Patent Docs

USPTO Kicks Off Semiconductor Technology Pilot Program to Expedite Semiconductor Manufacturing Applications

By Aaron Gin --

Today, the U.S. Patent and Trademark Office announced a new, no-cost, pilot program that accelerates the examination process for U.S. patent applications related to semiconductor manufacturing.  Starting tomorrow Friday, December 1, 2023, applicants can file a petition to participate in the pilot program.  Petitions that correspond to qualifying non-provisional utility patent applications, directed to "certain processes and apparatuses for manufacturing semiconductor devices," will be advanced out of turn for examination (accorded special status) until a first action.

• Qualifying applications must be "non-continuing original utility non-provisional applications" that may optionally claim priority to a prior provisional application or one or more foreign priority applications.

• At least one claim of the qualifying application must cover a process or an apparatus for manufacturing a semiconductor device and correspond to one or more of the technical concepts within H10 (Semiconductor Devices; Electric Solid-State Devices Not Otherwise Provided For) or H01L (Semiconductor Devices Not Covered by Class H10) in the Cooperative Patent Classification (CPC) system.

• Applicants must file the petition to make special with the application or entry into the national stage under 35 U.S.C. § 371 or within 30 days of the filing date or entry date of the application.

• The fee for the petition to make special under 37 C.F.R. § 1.102(d) has been waived for this program.

In a related news release, the Secretary of Commerce Gina Raimondo stated that "[t]his new USPTO program will ensure we are prioritizing critical intellectual property protections to incentivize investments in domestic semiconductor manufacturing."  The Director of the USPTO, Kathi Vidal said of the new pilot program: "Fast tracking examination of patent applications related to semiconductor device manufacturing speeds bringing key innovations to market and strengthens our nation's supply chain."

The pilot program aims to support the Creating Helpful Incentives to Produce Semiconductors (CHIPS) Act of 2022, which allocates investments designed to increase semiconductor manufacturing capacity and improve the resilience of the semiconductor supply chain.  The Semiconductor Technology Pilot Program supports the CHIPS Act by "encouraging research, development, and innovation in the semiconductor manufacturing space and providing equitable intellectual property protection to incentivize investments in the semiconductor manufacturing area."  The USPTO stated that the new pilot program can help achieve the goals of the CHIPS Act by expediting and encouraging innovations relating to semiconductor devices and their manufacture.

The petition (Form PTO/SB/467) will be available starting December 1, 2023.  The pilot program will be available until December 2, 2024 or until the USPTO accepts 1,000 grantable petitions, whichever is reached first.

For additional information, please see:

• USPTO Semiconductor Technology Pilot Program
• Pre-publication of the Federal Register 2023-26340

Allgenesis Biotherapeutics Inc. v. Cloudbreak Therapeutics, Inc. (Fed. Cir. 2023)

By Kevin E. Noonan --

The Federal Circuit dismissed an appeal from an unsuccessful challenger in an inter partes review (IPR) proceeding based on failure to satisfy the standing requirements for appeal in Allgenesis Biotherapeutics Inc. v. Cloudbreak Therapeutics, Inc.

The case arose in an IPR over U.S. Patent No. 10,149,820, directed to compositions and methods for treating pterygium, "an eye condition in which a tumor-like growth extends from the nasal or temporal side of the eye to the cornea."  Surgery was the recognized treatment, and also recognized was the likelihood of tumor recurrence.  The '820 patent claims use of multikinase inhibitors including nintedanib, pazopanib, and sunitinib.

Nintedanib

Claims 4 and 5 were the only claims remaining after the IPR; claim 4 was set forth as representative:

1.  [disclaimed] A method for reducing hyperemia or symptoms thereof in pterygium in an affected eye of a subject in need of such treatment, without surgically excising a pterygium, comprising administering to the affected eye of the subject a therapeutically effective amount of a multikinase inhibitor.

3.  [disclaimed] The method of claim 1, wherein the multikinase inhibitor is administered to the affected eye in the form of topical ocular formulation or ocular implant.

4. The method of claim 3, wherein the multikinase inhibitor is nintedanib and the nintedanib is administered to the affected eye in the form of a topical ocular formulation and is administered topically to the affected eye.

