Changes to AI Patent Practice in View of the USPTO’s Recent Actions
“Applicants will certainly be extremely cautious of filing a terminal disclaimer if the Proposed TD Rule passes, as doing so would make all the claims of a patent family vulnerable to widespread invalidation.”
Editor’s Note: This article was originally published with the incorrrect author listed; we regret the error.
The United States Patent and Trademark Office (USPTO, the Office or the Agency) has recently issued multiple guidance and proposed rules that potentially change the landscape of patent practice. On top of that, the USPTO has also proposed substantial terminal disclaimer fee increases that can induce early filing of terminal disclaimers. The Agency’s actions will significantly impact patenting artificial intelligence and it is important to understand the USPTO’s position from a holistic perspective with respect to this critical emerging technology. This paper specifically examines the impact of the following Agency actions on AI:
The AI Inventorship Guidance
Overview
The USPTO’s AI Inventorship Guidance was promulgated by directive of Presidential Executive Order with the policy goal of “promoting responsible innovation, competition, and collaboration will allow the United States to lead in AI and unlock the technology’s potential to solve some of society’s most difficult challenges.”
Under the AI Inventorship Guidance, inventions created with AI assistance are not categorially precluded from patentability. But AI cannot be listed as an author in light of the Federal Circuit’s decision in Thaler v. Vidal in which the court reiterated that inventors and joint inventors named on U.S. patents and patent applications must be natural persons. For claimed inventions that were invented with contribution from AI, such as generative AI, proper inventorship comes down to significant contribution from a human inventor. Particularly, the AI Inventorship Guidance states that “[p]atent applications and patents for AI-assisted inventions must name the natural person(s) who significantly contributed to the invention as the inventor or joint inventors.”
Therefore, inventorship comes down to “significant contribution” of the human inventor(s). The AI Inventorship Guidance thus requires listing all human inventors that provided significant contribution to the inventing process. If AI also contributed significantly to the claimed invention, AI cannot be named on the ADS. But significant contribution or use of AI does not preclude patentability as long as there are natural person(s) who significantly contributed to the claimed invention.
Practice Tips for Applicants and Practicing Attorneys
Firstly, the AI Inventorship Guidance states, “Generally, the USPTO presumes those inventors named on the application data sheet [the ADS] or oath/declaration are the actual inventor or joint inventors of the application. However, examiners and other USPTO personnel should carefully evaluate the facts from the file record or other extrinsic evidence when making determinations on inventorship.” Therefore, the ADS is presumed to be without error and the Agency is not to question the ADS without cause. If AI is not a listed inventor on the ADS, then the Agency should not question whether AI contributed significantly to the invention without cause.
Practitioners and applicants should then take care to not give the USPTO cause to question the ADS. For example, oftentimes, named inventors of inventions utilizing or implementing AI will publish a paper of their invention while also filing for patent protection. Applicants should be careful to not publish a paper having authors that do not correspond with the named inventors of their patent application. In some instances, interns or managers can be included or excluded from either one of these documents such that the publication authors and the inventors on the ADS are not consistent. These types of inconsistencies can be considered extrinsic evidence that can be cited by the Agency to question the validity or correctness of the ADS. Once that question is raised, a duty of candor or disclosure is triggered on the part of the applicant, which can be very problematic.
When it comes to the duty of candor, the AI Inventorship Guidance states that “when an examiner or other USPTO employee has a reasonable basis to conclude that an individual identified under 37 CFR 1.56(c) or 37 CFR 1.555(a) or any assignee has information reasonably necessary to the examination of the application or treatment of some matter, the examiner or other USPTO employee may require the submission of information that is not necessarily material to patentability.” Therefore, the USPTO’s discretion over requesting additional information regarding inventorship seems to be limitless. “Additional information regarding inventorship” indicates information sufficient for a determination of inventorship under the Pannu factors, which can be very fact intensive. The great concern for practitioners and applicants is that if this information cannot be provided to the Agency with sufficiency, which is determined at their discretion, the USPTO can then find that the inventorship question cannot be determined such that the patent or patent application is invalid for lack of proper inventorship.
In view of the possibility of a duty of candor being triggered, organizations may want to consider keeping contemporaneous records of AI use and asking inventors to certify their AI use. But asking inventors to keep records of AI use is very cumbersome and may be unfeasible in many instances. In fact, I have heard in-house counsel mention that requiring this kind of recordkeeping would cause a “revolt.”
