PatentNext Takeaway: To date, the Federal Circuit has not reviewed many cases involving artificial intelligence (AI). However, in a recent case, the Federal Circuit found that a “machine learning” claim element lacked sufficient enablement because both the claim itself and the written description of the patent to which it belonged failed to describe “how” the claimed invention implemented this element.  In view of this ruling, patent practitioners should endeavor to explain sufficiently in the written description the specific aspects of how machine learning features (and other computer-implemented invention features) operate in order to demonstrate sufficient enablement.

****

Enablement of a claimed invention is one requirement of U.S. Patent Law (specifically 35 U.S.C. § 112(a)). Specifically, enablement requires a patent specification to contain sufficient information regarding the subject matter of the claims so as to “enable” a person skilled in the art to make and use the claimed invention. Typically, a computer scientist, computer engineer, or skilled computer programmer constitutes a person of skill for computer-implemented inventions.

To date, while the Federal Circuit has reviewed and ruled on many computer-implemented inventions, the Federal Circuit has yet to provide significant guidance involving artificial intelligence (AI), which is, of course, one of the fastest-growing fields of computer technologies in modern times. 

However, recently, the Federal Circuit provided some guidance regarding enablement in a 2023 case. See in re Starrett, 2023 U.S.P.Q.2d 684. (Fed. Cir., Jun. 8, 2023) (nonprecedential). While the decision is nonprecedential, the case allows a window into the Federal Circuit’s analysis and treatment of artificial intelligence-type claims with respect to Section 112 issues such as enablement.

In re Starrett, 2023 U.S.P.Q.2d 684. (Fed. Cir., Jun. 8, 2023) 

In Starrett, the Federal Circuit considered an appeal from the Patent Trial and Appeal Board (PTAB) regarding U.S. Patent Application 15/299,124 (“the ‘124 application”). In re Starrett, 2023 U.S.P.Q.2d 684. (Fed. Cir., Jun. 8, 2023).  The ’124 application claimed an invention for maintaining “data structures representing categories of biological signals in a body such as a ‘Nervous System’ and a ‘Sensory System.’” Id. at *1.

Claim 1 of the ’124 patent recited, in part, a “machine learning” element directed to a specific “configuration.” This “machine learning” element is reproduced below:

[b)] configur[ing] to receive, relay, transmit, or distribute one or more signal [sic] wherein at least one signal comprising data representative of information about one or more biological body [sic] wherein the processing of biological systems data using at least one machine learning task intelligibly recovering perceived, experienced, remembered, or imagined imagery, sounds, or feelings as one or more computational, visual, auditory, textual, numeric, symbolic, coordinate, or haptic representation; or …

Id. (citing ’141 application, claim 1.) (emphasis added).

During prosecution, the examiner had rejected all claims (including claim 1) for lacking enablement pursuant to 35 U.S.C. § 112(a). The rejection was appealed to the PTAB, which affirmed the examiner. Id. at *2. The PTAB found that claim 1 was a type of genus claim that  “contain[ed] forty-seven ‘or’ clauses, thereby allowing it to cover over 140 trillion embodiments.” Id. In addition, while the patent applicant had argued that claim 1 was “fully enabled” by the patent application’s “laboriously detailed” specification, the PTAB disagreed finding such assertions merely conclusory. Id.at *2-*3. Finally, the Board noted that the patent applicant’s contentions essentially amounted to “argu[ing] that if an apparatus is well-known . . . , then any function that [the inventor] claims for that apparatus is also fully enabled.” Id.at *3 (citations omitted).

On appeal, the Federal Circuit affirmed, citing the Supreme Court’s precedent regarding enablement in the recent Amgen Inc. v. Sanofi decision:

“If a patent claims an entire class of processes, machines, manufactures, or compositions of matter, the patent’s specification must enable a person skilled in the art to make and use the entire class. In other words, the specification must enable the full scope of the invention as defined by its claims. The more one claims, the more one must enable. 

Id. at *4 (citing 143 S. Ct. 1243, 1254 (2023)). 

The Federal Circuit then applied this enablement principle to the ’141 application, finding that “[h]ere, much is claimed, and little is enabled.” Id. In particular, the Federal Circuit found particularly troubling the ‘141 patent application’s failure to explain how the claimed features would operate without undue experimentation: 

The application’s disclosure of a broad and abstract organizational structure used to accomplish the maintenance of augmented telepathic data amounts to little more than a “research assignment” requiring a skilled artisan to undertake undue experimentation to discover what types of devices are encompassed by the claim limitations and how they would function.

Id. at *5 (citing Amgen, 143 S. Ct. at 1256) (emphasis added). 

While the Federal Circuit did not specifically address the “machine learning” element of claim 1, it did find, more generally, that claim 1 was “rife with broad, vague concepts.” Id. at *5. For this reason, the Federal Circuit invalidated the claim based on a lack of sufficient enablement. Id.

The Federal Circuit also addressed the applicant’s contention that the claimed features were “well-known” and, as a consequence, allegedly “fully enabled.” See id. (discussing this aspect as part of a Wands factor analysis, of which consideration of “well-known” components is a part. See In re Wands, 858 F.2d 731, 737 (Fed. Cir. 1988)). However, the Federal Circuit found that whether a feature is “well-known” (or not) is but one factor of the Wands analysis and is not dispositive on its own. See id. Again, the Federal emphasized the importance of describing how the claim elements function. Id. (stating that “the Examiner’s discussion of the Wands factors properly faulted the specification for failing to describe how the claim elements function.”). 