Relevant to the issues on appeal, the Board held that Allgenesis had not shown the remaining (not disclaimed) claims of the '820 patent were invalid for anticipation or obviousness over its own PCT application because the earliest priority date for the '820 patent (June 5, 2015) was earlier than the priority date of the PCT application (June 22, 2015) and Cloudbreak's provisional application provided sufficient written description support for claims 4 and 5 of the '820 patent.  The Board also determined that the '820 claims were not obvious in view of the combination of two other references because of the '820 patent's disclosure of unexpected results of improved efficacy and safety.  Allgenesis appealed both grounds of the Board's determination of non-obviousness.

The Federal Circuit dismissed the appeal based on Allgenesis' failure to satisfy Article III requirements, in a decision by Chief Judge Moore joined by Judges Stoll and Cunningham.  The opinion sets forth the basis for Allgenesis' deficiency as being rooted in the requirement for standing of injury in fact under Spokeo, Inc. v. Robins, 578 U.S. 330, 338 (2016).  This requirement, while not arising for standing to file an IPR petition, Cuozzo Speed Techs., LLC v. Lee, 579 U.S. 261, 279 (2016), is required to appeal the Board's decision under Consumer Watchdog v. Wis. Alumni Rsch. Found., 753 F.3d 1258, 1261 (Fed. Cir. 2014).  Allgenesis asserted standing based on "(1) its potential infringement liability and (2) the Board's priority determination."  Neither assertion was sufficient, according to the opinion, because Allgenesis' allegations of injury were speculative and insufficiently concrete to establish injury in fact.  With regard to potential liability for infringement, the opinion states that the evidence, a declaration from Allgenesis' Vice President of Finance, relied on a Phase II clinical trial in 2020 and vague assertions of future plans to produce a nintedanib commercial product.  The opinion found these assertions to be unconvincing, saying that they "fail[] to identify any specific, concrete plans for Allgenesis to develop a nintedanib product that might implicate claims 4 and 5 of the '820 patent."  Nor did Allgenesis provide testimony regarding plans for further development such as a Phase III trial.  Instead, according to the panel, Allgenesis provided "generic" statements that it had not abandoned plans for a commercial product.  However, according to the opinion, "[s]uch conclusory testimony is insufficient to establish that Allgenesis has any concrete plans to develop and bring to market a nintedanib treatment for pterygium."  Nor did Allgenesis assert any threat from Cloudbreak to file suit for infringement.  The panel concludes that "Allgenesis has failed to establish it has nonspeculative, concrete plans for future activity that creates a substantial risk of future infringement."

Allgenesis was no more successful in its second assertion of injury in fact, that the Board's decision on priority of invention in favor of Cloudbreak would impair its ability to obtain its own patent on using nintedanib to treat pterygium.  Once again, the Federal Circuit found Allgenesis' evidence and argument to be insufficient to support its injury in fact claim.  The opinion analogizes the circumstances with the Court's  earlier decision in Best Medical International, Inc. v. Elekta Inc., 46 F.4th 1346 (Fed. Cir. 2022) (despite significant distinctions between the factual circumstances, recognized by the Court).  What the cases had in common, in the Federal Circuit's view, was reliance on collateral estoppel as the basis for the injury and, as the Court notes, "[c]ollateral estoppel will not attach to the Board's non-appealable priority determination," citing SkyHawke Techs., LLC v. Deca Int'l Corp., 828 F.3d 1373, 1376 (Fed. Cir. 2016).  And again the Court states that "Allgenesis has, based on these quite vague allegations, failed to establish a concrete injury."  Even when eschewing estoppel arguments Allgenesis "failed to articulate with any specificity how the Board's priority determination will impact its issued patents or pending continuation applications which claim priority to its PCT application," those articulations being found in "a single paragraph containing only vague allegations in its opening brief and reply brief, respectively."

Having decided that Allgenesis failed to establish injury in fact as required for standing the Federal Circuit dismissed the appeal without reaching the merits.

