positions. Intermediate technologies， particularly at the per-ception and cognitive levels， would likely become the focus of the patent war. At the perception level， as the associ- ated technologies are relatively basic and the threshold of the patent application is comparatively low， the fierce pat- ent competition among enterprises often results in a "Patent Thickets" dilemma in the form of a large number of patents[34].On the other hand， at the cognitive level， due to the great difficulty of related technology， few rights hold- ers often control the core patents， which leads to the tragedy of "Anti-commons" in the form of the unilateral mo- nopoly of technologies[35].Therefore， for the purpose of balancing the interests of all parties in the operation of artifi- cial intelligence technology and realise effective and reasonable patent protection of artificial intelligence technolo- gies， we should propose the multiple patent operation strategy of artificial intelligence technology， as well as pro- mote the rational patent layout， alliance patent management， and platform patent transaction of artificial intelligence technology， which start from the patent operation risk of artificial intelligence technologies in different levels， in combination with the development needs and operation practice of enterprises， industries， and society， so as to re- solve the patent operation issues of artificial intelligence technology[36].

References：

[1] ABBOTT R. I think， therefore I invent： creative computers and the future of patent law[J]. Boston College Law Re- view， 2016， 57（4）： 1079-1126.

[2] TURNER K. Meet "Ross" the newly hired legal robot[N]. The Washington Post， 2016-05-16.

[3] JONATHON K. John Koza has built an invention machine[J]. Popular Science， 2006， 268（5）： 66.

[4] SCHERER M U. Regulating artificial intelligence systems： risks， challenges， competencies， and strategies[J]. Harv. J. L. & Tech.， 2016， 29（2）： 363-369.

[5] DAVIES C R. An evolution step in intellectual property rights—artificial intelligence and intellectual property[J]. Computer Law & Security Review， 2011， 27 （1）： 601-619.

[6] YUAN Z. The examination of limited legal personality of artificial intelligence[J]. Oriental Law， 2017， 10（5）： 50-57.

[7] WU H D. Institutional arrangement and legal regulation in the age of artificial intelligence[J]. Legal Science （Jour- nal of Northwest University of political science and law）， 2017， 36（5）： 130-136.

[8] XIONG Q. Is artificial intelligence generated content copyrightable[J]. Intellectual Property， 2017， 31（3）： 3-8.

[9] BRIDY A. Coding creativity： copyright and the artificially intelligent author[J]. Stanford Technology Law Review， 2012， 5（1）： 1-28.

[10] HATTENBACH B， GLUCOFT J. Patents in an era of infinite monkeys and artificial intelligence[J]. Stanford Tech- nology Law Review， 2015， 19（1）： 32-51.

[11] FRASER E. Computers as inventors—legal and policy implications of artificial intelligence on patent law[J]. SCRIPTed： A Journal of Law， Technology and Society， 2016， 13（ 3）： 305-333.

[12] ZHU X Z， ZHANG G W. On the patentability and ownership of technological achievements brought about by artifi- cial intelligence[J]. Journal of Intelligence， 2018， 37（2）： 69-75.

[13] JI D M. On the challenges of patent system aroused by inventive achievements of artificial intelligence[J]. Intellec- tual Property， 2017， 31（11）： 59-66.

[14] YI J M. Is the creation of artificial intelligence a work-[J]. Legal Science （Journal of Northwest University of politi- cal science and law）， 2017， 36（5）： 137-147..

[15] LIANG Z W. On legal protection of creations generated by artificial intelligence[J]. Legal Science （Journal of Northwest University of political science and law）， 2017， 36（5）： 156-165.

[16] LIU X， QIN C X. On problems， challenges and responses of patent law in the era of artificial intelligence[J]. Elec- tronics Intellectual Property， 2021， 20（1）： 73-82.

[17] HE Q. Research on the development of artificial intelligence technology[J]. Modern Science & Technology of Tele- communications， 2016， 23（2）： 18-21.

[18] WANG L H. Patent law[M]. Beijing： Peking University Press， 2007： 82-83.

[19] LI M D. American intellectual property law （2nd edition）[M]. Beijing： Law Press， 2014： 43-68.

