Is an invention autonomously generated by artificial intelligence patentable? This is a question that is being studied including by the United States Patent and Trade Mark Office (USPTO) which launched an investigation into issues associated with patenting artificial intelligence inventions. In the meantime, the USPTO just released a decision denying the application for a such a patent holding that under the U.S. patent law, 35 USC §§ 1 et seq. an inventor must be a natural person.
The decision was based on an application listing a single inventor with the given name “[DABUS]” and the family name “(Invention generated by artificial intelligence).” The application was for an invention conceived by a “creativity machine” named “DABUS” which claimed it should be named as the inventor.
According to the petition, which was advanced by The Artificial Inventor Project, a project that seeks intellectual property rights for the autonomous output of artificial intelligence, the invention was generated by a machine named “DABUS.” This “creativity machine” was programmed as a series of neural networks that have been trained with general information to independently create the invention. DABUS was not created to solve any particular problem, and it was not trained on any special data relevant to the invention. Instead, it was the machine, not a person, which recognized the novelty and salience of the instant invention.
The petitioner contended that inventorship should not be limited to natural persons and, therefore, the naming of DABUS as the inventor in the application was proper. The petitioner accepted that an AI system could not under current law own property including the property rights in inventions the machine itself created. This further called into question whether an assignment document satisfied the requirements of the applicable law and the ability of the assignee to even file the application.
In rendering the decision, the USPTO referred to a number of provisions in the U.S. patent statute including the definition of “inventor” which means “the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention”. It also referred to other provisions including 35 U.S.C. § 101 which prescribes who can obtain a patent which states “Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.”
The petition was dismissed and the decision was affirmed on reconsideration. According to the USPTO:
To the extent the petitioner argues that an “inventor” could be construed to cover machines, the patent statutes preclude such a broad interpretation. Title 35 of the United States Code consistently refers to inventors as natural persons. For example, 35 U.S.C. § 101 states “Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter…may obtain a patent therefore, subject to the conditions and requirements of this title” (emphasis added). “Whoever” suggests a natural person.9 35 U.S.C. § 115 similarly refers to individuals and uses pronouns specific to natural persons—”himself” and “herself”—when referring to the “individual” who believes himself or herself to be the original inventor or an original joint inventor of a claimed invention in the application. It further states that the inventor who executes an oath or declaration must be a “person.”11 Other sections of Title 35 take the same approach.12 Therefore, interpreting “inventor” broadly to encompass machines would contradict the plain reading of the patent statutes that refer to persons and individuals.
The USPTO then referred to some other U.S. cases that, while not dealing with AI inventions, touched on the requirement that an inventor be a natural person.
In addition, the U.S. Court of Appeals for the Federal Circuit (Federal Circuit) has explained that the patent laws require that an inventor be a natural person. For example, in Univ. of Utah v Max-Planck-Gesellschaft zur Forderung der Wissenschaften e.V.,13 the Federal Circuit explained that a state could not be an inventor, stating—The inventors of a patent are “the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention.” 35 U.S.C. § 100(f). It is axiomatic, that inventors are the individuals that conceive of the invention:
Conception is the touchstone of inventorship, the completion of the mental part of invention. It is the formation in the mind of the inventor, of a definite and permanent idea of the complete and operative invention, as it is hereafter to be applied in practice. Conception is complete only when the idea is so clearly defined in the inventor’s mind that only ordinary skill would be necessary to reduce the invention to practice, without extensive research or experimentation. [Conception] is a mental act. . . .
Burroughs Wellcome Co. v. Barr Labs., Inc., 40 F.3d 1223, 1227-28 (Fed.Cir.1994) (internal quotation marks and citations omitted). To perform this mental act, inventors must be natural persons and cannot be corporations or sovereigns…
While these Federal Circuit decisions are in the context of states and corporations, respectively, the discussion of conception as being a “formation in the mind of the inventor” and a “mental act” is equally applicable to machines and indicates that conception—the touchstone of inventorship—must be performed by a natural person.