Allgenesis Biotherapeutics Inc. v. Cloudbreak Therapeutics, Inc. (Fed. Cir. 2023)
Panel: Chief Judge Moore and Circuit Judges Stoll and Cunningham
Opinion by Chief Judge Moore

College Football's Latest Controversy: NCAA IP-Like "Sign Stealing" Rules Not Ready for Big Data or AI

By Joshua Rich --

Aside from the actual games on the field, the college football press has been fixated on one story over the past several weeks:  the Michigan "sign stealing" controversy.  Michigan head coach (and former Chicago Bears quarterback) Jim Harbaugh has been suspended from appearing at the field for the last three games of the season -- including the critical game against Ohio State -- by the Big Ten conference, and may see further punishment when the NCAA completes its investigation.  But the entire process could have been prevented by a more innovative approach to technology, which could have brought even greater opportunities to forecast opposing teams' plays.

The existence of "signs", let alone stealing them, is a relatively recent innovation in the history of college football.  For decades, coaches chose a play and told it to a player on the sidelines, who would then run onto the field and convey it to the entire team in a huddle.[1]  As offensive coaches decided in 1980s and 1990s that they wanted to speed up the pace of play and limit defensive substitutions, they realized the biggest delay between plays was huddling.  So they started signaling plays to the team with hand signals.  However, those hand signals were not only clearly visible to the other team (and therefore readable and decipherable), but also limited to a relatively small number of plays that could be communicated by gestures.  Then came an innovation that allowed a wider variety of signals -- large placards or signs that could be held up by student assistants in numerous permutations and various orders to indicate different plays, as well as different formations.[2]  However, the problem of visibility of the signs remained.

The NCAA became concerned that the visibility of signs could create even greater inequality between college football teams because large, well-funded programs would be able to scout or "steal" signs at an opponent's games earlier in the season.  It therefore put in a rule in place in 1994 that bears some resemblance to trade secret rules.  Trade secret laws, such as the Federal Defend Trade Secrets Act ("DTSA"), define protectable trade secrets as any type of information if "(A) the owner thereof has taken reasonable measures to keep such information secret; and (B) the information derives independent economic value, actual or potential, from not being generally known to, and not being readily ascertainable through proper means by, another person who can obtain economic value from the disclosure or use of the information."[3]  Similarly, NCAA Bylaw 11.6.1 provides that "[o]ff-campus, in-person scouting of future opponents (in the same season) is prohibited."

The NCAA rule protects only information that is not known to the general public without a license (such as a game ticket).  For example, if a coach shouts the play from the sidelines so that it is audible to players on the field, there is no protection against the other team using that information; that is why the placards with seemingly random symbols have been adopted as "reasonable measures" of keeping the information secret.  But in this case, there are two "proper means" of scouting that avoid a complete ban.  First, in-game sign stealing (that, deciphering the signs while the game is being played) is permissible.  Second, reading signs from the television broadcast of an earlier game -- an activity that can be performed on campus -- is acceptable.  Furthermore, like trade secret laws, a team is allowed to use reverse engineering to determine the play from the publicly-visible signs.[4]

Michigan's problems with the NCAA's sign stealing rule arose out of a mysterious private investigation[5] of a Michigan staffer with the colorful name of Connor Stalions.  Stalions came to the Michigan football program from the Navy rather than the traditional route of playing the sport.  He appears to have started his scouting work in compliance with the rules, by watching recorded television broadcasts of games.  He allegedly eventually crossed far over the line, however, by buying tickets to opponents' earlier games and having associates record the sideline signals throughout the game with smartphones.  Allegedly, the complete-game videos were then synched to the on-field action to allow a correlation of the signs to the plays.  Stalions then would stand near the coaches during Michigan games, allegedly to communicate the upcoming play to them from the signs shown on the far sideline.

While such a scheme would clearly run afoul of the NCAA sign stealing rule -- if the allegations are true -- a crowdsourcing, big data approach might very well avoid any rule violation.  Many college football programs have rabid fanbases that would be willing to do almost anything to support their chosen team.[6]  If fans learned that a team would benefit from having recordings of the sidelines of upcoming opponents, many likely would be willing to make and provide such a recording.  They could then post the recoding on the internet (for example, hosted on YouTube) and coaches would be able to review and decipher the signs from on campus after the games.  So long as the fans were not employed or directed by the school, such a crowdsourcing approach would appear to be legal.[7]

In addition, crowdsourcing a future opponent's signs from several sources so that the correlation to plays could be made would allow deeper analysis of the data.  For example, once such a correlation is identified, an AI-based approach could not only allow prediction of which plays an opponent would like run in a given situation (with stolen signs providing a failsafe for the prediction) and identify the best players, formations, and plays to combat the opponent's plays.  Essentially, rather than relying on a coach's "gut instinct", an AI could serve as a predictive tool that would suggest the best approach -- essentially serving as a replacement for an existing coach.  Thus, one of the first jobs that could be replaced with AI would be one of the jobs that nobody has suggested would be vulnerable:  football coach.

