Can AI algorithms be patented under AIA?

Before reading this article, you may find helpful on the software-related issue on means plus function: How Means-Plus-Function Limitation Interacts With Software Industry? Do “Processor,” “Controller,” Or “Program” Invoke 35 U.S.C. §112? Or, you may take a look at basic introduction of software in Taiwan: 電腦軟體的智慧財產權保護 Anyway, can AI algorithms be patented? Well, the short answer is “It depends!” Yeah, law people like “It depends.” The following article is a summary on cases addressed on the subject matter relevant to software or algorithms. Hope you enjoy it.

I. Introduction

Artificial intelligence (AI) encompasses technologies have become increasingly important to companies of all sizes across a wide range of fields in recent decades. For instance, AI technologies are crucial to visual recognition, autonomous vehicles, and assembling robotic arms. For the AI applications, an algorithm in a system is capable of learning from data, optimizing decision, and further enhance the operating performance. Also, an AI algorithm can combine with other computing systems. For example, the Neural Network (NN), a framework based on many different machine learning algorithms to work together and process complex data inputs. Such widespread applications of AI may provide an economic boost of approximately $14 trillion within the next two decades. Accordingly, companies pour billions into the research and development of AI technologies as well as the intellectual property rights protection of such investment.

In the United States, most enterprises may secure its AI-related technologies through the patent system. However, certain AI technologies can encounter scrutiny at the United States Patent and Trademark Office (USPTO). Upon receiving an application from an inventor, the examiner at USPTO inspects the patent applications closely with respect to the patent-eligible subject matter. Some of the AI technologies may involve “abstract ideas, laws of nature, and natural phenomena,”[1] which are not patentable under section 101 of the Patent Act.[2] Nevertheless, the judiciary and administration system both hold a positive attitude toward the patent-eligible of AI algorithm. The courts have ruled such patent claims recited with abstract ideas, laws of nature, or natural phenomena can still satisfy the patent eligibility with showing the “significantly more” than the aforementioned concepts.[3] Further, concerning the policy consideration, Andrei Iancu, the director of USPTO, has declared that AI Algorithms are generally patentable and the policies must highly focus on incentivizing AI and Big Data related invention.[4]

II. Short Introduction of Patent Claims

To monopolize an invention, a patentee should disclose its invention by claiming a “claim,” which defines the scope of its invention. Generally, if one attempt to obtain a patent for AI algorithms, its claim may be a process, system, method, or machine. In particular, section 101 of the modern Patent Act authorized granting of patents for a broad range of subject matter— “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.”[5] The statue does not express limitation for an invention. However, courts came to address the patent with respect to the “abstract ideas, laws of nature, or natural principles” whether a patent overly expand the scope of protection to be the monopoly of such concepts.[6] Consequently, the navigation of the boundaries whether aforesaid concepts are patentable has been carefully studied. In recent, it turns to be a harsh issue to the courts, USPTO, software-based companies, and even the Congress.

III. The Evolution of Patentable Subject Matter

        Since the 1800s, the first Industrial Revolution took place in many different fields. The spark of genius brought a sequence of transition from hand production methods to machines; new chemical manufacturing and iron production processes drastically increasing the technologies development. Inventors turned to the patent system seeking protection for their great discoveries or inventions. Nevertheless, some patents might not suffice the requirement to be the subject matter under such patent system.

In Neilson v. Harford, a well-known English case, Neilson’s patent was challenged by merely claimed a natural principle, which injecting hot air into the furnace, instead of cold, and thereby increasing the intensity of the heat.[7] The English court upheld the patent, reasoning that Nelson’s invention did not claim the principle itself but an application of the principle. 