In denying the application the USPTO also relied on its own understanding, Title 37 of the Code of Federal Regulations, and the Manual of Patent Examining Procedure (“MPEP) which defines “conception” as “the complete performance of the mental part of the inventive act” and it is “the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention as it is thereafter to be applied in practice.”‘
The petitioner also advanced policy considerations to support the position that a patent application can name a machine as an inventor. For example, it contended that allowing a machine to be listed as an inventor would incentivize innovation using AI systems, reduce the improper naming of persons as inventors who do not qualify as inventors, and support the public notice function by informing the public of the actual inventors of an invention. These were rejected as they “do not overcome the plain language of the patent laws as passed by the Congress and as interpreted by the courts”.
The decision of the USPTO followed similar decisions reached by the European Patent Office and the UK patent office. These corresponding decisions are summarized in a blog post written by Alfred Macchione and Ben Lai of Mccarthy Tetrault.
The decision of the USPTO did not address the circumstances in which a natural person can contribute to conception of an AI invention and be eligible to be a named inventor. This issue is also being examined by the USPTO. According to the Association for the Protection of Intellectual Property (AIPPI) in a submission to the USPTO this could arise in a variety of situations:
… the ways in which a natural person may contribute inventively towards machine learning can include the identification and/or selection of an available intelligence model or models, in the selection or configuration of model parameters, the training methodology to be applied and the selection of appropriate training parameters. Moreover, inventive activity may present itself in the selection, collection, cleansing and formatting of training data, and/or in the preparation and transformation of data for a particular training model.
Furthermore, a contribution to an invention involving AI may reside in the decision to use AI for a specific purpose or problem and/or directing the development of a specific AI solution for such purposes. A further contribution can lie in the abstraction of a specific solution provided by the 5 AI, e.g. a specific design of a mechanical part, to a general concept, e.g. a design principle for certain mechanical parts.
The DABUS decision did not touch on the very important question as to whether the goals of the patent system would be achieved if autonomously generated inventions could be protected by patents. The goals of patent law are generally recognized as seeking to foster and reward invention; promote disclosure of inventions to stimulate further innovation and to permit the public to practice the invention once the patent expires; and to assure that ideas in the public domain remain there for the free use of the public. Whether the law should be clarified or changed would depend on, among other things, how these goals would best be fostered, and in particular, whether permitting such inventions to be patentable would encourage innovation without unduly hindering competition. See, Barry Sookman, Law and Innovation: Is Intellectual Property a Path to Progress.
For more on the position in the U.S., you may want to read the USPTO Request for Comments on Intellectual Property Protection for Artificial Intelligence Innovation and the many submissions filed in response which can be accessed on the USPTO’s AI site.
Similar questions of whether AI generated subject matter can qualify for protection have arisen in other fields of intellectual property including copyright. Copyright statutes in many countries are premised on a work only being protected if the originality in the work arises from the authorship of a natural person. Accordingly, would works created using AI be protected by copyright? Some courts have rendered decisions that are relevant to this question.
In the U.S., “a curious male crested black macaque” named Naruto was denied copyright protection for a photograph by the Ninth Circuit Court of Appeals in Naruto v Slater 888 F.3d 418 (9th.Cir.2018) because Naruto was not a human being. In Southco, Inc. v. Cambridge Corp., 390 F. 3d 276 (3rd Cir. 2004), a U.S. Court declined to protect a listing of parts where the part numbers were generated using a predetermined and prescribed set of rules which precluded any originality by the persons who used the system.