Since the ultimate test for inventorship is the presence of significant contribution from a human inventor and NOT whether there is significant contribution from AI, as the Pannu factors are assessed with respect to human contribution and not the AI’s contribution, applicants should focus on providing records of human contribution rather than records of AI use, which may cause the USPTO to direct their focus on the degree and scale of AI use to the applicant’s detriment. For example, consider keeping clear records of design meetings, invention meetings, substantial experimentation, tests, modifications – particularly modifications of AI output, prototype development, as well as previous drafts, designs and prototypes, etc. The key is to demonstrate more than mere problem identification or appreciation of an invention by the human inventor(s). Applicants should focus on keeping records of having exercised expertise in the art and specific skills during the invention process on the part of the human(s) regardless of AI use.
Claiming AI in Light of the Enablement Guidance and the Means-Plus-Function
Overview
The Office issued the Enablement Guidance in view of the Supreme Court’s decision in Amgen Inc. et al. v. Sanofi et al. (hereafter Amgen) to inform USPTO personnel and the public on the Office’s implementation of the decision. Unlike the Enablement Guidance, the Means-Plus-Function Guidance was not prompted by a court decision but was to remind examiners of the resources available when examining limitations under 35 U.S.C. 112(f).
The Enablement Guidance and the Means-Plus-Function (together the 112 Guidance) should be examined for their impact on functional claiming, as functional claiming is often an (and sometimes the only) effective way to protect AI technology. When it comes to AI and AI-enabled technologies (e.g., internet of things), the physical components (e.g., hardware, machinery, etc.) are no longer provided on one thing. AI components are scattered – data processing and storage can take place on a cloud, machine transformation occurs remotely, output can be generated remotely. Traditional modes of positively claiming components do not protect most AI technology and pure method claims can be easy design arounds. Therefore, functional claiming and means-plus-function claims are useful and sometimes necessary for capturing the features of AI. In light of the criticality of functional claims for AI and AI-enabling technologies, it is important to understand the implications of the 112 Guidance.
In Amgen, the Supreme Court held that claims drawn to a genus of monoclonal antibodies, which were functionally claimed, were invalid due to a lack of enablement. The Supreme Court held that “[i]f 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”. The Enablement Guidance reiterates the Amgen Court’s determination that function can be deemed to encompass a genus and are therefore more subject to enablement considerations. Additionally, functional claims can be subject to being deemed genus claims if there are identifiable variants of performing said function which all also must be fully enabled under the Enablement Guidance and Amgen.
Regarding means-plus-function, the Means-Plus-Function Guidance reminds examiners to invoke the claim interpretation when a limitation recites functional language along with a generic placeholder term while failing to recite sufficiently definite structure for performing the function. Common nonce terms that are used in software claiming include “mechanism for,” “module for,” “device for,” “unit for,” “component for,” “element for,” “member for,” “apparatus for,” “machine for,” and “system for.” In order to avoid an indefiniteness rejection once a means-plus-function interpretation is invoked, the application needs to have provided disclosure of corresponding structure (e.g., hardware) that performs the entire claimed function. If there is no disclosure of structure for performing the entire recited function, the claim fails to satisfy the requirements of § 112(b) and should be found indefinite. The Means-Plus-Function Guidance specifies that for computer-implemented § 112(f) claim limitations, the specification must disclose an algorithm for performing the claimed specific computer function, or the claim should be found indefinite.
Practice Tips for Applicants and Practicing Attorneys
In view of the Enablement Guidance, practitioners should take care to identify whether functional claims can be deemed a genus when drafting claims, and then consider whether the disclosure has properly enabled all possible means of achieving the claimed function. Additionally, practitioners can consider modifying functional language to include a method limitation having at least one or more steps. That is, when trying to capture function, try drafting with some method, possibly even one step, followed by outcome which is the function. In this way, the claim is less likely interpreted as a genus claim, or at least a broad genus with innumerable variants or means that will be harder to enable. If the examiner argues that the steps of a function can be sub-genus or variants that need further enablement, that would be akin to arguing that the legs to a table are variants of the table which is improper.
Regarding means-plus-function, practitioners should also take care to identify all the potential “black boxes” of a disclosure that can be rampant in applications covering AI and machine learning (ML) technologies. The Patent Trial and Appeal Board has found that “machine learning classifier[s] and linear model[s] are like black box[es] with input coming in and produc[ing] output of tissue parameters. In these instances, the inputs going into the models need to be specified for each layer. Provide examples of input data and training data and how that data is transformed at each stage. Practitioners can also consider providing sample data sets. But when doing so, it is important to take care to not give off the impression that the data sets can be sufficiently processed by the human mind which can then endanger the invention as being deemed directed to a mental process.