In Le Roy v. Tatham, the Supreme Court followed the English approach, articulating that “principle, in the abstract, is a fundamental truth; an original cause; a motive; these cannot be patented, as no one can claim in either of them an exclusive right.” [8]

Next year in 1854, O’Reilly v. Morse, the Supreme Court again addressed the Neilson, endorsing that Morse could not claim the electric or galvanic current itself, whereas Neilson claimed an invention, improving the functioning of a hot blast furnace with applying the natural principle. [9]

In Funk Brothers Seed Co. v. Kalo Inoculant Co, the Court upheld that an application of a natural principle should be “new and useful.”[10] The patentee, Kalo Inoculant Company, claimed particular combinations of naturally occurring bacteria, which can promote plant growth. However, it is “no more than the discovery of some of the handiwork of nature” and hence is not patentable; the application of the invention is hardly more than an advance in the packaging of the substances for plants.[11] So far, the Courts repeatedly descanted the importance of the application of a natural principle. 

However, what if a patentee “invents” an algorithm like converting binary-coded decimal (BCD) numerals into pure binary numerals? In Gottschalk v. Benson, the Supreme Court rejected that mathematical formula to be patented on subject matter grounds.[12] It is a milestone of identifying cases for patenting an algorithm related invention. The patentee claimed an algorithm, a process claim, for converting BCD numerals into pure binary numerals. The Court invalidated the process claim by articulating three principles: (1) [p]henomena of nature, though just discovered, mental processes, and abstract intellectual concepts are not, as they are the basic tools of scientific and technological work,”;[13] (2) [t]ransformation and reduction of an article “to a different state or thing” is the clue to the patentability of a process claim that does not include particular machines,”;[14] (3) algorithms may not be patented if it has no practical application except in connection with a digital computer, the patent would “wholly pre-empt the mathematical formula.”[15]

However, it is not enough a patentee claims an application connected with its algorithms, rather the application should be inventive. In Parker v. Flook, the patentee has invented a procedure for updating an alarm limit during catalytic conversion processes—measuring the present value of the process variable (e.g., the temperature), using an algorithm to calculate an updated alarm—limit value, and adjusting the value of alarm limit.[16] The court declared that it is unpatentable not because it contains the mathematical algorithm, but there is no inventive application of the algorithm—providing the unpatentable formula for computing updated alarm limit for catalytic chemical conversion of hydrocarbons.[17](emphasis added).

By contrast, in Diamond v. Diehr, the patentee made a great application on rubber molding process. [18] It claimed a patent of automatic opening of the curing mold, a process of curing synthetic rubber, involved mathematical equation but do not seek to pre-empt the use of that equation. Again, the Court reiterated that “a process is not unpatentablesimply because it contains a law of nature or a mathematical algorithm.”[19] (emphasis added). Also, “an application of a law of nature or mathematical formula to a known structure or process may well be deserving of patent protection.”[20] The Court referenced Benson and said that “transformation and reduction of an article ‘to a different state or thing’ is the clue to the patentability of a process claim that does not include particular machines.”[21] It is clear that a process still can be patent only contains process with an algorithm. The question is whether an application is inventive, considering the claim as a whole— “it is inappropriate to dissect the claims into old and new elements and then to ignore the presence of the old elements in the analysis.”[22]

After 2000, along with the internet development burst, the United States Court of Appeal for Federal Circuit (CAFC) gradually eroded patent eligibility limitations. In re Alappat, building off of Diehr, the court held that the display of data on a computer screen could suffice subject matter.[23]

However, in re Bilski, which the patentee claimed a method of managing risks of commodity prices. [24] The CAFC articulated the “machine-or-transformation” (MoT) test “to determine whether a process claim is tailored narrowly enough to encompass only a particular application of a fundamental principle rather than to pre-empt the principle itself.” (emphasis added)[25]. Under the MoT test, “a claimed process is surely patent-eligible under § 101 if: (1) it is tied to a particular machine or apparatus, or (2) it transforms a particular article into a different state or thing.”[26] The Bilski patent failed both prongs and hence was unpatentable: it was not tied to a particular machine, and the transformation of avoiding business risks is not a tangible change. 