[20] GUO S M. Horizontal research on patent application examination standards involving algorithms[J]. Patent Agency， 2016， 2（2）： 37-39.

[21] VERTINSKY L， RICE T M. Thinking about thinking machines： implications of machine inventors for patent law[J]. B. U. J. Sci. & Tech. L.， 2002， 8（2）： 590.

[22] LIU X. On the patentability of human gene—peflections on the myriad case[J]. China Postgraduates， 2016， 15（6）： 29-33.

[23] LIU Y L. Ten year expansion and reincarnation of business method patents in the United States： a historical review from the state street case to the bilski case[J]. Intellectual Property， 2010， 24（6）： 90-92.

[24] ZHENG G. The law of algorithm and the algorithm of law[J]. China Law Review， 2018， 5（2）： 68.

[25] CCID think tank， ZHOU Y. It is urgent to establish artificial intelligence algorithm review mechanism in China[N]. China Computer News， 2018-05-14， 012.

[26] JIANG Y. Discipline of algorithm and the disciplined algorithm： the legal regulation of the algorithm in the AI era[J]. Hebei Law Science， 2018， 36（12）： 142-153.

[27] ZENGCMR.Chinese，EuropeanandAmericanpatentlaw[M].Beijing：ChinaRenminUniversityPress，2007：39-40.

[28] LIU X. On patent regulating of technical schemes generated by artificial intelligence—theoretical disputes， practi- cal problems and legal solutions[J]. Science of Law（Journal of Northwest University of Political Science and Law）， 2019， 37（5）： 82-92.

[29] WU H D， etc. Research on the basic problems of intellectual property rights （2nd Edition）[M]. Beijing： China Ren- min University Press， 2009： 245.

[30] LI C L， YE J Y. Patent publication，spillover and the private value of patents[J]. Journal of Central University of Fi- nance & Economics， 2016， 38（9）： 112-121.

[31] LIU X. On Philosophy and legitimacy of patent[J]. Journal of Foshan University （Social Science Edition）， 2016， 34（2）： 29-34.

[32] WU H D. Patent technology transformation and intangible assets operation[J]. Patent Agency， 2016， 2（2）： 13-16.

[33] FU L X. Industry revolution and intellectual property strategic response under the global value chain—based on the analysis of world intellectual property report 2017[J]. China Invention & Patent， 2018， 15（6）： 47-52.

[34] SHAPIRO C. Navigating the patent thicket： cross licensing， patent pool and standard setting[J]. Chapter in NBER book Innovation Policy and the Economy， 2001， 1（1）： 119-150.

[35] HELLER M A. The tragedy of the anticommons： property in the transition from marx to markets[J]. Harvard Law Review， 1998， 111（3）： 621-688.

[36] QIN J， LIU X. On Risks and measures of market operation of artificial intelligence patents[J]. Electronics Intellec- tual Property， 2019， 18（2）： 66-71.

人工智能技术的基本架构及其可专利性分析

刘鑫

（中南财经政法大学知识产权研究中心，武汉430073）

摘要：人工智能作为当今最为尖端的技术类型之一，涉及算法设计、数据整合、语音识别、图像处理等诸多技术内容，其基本架构包括基础层、感知层、认知层和应用层四个技术层次，每个层次中相关技术的不同技术特征与功能形态，导致了不同的可专利性判定结论。为实现对人工智能技术的充分专利保护，应对人工智能技术的基本架构进行全面解析，并根据我国专利法的相关规定，展开对人工智能技术中所涉及的各类技术方案的可專利性分析。基础层中的算法模型在现行专利授权规则下往往被认为是自然法则或抽象概念而被排除出专利保护客体范围；感知层和认知层中通过算法模型与特定硬件或应用场景相结合而形成的技术方案，须与算法模型相融合并有所提升，满足专利授权的创造性条件，才能成为我国专利法所保护的技术类型；应用层中最终产品的可专利性，则取决于产品的性质与结构是否符合专利授权的基本要件。

关键词：人工智能；技术架构；可专利性；算法模型；技术方案