[1] In fact, as Michigan has simplified its playbook and started calling more running plays -- 32 in a row against then sixth-ranked Penn State! -- it has reverted to huddling much more frequently and communicating plays to the huddle without signs.

[2] An example of such signs used by the Michigan football team last week are shown here:

[3] See 18 U.S.C. § 1839(3).

[4] 18 U.S.C. § 1839(6) provides that "the term 'improper means' . . . does not include reverse engineering, independent derivation, or any other lawful means of acquisition."

[5] It is not clear who retained and paid the private investigator who uncovered the information at the heart of the controversy.

[6] As one notorious example, in order to "punish" Auburn for a victory over Alabama, one Alabama fan poisoned the oak trees standing at Toomer's Corner by the Auburn campus merely because Auburn fans would celebrate victories by wrapping those trees in toilet paper.  The trees had otherwise played no part in the game.

[7] Other Big Ten teams, including Ohio State, are alleged to have done something similar by providing their in-game sign stealing to other teams playing against Michigan.

Happy Thanksgiving from Patent Docs

The authors and contributors of Patent Docs wish their readers and families a Happy Thanksgiving.

The authors and contributors of Patent Docs wish their readers and families a Happy Thanksgiving.

Medtronic, Inc. v. Teleflex Life Sciences Ltd. (Fed. Cir. 2023)

By Kevin E. Noonan --

In what was an otherwise run-of-the-mill affirmance of a decision by the Patent Trial and Appeal Board (PTAB) (albeit somewhat noteworthy in affirming the Board's determination that the challenged claims were not invalid), the Federal Circuit heard but deigned not to consider the question of whether claims to methods for achieving a therapeutic outcome needed to be supported by a showing of actual reduction to practice (i.e., that the method had been shown to be operative for the claimed result) in Medtronic, Inc. v. Teleflex Life Sciences Ltd. (precedential).

The case arose in two inter partes review decisions over a reissue patent U.S. Patent No. RE46,116 directed to methods for "using a guide extension catheter with a guide catheter."  A portion of claim 25 of the '116 patent was reproduced in the opinion as being representative:

25.  A method, comprising:
    advancing a distal end of a guide catheter having a lumen through a main blood vessel to an ostium of a coronary artery; . . .

The original patent was filed in 2006 and the Board's assessment of the sufficiency of Medtronic's validity challenge was carried out under the provisions of the 1952 Patent Act.  Accordingly, the Board considered evidence of conception in 2005 and diligence from its conception date until filing a patent application in May 2006 (constituting constructive reduction to practice).  Medtronic asserted U.S. Patent No. 7,604,612 against claims 52 and 53 of the '116 patent for anticipation (the Board held for Medtronic and this decision was not appealed); asserted the combination of the '612 patent and U.S. Patent No. 7,736,355 for obviousness of claims 25-40, 42, 44-48, 52 and 53; and asserted the '612 and '355 patents in combination with U.S. Patent Application Publication No. 2005/0015073 for obviousness of claim 45 (wherein in each case in which it was asserted Medtronic argued the '355 patent was prior art under 35 U.S.C. § 102(e)).*  The opinion notes that Medtronic did not challenge the sufficiency of Teleflex's evidence for conception prior to the critical date of the '355 patent (September 21, 2005).  With regard to the '355 patent, the Board held that Teleflex had reduced to practice the invention claimed in the remaining challenged claims prior to the filing date of the ‘355 patent and had also satisfied the diligence requirements for its May 2006 filing date to qualify as a constructive reduction to practice.  This decision, the opinion notes, was consistent with the Board's earlier decision on whether the '355 patent qualified as prior art for claims to Teleflex's corresponding apparatus claims.