In Bilski v. Kappos, the Supreme Court reviewed in re Bilski, affirmed Bilski’s claim was unpatentable;[27] however, the Court denied the MoT test, stating it was too rigid given the broad statutory definition under section 101 of Patent Act. However, it still is “a useful and important clue or investigative tool.”[28]

In Mayo Collaborative Servs. v. Prometheus Labs., Inc., has the process been transformed?[29] The Court has made clear that in order to transform an unpatentable law of nature into a patent-eligible application of such a law, a patent must do more than simply state the law of nature while adding the words “apply it.”[30] Thus, the key is whether a process involved natural law suffices an “inventive concept,” sufficient to ensure that the patent in practice amounts to significantly more than a patent upon the natural law itself. The Court again addressed and distinguished Flook and Diehr, and stated that Flook’s claim did not limit it to a particular application, all the steps in its processes were well-known, and there was no “inventive concept” in it; in particular, the steps involved “well-understood, routine, conventional activity” previously engaged in by researchers in the field.[31] Such activity was not sufficient to transform unpatentable concepts into a patent-eligible application.[32]

IV. Alice Two-Step Test

The Supreme Court this time established a two-step test for patent eligibility.[33] Alice’s claim stated a method that creating a debit and credit account for each party, checking the account balances in the morning, adjusting the account balances through the day, and paying the parties at the end of the day if both parties have performed. Such case reminded ofthe Supreme Court Bilski and Mayo, and it further set a framework based on Mayo as a two-step inquiry:[34]

Step 1: Does the patent claim a patent-ineligible law of nature, natural phenomena, or abstract idea?

Step 2: If so, does the claim contain an inventive concept sufficient to transform the ineligible law of nature, natural phenomena, or abstract idea into a patent-eligible application of the ineligible subject matter? 

The “inventive concept,” as the court stated, may be an element or combination of elements that is “sufficient to ensure that the patent in practice amounts to significantly more than a patent upon the ineligible concept itself.”[35]

Such analysis has become a canon to a court and the patent draftsman since 2014. Courts applied the two-step test and tried to delineate the boundary of “significantly more.”  For instance, in Finjan, Inc. v. Blue Coat Systems, Inc., the Federal Circuit upheld a finding that Finjan’s patent was patent-eligible and not directed to an abstract idea, even though it was a software innovation.[36] The patented method, unlike well-known and conventional virus screening program only ensures a file being scanned before a user’s access, did a good deal more. In specific, the method “is distinguished from traditional, “code-matching” virus scans that are limited to recognizing the presence of previously-identified viruses, typically by comparing the code […] to a database of known suspicious code.” Rather, it claimed a “behavior-based” method could generate security profile attaching to a potentially hostile operation, which can avoid from detection by cosmetically modified its name or deputy file name, and prevent a user from potential threats; and such approach constituted an improvement in computer functionality hence was deemed to be a non-abstract idea. The court confirmed that software-based innovations could make “non-abstract improvements to computer technology” and be deemed patent-eligible subject matter.[37]

However, is Alice Two-Step test the best solution for subject matter inquiry? In Berkheimer v. HP Inc., LOURIE, Circuit Judge, with whom NEWMAN, Circuit Judge, joins, concurring in the denial of the petition for rehearing en banc, question the Alice test that “if a method is entirely abstract, is it no less abstract because it contains an inventive step?” [38] The step two’s requirement whether a claim element or combination of elements is “well-understood, routine, and conventional” to a skilled artisan in the relevant field is essentially a factual issue, the section 102 and 103 inquiry.[39] Moreover, it cast doubt on the subject matter does not need a two-step test. 

Instead, the judge suggested that there shall be a further authoritative treatment beyond the Alice test, which considers the patent-eligibility issue along with novelty, nonobviousness, and written description. The author suggests such complicated matter may seek to the Congress for another resort, as the court in Benson has questioned the Congress whether a computer program ought to be patentable: 

If these programs are to be patentable, considerable problems are raised which only committees of Congress can manage, for broad powers of investigation are needed, including hearings which canvass the wide variety of views which those operating in this field entertain. The technological problems tendered in the many briefs before us indicate to us that considered action by the Congress is needed.[40]

V. The Patent Eligible of Artificial Intelligence Algorithm 

An AI algorithm can be defined as a set of unambiguous instructions that a mechanical computer can execute; in particular, it usually is a mathematical equation or formula functioning a program or process. Such an algorithm can be applied to AI-related application like Machine Learning (ML), Neural Network (NN), Natural Language Processing (NLP), Heuristic Algorithm, and so on. Due to the wide range of the algorithm’s application, can one still monopolize such algorithm itself? 