Some cases in the U.S. have suggested that copyright can subsist in a work that is created by a computer program that leaves little involvement for users. For example, in REARDEN LLC v. The Walt Disney Company, (N.D. Feb. 21, 2108), a U.S. District court stated that “some authorities “suggest that the copyright protection afforded a computer program may extend to the program’s output if the program `does the lion’s share of the work’ in creating the output and the user’s role is so `marginal’ that the output reflects the program’s contents””, referring to Design Data Corp. v. Unigate Enter., Inc., 847 F.3d 1169, 1173 (9th Cir. 2017) (quoting Torah Soft Ltd. v. Drosnin, 136 F. Supp. 2d 276, 283 (S.D.N.Y. 2001). On the facts, the court did not find this to be the case stating:
The Court does not find it plausible that the MOVA Contour output is created by the program without any substantial contribution from the actors or directors. Unquestionably, the MOVA program does a significant amount of work to transform the two dimensional information captured on camera into three dimensional Captured Surface and Tracking Mesh outputs. But this cannot be enough, since all computer programs take inputs and turn them into outputs. See 17 U.S.C. § 101 (“A `computer program’ is a set of statements or instructions to be used directly or indirectly in a computer in order to bring about a certain result.”). Here, Rearden must allege that the MOVA program has done the “lion’s share of the work,” and in particular “the lion’s share of the creativity” in creating the outputs. See Torah Soft, 136 F. Supp. 2d at 283.
Rearden has not met this burden. Here, unlike in Torah Soft, where the user merely inputs a word into the program, MOVA Contour’s user inputs a two dimensional camera capture that may range from Dan Stevens’ “facial expressions of all the scenes we had done on previous days” to the “subtle and dynamic motions performed by the actor [Josh Brolin playing Thanos in Guardians of the Galazy)” to “Brad Pitt’s 44-year-old face.” Disney Compl. ¶¶ 2, 4, 36. Defendants’ role in creating the end-product is not so `marginal’ that the output reflects the program’s contents.” See Design Data Corp, 847 F.3d at 1173 (internal citations omitted).
The decision in the REARDEN case did not analyze the difference between two very different questions. First, whether a person that uses software that is solely responsible for the generation of the program’s output can be entitled to claim copyright in the output. Second, whether the creator of a program that generates an original output with only negligible contribution to the work’s originality by the program developer is eligible for copyright protection.
There have been several decisions in China addressing copyright for AI created works. In Feilin v Baidu, as summarized in the IPkat IP blog, a Beijing court ruled that for copyright to subsist in a work it must be created by a natural person. The court ruled a work created through the use of software must sufficiently pass on the original expression of the idea or feeling of the software user. The court expressed the view that a report generated by a visualisation function of software with the only participation of the user being to choose keywords used to generate the report was insufficient. In a second case, also reported by the IPkat, among others, Shenzhen Tencent v Yinxun, a Chinese court protected a work created by an intelligent writing assistant system called “Dreamwriter” where it found sufficient individual choices of the author. As summarized in a blog post by the IPKat:
In assessing the originality of the work, the court examined: (1) whether the work was created independently; and (2) whether the appearance of the work was sufficiently different from any existing work, or that it would at least contain minimum creativity. The court held that the article met the formal requirements of a literary work on the basis of its appearance. The contents of the article reflected the selection, analysis and judgment of the data and information in the stock market on that day. The court also stated that the article’s structure was reasonable and its expression logical and clear. In this sense, therefore, it was original.
The court then determined whether the process of creation of the article revealed the individual choice, judgement and skill of the creator. The court noted that the difference between how this article was created – in comparison with the typical way an article is composed – resided in the fact that there was a time gap between the acts of selection and arrangement carried out by the creating group and the actual act of writing executed by the software ‘Dreamwriter’, with this lack of synchronicity a function of the technical means used by the plaintiff.
Nonetheless, the selection of the contents by the creating group met the requirement of creation under the copyright law. The test was ultimately whether such selection and arrangement was an act of intellectual activity and whether such activity had a direct connection with the specific expression of the work. The court ruled that the answers were both “yes”. The operation of the algorithm merely gave technical effect to the creative work of the group. The creating group input the data, set up the trigger conditions, and chose and arranged templates and resources. These activities directly related to the creation of the article.