The distinction ("unique to this case") that raised one of Medtronic's arguments in this appeal was the question of "whether or not in vivo testing was required for actual reduction to practice" because here the challenged claims recited the step of "advancing . . . a guide catheter . . . through a main blood vessel to an ostium of a coronary artery."  The Board held such evidence was not necessary and that evidence from alternative physical models could suffice.  This appeal followed.

The Federal Circuit affirmed, in an opinion by Judge Lourie, joined by Judges Prost and Chen.  The sole issues raised in this opinion were "(1) whether or not in vivo testing was required for actual reduction to practice and (2) whether or not the patentee exercised reasonably continuous diligence until constructive reduction to practice," wherein the panel determined that Teleflex had established constructive reduction to practice and did not reach the issue of actual reduction to practice (noting that a showing of either species of reduction to practice was enough to affirm the Board's opinion).  On this question, the Federal Circuit held that Medtronic had waived any challenge regarding the diligence issue by attempting to incorporate its arguments in that regard by reference to the Federal Circuit's judgment in a separate, related IPR.  The panel rejected this attempt, first, because in that related case the Court had not vacated the Board's decision on sufficiency of asserted diligence, and second, that trying to "incorporate by reference twenty pages from another brief in another case, amounting to over 4,000 extra words" was "a clear violation of both the motions panel's order [denying Medtronic's motion for leave to expand its brief to 20,000 words] and our rules" (emphasis in opinion).  The panel characterized this attempt as a strategic decision by Medtronic not to include arguments directed at the diligence issue into its principle brief and, its attempt having failed held that Medtronic's challenge of the Board's finding of diligence had been waived.  Combined with Medtronic's decision not to contest conception of the claimed invention the panel affirmed the Board's decision that the '355 patent was not prior art and that Medtronic's obviousness challenges failed.

*Medtronic asserted unsuccessfully these references as well as U.S. Patent Application Publication No. 2007/0260219 in a second IPR against the '116 patent, but Federal Circuit decisions in three related cases rendered moot Medtronic's appeal of the Board's decisions in this IPR, which was not further discussed in this opinion.

Medtronic, Inc. v. Teleflex Life Sciences Ltd. (Fed. Cir. 2023)
Panel: Circuit Judges Lourie, Prost, and Chen
Opinion by Circuit Judge Lourie

An Analogy for the Current Wave of AI Copyright Lawsuits

By Michael Borella --

We are at the beginning of what promises to be a wave (potentially a tsunami) of complaints filed against the companies behind generative AI models (e.g., OpenAI).  Recent lawsuits from Paul Tremblay and Mona Awad (Tremblay and Awad v. OpenAI Inc. et al. -- Northern District of California, No. 3:23-cv-03223), Sarah Silverman (Silverman v. OpenAI, Inc. -- Northern District of California, No. 3:23-cv-03416-AMO), and the Authors Guild (Authors Guild et al v. OpenAI Inc. et al. -- Southern District of New York, No. 1:23-cv-8292)[1] contend that OpenAI and others have hoovered up thousands of copyrighted publications, including those of the named plaintiffs, and used them to train large language models (LLMs) such as GPT-4.  As these initial cases proceed, and possibly go up on appeal, they are likely to define the contours of how copyright law applies to the new world of generative AI and whether is it proper to train such models on copyrighted works without permission to do so.

The authors' theories of infringement vary as do their ancillary claims.  While acknowledging the risk of over-simplify complex issues, we can boil the merits of these cases down to two main questions:

1)  Is the ingestion of a copyrighted work into the training process of an LLM without the author's permission an infringement of the copyright?

2)  What if an LLM trained in this fashion produces a new work that is substantially similar to the copyrighted work?

These questions can be thought in terms of pigs and sausage.[2]  Pigs can be turned into sausage, but it is generally accepted to be impossible to turn sausage back into a pig.  Mathematicians would consider the transformation from pig to sausage to be an irreversible one-way function.

It is important to understand that all computer data are just organized collections of numbers.  This includes digital copies of books, images, audio, video, web sites, etc.  When a machine learning model such as an LLM is trained on a digital book, the arrangement of numbers representing the words, punctuation, front matter, and so on are transformed into a different arrangement of numbers -- weights in a complex set of neural networks.