The author starts from the statue that provides for patent “any new and useful process, machine, manufacture, or composition of matter.”  However, the Court, in Le Roy and O’Reilly, upheld that “principle, in the abstract, is a fundamental truth; an original cause; a motive” cannot be patented. An inventor cannot claim laws of nature, abstract ideas, natural phenomena. [41][42]

Although algorithms itself are so abstract, the Judge, in Benson, concedes algorithms can be patent if it has any transformation and reduction of an article to a “different state or thing,” or further “apply it.”[43] Therefore, a drafter may connect an algorithm with a machine to cooperate such algorithm, and also claim a particular application complying transformation requirement; noting that such co-work machine cannot solely process the algorithm.[44] The Court worries about a patent wholly pre-empt the algorithms, as basic tools of scientific and technological work, impeding junior researches by availing the algorithms. 

An algorithm as a basic tool and cannot be patented.  However, what if the algorithm’s application gives an improvement in the AI area, and enhance the efficiency of basic operations such as Big Data Training, Machine Learning, or Neural Network? Do not such refinements of mathematical algorithm deserve patent protection? Following Parker and Diehr, the Court’s answer is affirmative. The Parker suggests an inventive application of a well-known algorithm can still be patented;[45] Diehr further emphasizes elements in a claim should be analyzed as a whole to determine whether such patent involves the “inventive application.”[46] For instance, an inventive application of Convolutional Neural Network (CNN), often being used as tools for processing image recognition, may be patentable if its claimed invention, considered as a whole. Still, it is hard to identify and distinguish whether an invention has sufficed the “inventive application.” 

According to in re Bilski, the application of CNN algorithm, narrowly tailored into a specific field, may suffice subject matter.[47] Furthermore, under the MoT test, an important clue or investigative tool, the application better has tangible changes such as providing high accuracy to track vehicle movement. Also, is the application of CNN algorithm beyond the “well-understood, routine, or conventional activity?” [48]

Following Alice, under the step one of Two-Step test, to avoid the application of CNN algorithm to be an abstract idea, a claimed algorithm can be drafted by complying MoT test, inventive, and narrow enough application. If one failed under the step one, an inventor could show that it is not “well-understood, routine, conventional activity,” which “significantly more” than the ineligible concept in step two. 

Compared with Finjan, it is patentable because it has “non-abstract improvement to computer technology.”[49] Is the CNN algorithm application abstract? In Alice, the court has stated that “monopolization of [abstract ideas] through the grant of a patent might tend to impede innovation more than it would tend to promote it.”[50] In other words, the improvement of an algorithm may not be abstract if it will not pre-empt such an algorithm. However, the result may vary by the patent drafting skills and the innovative algorithm itself; the author believes when a judge adopts the rules or the framework as aforesaid, the inquiry may probably turn to ascertain the novelty and non-obviousness issue comparing with prior arts, instead of patent-eligibility.

VI. Conclusion 

Undoubtedly, an AI algorithm can be patented as long as it suffices the requirement of the subject matter as the statue state: “whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter….” In particular, an inventor cannot claim the “laws of nature, abstract ideas, and natural phenomena,” which is the basic tools of scientific and technological work. Instead, the algorithm needs to be applied in a specific field of art, transforming it into a “new and useful” application, and “different state or thing.” Furthermore, the claim of AI algorithm application should be considered “as a whole,” whereby the claim better cooperates with a tangible machine or processor. Last but not least, such AI algorithm has to have an “inventive concept” which the AI algorithm application should be “significantly more” an “well-understood, routine, conventional activity” previously engaged in by researchers in the field. 