Ultimately, claim 1 (containing the “machine learning” element) was found non-enabled because the applicant had failed to describe how this (and other) aspects of the invention worked, and the applicant could not rely on the knowledge of a person of skill in the art to cure this defect, no matter how “well-known” such prior art elements were. As the Federal Circuit noted, “[a]lthough the knowledge of one skilled in the art is indeed relevant, the novel aspect of an invention must be enabled in the patent.” Id. (citing Auto. Techs. Int’l, Inc. v. BMW of N. Am., Inc., 501 F.3d 1274, 1283 (Fed. Cir. 2007)).

Realtime Data v. Array Networks, 2023 WL 4924814 (Fed. Cir., Aug. 2, 2023).

Enablement could become a more prominent area of focus with respect to computer-implemented inventions if, for example, Section 101 is resolved via legislation. This topic was recently previewed in Realtime Data v. Array Networks in a dissenting opinion by Judge Newman. 2023 WL 4924814 (Fed. Cir., Aug. 2, 2023). 

The Majority opinion (focusing on Section 101)

In Realtime Data, the patent-at-issue generally related to computer-implemented technology (and not artificial intelligence), where the claims recited methods and systems for digital data compression. The majority opinion, by Judge Reyna, affirmed a district court’s decision invalidating the claims of the patents-at-issue as abstract ideas pursuant to Section 101. Id at *1. In particular, the Federal Circuit agreed with the district court that the claims were directed to the abstract idea “of manipulating information using compression.” Id. at *7. The court admonished the claims and specification, stating that “[w]e have determined that “the claim itself … must go beyond stating a functional result” and that “the claim must “identify ‘how’ th[e] functional result is achieved by limiting the claim scope to structures specified at some level of concreteness, in the case of a product claim, or to concrete action, in the case of a method claim.” Id. (citing Am. Axle & Mfg., Inc. v. Neapco Holdings LLC, 967 F.3d 1285, 1302 (Fed. Cir. 2020)) (emphasis added). 

Because the claims “failed to do this,” the majority opinion held the claims to be invalid pursuant to Section 101. Id. at *8.  In particular the majority found that “none of the claims at issue specifies any particular technique to carry out the compression of data—the particular rules for producing a smaller set of data out of a larger starting set.” Id.  Rather, the claims “all take the availability of compression techniques as a given and address the threshold matter of choosing to use one or more such available techniques.” Id. The majority further faulted the abstract nature of the claims stating that “even as to making such a selection, the claims are directed to only abstract ideas, calling for unparticularized analysis of data and achievement of general goals.” Id.

Judge Newman’s Dissent (focusing on Section 112)

Judge Newman dissented, arguing that the proper lens for determining was not Section 101 but Section 112, in particular, enablement. See id. At *12 (Newman, J., dissenting) (stating that “This is properly an enablement case.”). Judge Newman did not analyze the claim under Section 112. Rather she advocated that the proper review belonged under Section 112, and not Section 101:

I write separately to note once again that § 101 was never intended to bar categories of invention in this way. This judicial exception to eligibility is an unnecessary and confusing creation of the courts. This case is an example, for the enablement requirement of § 112 is better suited to determining validity of these claims than is the distortion of § 101. I respectfully dissent, and would remand for determination of validity under § 112.

Id. (Newman, J., dissenting) 

She ended her dissent by noting that “[e]ligibility law has been called a ‘morass of seemingly conflicting judicial decisions’” (citations omitted) and that “[w]e should not wade further into this morass.” Id. at *13.  “This case is another example that conforms with our flawed precedent. I respectfully dissent. I would remand for a determination of validity under § 112 and, if applicable, §§ 102 and 103.” Id.

Conclusion

In view of the Starrett decision, practitioners should endeavor to explain sufficiently in the written description the specific aspects of machine learning (and other computer-implemented invention features). In particular, practitioners should endeavor to describe how a claimed computer-implemented invention (e.g., an artificial intelligence (AI) invention) operates or otherwise works. Previous articles on PatentNext discuss this important objective and cases and examples demonstrating how to achieve it. See PatentNext, Why including an “Algorithm” is Important for Software Patents (Part 2)

Finally, in view of the Realtime Data, practitioners can get a preview of what may yet come in the event of a legislative change to Section 101 (or a ruling from the Supreme Court bringing about the same), where the new invalidity battleground is not Section 101, but instead Section 112.  

****

Subscribe to get updates to this post or to receive future posts from PatentNext. Start a discussion or reach out to the author, Ryan Phelan, at rphelan@marshallip.com (Tel: 312-474-6607). Connect with or follow Ryan on LinkedIn.

Photo of Ryan N. Phelan Ryan N. Phelan

PatentNext is moderated by Ryan N. Phelan, a registered U.S. Patent Attorney and Software and Computer Engineer. Ryan previously worked in the IT industry as a consultant at Accenture, where he regularly consulted Fortune 500 companies in software and computing technologies. Ryan is…

PatentNext is moderated by Ryan N. Phelan, a registered U.S. Patent Attorney and Software and Computer Engineer. Ryan previously worked in the IT industry as a consultant at Accenture, where he regularly consulted Fortune 500 companies in software and computing technologies. Ryan is featured in the IAM Strategy 300 & 300 Global Leaders guides, and was selected for inclusion in The Best Lawyers in America© list in the practice area of Patent Law. Ryan is also an adjunct professor at Northwestern University’s Pritzker School of Law where he teaches coursework on Patenting Software Inventions. Learn more about Ryan.