In most cases, there is no one-to-one mapping between the numbers used before and after transformation.  One cannot point to a particular set of numbers in an LLM and identify a Game of Thrones novel.  Indeed, the weights in an LLM are a complex amalgam of most or all data on which it was trained.  Even the entities that design and build LLMs have yet to provide an understanding of what the weights actually represent.

So this leads to a likely answer to the first question.  A similar set of facts were considered by the Second Circuit in Authors Guild, Inc. v. Google, Inc., in the context of using copyrighted books for search purposes.  The Court ultimately ruled that the conversion of the copyrighted content into a form useful for searching was highly transformative, displaying small portions of the books was fair use, and such search and display did not provide a significant market substitute for the original works.  Therefore, the mere use of a copyrighted work to train an LLM, even without permission, is unlikely to be a winning fact pattern.

But the emergent magic of LLMs is that they might know enough about an ingested Games of Thrones novel to be able to produce its plot summary, a list of main characters, and even quote a section or two.[3]  These uses might also fall under the Second Circuit's definition of fair use.

But an LLM may be able to produce significant portions of the work or the work as a whole.[4]  Or, the LLM may be able to generate alternative endings to the novel, new works in the style of the author, or new works involving the same characters and relying on the authors' world-building.

Thus, the answer to the second question is not clear, though it seems that the LLM would have to provide "more than just a little" of the copyrighted work.  For example, copyright famously protects actual works and not styles.  This issue may boil down to whether an LLM can reverse the transformation function and turn sausage back into a reasonable semblance of a pig, as well as whether an LLM operator can successfully prevent it from doing so.

As noted, the cases currently being litigated may provide some clarity -- or, depending on how they proceed, maybe not.  Also, Congress may step in and define new causes of action that specifically target LLMs and similar fact patterns.

Authors may ultimately have their strongest positions where they can argue that the operator of the LLM is unjustly enriching itself on the backs of the authors' labor or effectively competing in the same marketplace as the authors.  At first blush it seems that imaging tools based on generative AI (e.g., DALL-E, Midjourney, and others), the use of which can eliminate the need for human illustrators, might be a better target for such claims.

[1] Here, the group of authors named in the complaint include John Grisham, George R. R. Martin, Jodi Picoult, and Scott Turow.

[2] Vegans should feel free to replace "pigs" with "plant-based protein."

[3] OpenAI appears to be aware of the issues that this capability might raise.  If you ask ChatGPT 4 to "provide a Jon Snow quote from Game of Thrones," it falls back on a Bing search to do so.

[4] This is theoretically possible, though OpenAI and others have put guardrails in place in attempts to prevent their models from such blatant infringement.

International Trade Commission Issues Report of COVID-19 IP Waiver

By Kevin E. Noonan --

It came as something of a shock in May, 2021 when the Biden administration voiced support for waiver of intellectual property protection mandated under the TRIPS agreement for COVID-19 vaccines (see "Biden Administration Supports Waiver of IP Protection for COVID-19 Vaccines").  The waiver had originally been proposed to the World Trade Organization (WTO) by India and South Africa; in that incarnation a much broader swath of intellectual property protections had been proposed (encompassing diagnostics and therapeutic drugs related to the pandemic).  When eventually imposed under the WTO's June 17, 2022 Ministerial Decision, the waiver was limited to vaccines with further consideration of waivers having broader scope postponed (until later this year, in fact).

In anticipation of negotiations over expanding the IP waiver, the U.S. Trade Representative, Ambassador Katherine Tai sent a letter on December 16, 2022 to the International Trade Commission requesting the ITC to prepare a Report under section 332(g) of the Tariff Act of 1930 on the needs and justification for further expansion of the IP waiver.  The Trade Representative asked that the ITC's Report be a product of market and other research within the Commission's expertise and also reflect the views of stakeholders and the public, to be garnered from public hearings and solicited comments from "foreign governments, non- governmental health advocates, organizations such as MPP and Foundation for Innovative New Diagnostics(FIND), and diagnostic and therapeutic manufacturers."  The letter expressly disclaims any request for the Report to "draw any policy conclusions" and asked for the Report to be provided no later than October 17, 2023.