This image has an empty alt attribute; its file name is by-nc-nd.png



楊杰凱 專利師
Jack Yang Patent Attorney

總承國際專利商標事務所|總承會計師事務所|主持專利師|商標代理人|國立清華大學工業工程管理學系|清大創新創業學分學程|華盛頓大學法學院法學碩士|接受新事物,突破舒適圈。主要提供公司成立及智財服務,從工商登記、商標設計、商標申請到專利申請。“Stay Positive. Work Hard. Make it happen.”


  • [1] Diamond v. Diehr, 450 U.S. 175, 185 (1981).
  • [2] 35 U.S.C.A. § 101 (West, Westlaw through 1952 legislation)
  • [3] Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208
  • [4]
  • [5] 35 U.S.C.A. § 101 (West, Westlaw through 1952 legislation)
  • [6] Diamond v. Diehr, 450 U.S. 175, 185 (1981)
  • [7] Neilson v Harford, 151 ER 1266 (1841).
  • [8] Le Roy v. Tatham, 55 U.S. 156, 175 (1853).
  • [9] O’Reilly v. Morse, 56 U.S. 62 (1854).
  • [10] Funk Brothers Seed Co. v. Kalo Inoculant Co., 333U.S. 127, 131 (1948).
  • [11] Id.
  • [12] Gottschalk v. Benson, 409 U.S. 63 (1972).
  • [13] Id, at 67.
  • [14] Id, at 70.
  • [15] Id, at 70.
  • [16] Parker v. Flook, 437 U.S. 584 (1978).
  • [17] Id. at 594.
  • [18] Diamond v. Diehr, 450 U.S. 175 (1981).
  • [19] Parker, 437 U.S. at 594.
  • [20] Diehr, 450 U.S. at 187.
  • [21] Gottschalk v. Benson, 409 U.S. 63, 70 (1972).
  • [22] Diehr, 450 U.S. at 188.
  • [23] In re Alappat, 980 F.2d 1439 (1992).
  • [24] In re Bilski, 545 F.3d 943 (Fed. Cir. 2008) (en banc).
  • [25] Id. at 954.
  • [26] Id.
  • [27] Bilski v. Kappos, 561 U.S, 592 (2010).
  • [28] Id. at 594.
  • [29] Mayo Collaborative Servs. v. Prometheus Labs., Inc., 566 U.S. 66 (2011).
  • [30] Gottschalk v. Benson, 409 U.S. 63, 71-72 (1972).
  • [31] Mayo 566 U.S. at 73.
  • [32] Id. at 67.
  • [33] Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208 (2014).
  • [34] Id. at 218.
  • [35] Id.
  • [36] Finjan, Inc. v. Blue Coat Sys., 879 F.3d 1299 (Fed. Cir. 2018).
  • [37] Id. at 1305.
  • [38]Berkheimer v. HP Inc., 890 F.3d 1369, 1374 (2018).
  • [39] 35 U.S.C.A. §§ 102-103 (West, Westlaw through 1952 legislation).
  • [40] Gottschalk v. Benson, 409 U.S. 63, 73 (1972).
  • [41] Le Roy v. Tatham, 55 U.S. 156, 175 (1853).
  • [42] O’Reilly v. Morse, 56 U.S. 62 (1854).
  • [43] Gottschalk v. Benson, 409 U.S. 63, 70 (1972).
  • [44] Id.
  • [45] Parker v. Flook, 437 U.S. 584, 594 (1978).
  • [46] Diamond v. Diehr, 450 U.S. 175, 188 (1981).
  • [47] In re Bilski, 545 F.3d 943, 954 (Fed. Cir. 2008) (en banc).
  • [48] Mayo Collaborative Servs. v. Prometheus Labs., Inc., 566 U.S. 66, 73 (2011).
  • [49] Finjan, Inc. v. Blue Coat Sys., 879 F.3d 1299, 1305 (Fed. Cir. 2018).
  • [50] Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208, 216 (2014).


這個網站採用 Akismet 服務減少垃圾留言。進一步了解 Akismet 如何處理網站訪客的留言資料

Powered by

Up ↑