It is also important to fully disclose and describe all neural network structure including network topology, activation functions (e.g., sigmoid, Tanh, ReLU), and layer connection types (e.g., fully connected layers, convolutional layers, etc.). Even if conventional, it is also advisable to include examples of neural network types (e.g., feedforward neural networks, convolutional neural networks, etc.) to meet the disclosure requirement because the architecture is needed to be coupled with data inputs and transformation to meet the disclosure requirement.
Terminal Disclaimer and Continuation Practice
Overview
The Proposed TD Rule sets forth that filing a terminal disclaimer will include an agreement that any patent granted on the subject application would be unenforceable if the patent is tied through the terminal disclaimer to another patent with a claim that has been held unpatentable or invalid under 35 USC §§ 102 or 103 and all appeal rights have been exhausted, or a statutory disclaimer of a claim is filed after any challenge based on anticipation or obviousness to that claim has been made. In laymen’s terms, the proposed rule would make it such that knocking out one claim in a patent based on prior art leads to knocking out the entire family of patents if the patents are tied by terminal disclaimer(s).
Shortly prior to the release of the Proposed TD Rule, the USPTO prefaced its hardened stance towards terminal disclaimers by the new Proposed TD Fees that introduced surcharges that could significantly increase the cost of waiting to file terminal disclaimers and continuations. Under the Proposed TD Fees, the cost of filing a terminal disclaimer is subject to a graduated fee structure after waiting for a first action on the merits. Particularly, if the applicant waited to file a terminal disclaimer is filed after a first office action on the merits, the cost is almost three times the current fee (194 percent increase). The surcharge increase jumps to four and a half times the current fee for waiting until after final action or allowance, then to eight times the current fee if filing after the patent issues.
At the backdrop of all this Agency action is the Federal Circuit’s decision in In re Cellect. Because Cellect did not file a terminal disclaimer in a later filed patent in which the claims should have been rejected for obviousness type double-patenting (ODP), the rejected claims of the later filed patent were invalid despite Cellect never receiving an ODP rejection during prosecution and, by the time it was raised during reexamination, they were left without the option because all of the patents had expired.
Practice Tips for Applicants and Practicing Attorneys
Applicants will certainly be extremely cautious of filing a terminal disclaimer if the Proposed TD Rule passes, as doing so would make all the claims of a patent family vulnerable to widespread invalidation. Yet, the Proposed TD Fees along with the Cellect decision augur expensive penalties for the applicant for not only refusing to file a terminal disclaimer, but for failing to preemptively file a terminal disclaimer even when ODP has not been raised. Prior to Cellect there was little to no incentive for applicants to file terminal disclaimers without an ODP rejection. But by introducing a treble fee surcharge for not disclaiming prior to a first office action on the merits, the USPTO is lending a ton of flex strength to the Cellect ruling because the applicant definitely could not have received an ODP prior to substantive examination.
So applicants are left with only horrible choices – file a terminal disclaimer after ODP and subject all the claims in the patent as far more vulnerable to invalidation, or attempt to traverse an ODP rejection (which can be very difficult) only to fail and have to file a terminal disclaimer anyway but now with a massive surcharge, or be good and file a terminal disclaimer upon filing the application and possibly play a guessing game as to the relevant reference patent when a large family is at play, such as the family in Cellect.
It almost seems as if the least horrible route is to avoid patent families going forward. When it comes to AI, it may be advisable to introduce a range of apparatus, system and method claims in separate applications as there can be so many ways to implement, deploy or use AI. Sometimes the applicant does not know which type of these claims is more valuable from an enforcement standpoint and will end up filing one or more continuations introducing different claims after initial filing as the technology matures on the marketplace. Therefore, it may be advisable to concurrently present these claims in separate applications and later abandon the one(s) that are more difficult to enforce. Alternatively, the applicant can introduce all of these types of claims in a single application and hope for a restriction, which will shield the subsequent divisional applications from an ODP rejection. Again, a restriction may not happen. There is also the fact that method claims are often more subject to being directed to a mental process for Section 101 purposes, which can make the entire invention vulnerable to being directed to a mental process and which is an added consideration for introducing method claims on their own. There is no clear path forward when it comes to dealing with this potential new age of continuation practice—only more risks and more costs.
Image Source: Deposit Photos
Author: gustavofrazao
Image ID: 71124389