The issue of copyright protection for computer generated works has also arisen in a number of Commonwealth decisions. The Australian case, Telstra Corporation Limited v. Phone Directories Company Pty Ltd.,  FCAFC 149 (15 December 2010) involved the protection of Telstra’s white papers and yellow pages directories. The Full Court made clear that for copyright to subsist in a work that has been created using a tool such as software, there must be sufficient intellectual effort in the creation of the material form of the work. Perram J., expressed this opinion as follows:
The Act does not presently deal explicitly with the impact of software on authorship… But a computer program is a tool and it is natural to think that the author of a work generated by a computer program will ordinarily be the person in control of that program. However, care must taken to ensure that the efforts of that person can be seen as being directed to the reduction of a work into a material form. Software comes in a variety of forms and the tasks performed by it range from the trivial to the substantial. So long as the person controlling the program can be seen as directing or fashioning the material form of the work there is no particular danger in viewing that person as the work’s author. But there will be cases where the person operating a program is not controlling the nature of the material form produced by it and in those cases that person will not contribute sufficient independent intellectual effort or sufficient effort of a literary nature to the creation of that form to constitute that person as its author: a plane with its autopilot engaged is flying itself. In such cases, the performance by a computer of functions ordinarily performed by human authors will mean that copyright does not subsist in the work thus created.
A similar line of reasoning was applied in the Australian case, Acohs Pty Ltd v Ucorp Pty Ltd  FCA 577 (10 June 2010).
In Canada, for a work to be protected by copyright, there must be a sufficient degree of skill and judgment in its creation. CCH Canadian Ltd. v. Law Society of Upper Canada, 2004 SCC 13. As in Australia, even where tools are used in the creation of a work, if there is a sufficient degree of a natural person’s skill and judgment that goes into creating the material form of the work, it may be protected. This was a recent holding in the Alberta case in Geophysical Service Incorporated v. EnCana Corporation, 2016 ABQB 230, affirmed 2017 ABCA 125. This case involved whether copyright could subsist in seismic data. created using technical tools. According to the court, which relied on the Australia Telstra decision, the data was protected as an original literary and artistic work:
Clearly a human author is required to create an original work for copyright purposes. The Defendants argue, however, at para 102 of their brief that “copyright does not subsist in a work which is created by a computer with little if any human input.” They rely on the Australian case Telstra Corporation Limited v Phone Directories Company PTY Ltd, (2010) FCAFC 149.
First, I do not accept that seismic data is created with “little if any human input”. I reject the suggestion that the data is created by simply throwing a microphone in the ocean and pushing the “record” button on a computer. Human input is involved continuously through the acquisition stage, like creating a sound recording, as eloquently described by Dr. Wren’s report…
I agree with the conclusion in Telstra that putting together a generic telephone book may not require the human skill and judgment necessary to attract copyright protection if all that is being done is listing names in alphabetical order…
Ultimately, however, these are all factual determinations. In Telstra, the court concluded that the process was mechanical and involved little human input. Chief Justice Keane said at para 4, “the compilations were brought into the form in which they were published primarily by an automated computer process”. And later at para 33, “A majority of the creation process of the WPD and the YPD was heavily automated. Human intervention was regulated and controlled…” In that case, the computers virtually took over all the necessary decisions involving skill and judgment. That is not the case on the facts before me. Even though many technical instruments are used in the production of seismic data, they require human intervention, in the form of expert scientific skill and judgment to make them work. The seismic data produced through this exercise of skill and judgment is “tailored and unique” to the author. In my view, the human authorship element has been satisfied.
The issue of whether the Copyright Act would protect a work whose form could not be traced sufficiently back to a human author is something that could end up in the Supreme Court. Decisions of that court (including in CCH which addressed the level of originality required for copyright subsistence) take into account policy considerations such as how any decision would affect the Court’s view of the appropriate copyright balance. The Court also applies other malleable principles in construing the Act including the principle of technological neutrality. These principles cannot, however, trump clear wording in the Act. See, Canadian Broadcasting Corp. v. SODRAC 2003 Inc., 2015 SCC 57
The issue of whether works created using AI tools should be protected by copyright was debated during the last s.92 review of the Copyright Act. The report of the INDU Committee recommended “That the Government of Canada consider amending the Copyright Act or introducing other legislation to provide clarity around the ownership of a computer-generated work.” The Committee fell short of endorsing the neighboring right approach to protecting “computer generated works” as has been adopted in the UK and elsewhere which protects works generated by a computer in circumstances where there is no human author of the work.