The ITC (Chairman David S. Johanson, Rhonda K. Schmidtlein, Jason E. Kearns, Randolph J. Stayin, and Amy A. Karpel) issued its almost 500-page Report, entitled "COVID-19 Diagnostics and Therapeutics: Supply, Demand, and TRIPS Agreement Flexibilities," on the deadline date (Ambassador's Tai's December 16, 2022 letter can be found as Appendix A to the Report).  Containing as a prelude a list of abbreviations and glossary (necessary and advantageous for a report from any government agency or commission), which also identifies various companies, organizations, agencies in the U.S. and abroad, regulations, and proposals, the Report provides a succinct Executive Summary.  Contained in this Summary are the sentiments that the COVID-19 pandemic had caused almost 7 million deaths worldwide and that there was a possibility (espoused by some) that IP could "act as a barrier to access, particularly in developing countries."  Several highlights of this summary are:

• The TRIPS Agreement sets minimum standards for the protection and enforcement of IP rights. It also includes flexibilities (although there were differences of opinion regarding the scope of those flexibilities).

• Determining a definitive scope of what products are covered by the terms "diagnostics" and "therapeutics" as they pertain to COVID-19 and what constitutes relevant COVID-19 diagnostics and therapeutics covered by patents is complicated and subject to interpretation.

• The development and commercialization of COVID-19 diagnostics and therapeutics occurred on an extremely compressed timeline.  Manufacturing of diagnostics and therapeutics involves multiple stages, each of which requires careful attention to detail and strict quality control measures.  The two fields of diagnostics and therapeutics are composed of different producers, inputs, know-how, and so on (recognizing the unique characteristics of COVID-19 and thus suggesting caution in applying this recent history to future pandemics, but the commonalities such as diagnostics being marketed sooner than therapeutics).  Data points: As of summer 2023, China (UMIC) reportedly had the highest number of manufacturers of COVID-19 diagnostics (247), and India (LMIC) had the largest number of reported therapeutics manufacturers (56).

• A variety of advantages and challenges are associated with the use of voluntary licenses to provide access to IP associated with COVID-19 therapeutics and compulsory licenses to access COVID-19 therapeutics and other patented pharmaceutical products (wherein neither licensing type was used during the pandemic).

• The availability of supplies to meet global demand for COVID-19 diagnostics and therapeutics has been a moving target throughout the pandemic (market demand or need?  Varied over the time course of the pandemic).

• The disparity among countries of different income groups is wide in terms of access and availability to COVID-19 diagnostics and therapeutics (regarding government procurements: 80% HIC; 14% UMICs, 5% LMICs, none by LICs but multilateral organizations help in LICs. Recognition that this is multivariable, depending on "access to IP, prices and affordability, regulatory approvals, healthcare infrastructure, and the healthcare priorities of governments").

• Academic literature on the effects of patent protection, compulsory licenses, and the MPP is limited and would benefit from additional research ("From the available evidence, patent protection is generally found to be more beneficial to innovation in the health sector for developed countries and less so for developing countries.")

The Executive Summary also sets forth both the source of the information considered by the Commission (from academic and governmental literature, public and private sources, more than 120 interviews with stakeholders internationally, public hearings held March 29-20, 2023, and over 300 written submissions) and gaps in that information, which vary by country and region ("greatly"), in particular with regard to production costs.  Also noted were gaps in information regarding IP-related agreements, including compulsory licenses and the existence of "very little price transparency" (which raises the question of the factual basis for the purported of the necessity for the waiver in the first place).  Nor was there a great deal of information on consumption of COVID-19 related diagnostic or therapeutic moieties outside high income countries.

The Executive Summary also contains a brief review of IP and the TRIPS agreement, including an even briefer history of WTO grant of compulsory licenses in limited instances for pharmaceutical products.  This portion of the Report identifies IP protections for COVID-19 related diagnostics and therapeutics (and expiry of the foundation IP on diagnostic methods) ("the best evidence that patents did not act as a primary barrier to the production of these tests is that many new producers and products, including in developing countries, came online relatively quickly").  The results in the Report include a "broad and varied" (and indeterminate) universe of the diagnostics that include "(1) tests used to diagnose an active infection of the SARS-CoV-2 virus, (2) tests that measure an adaptive immune response to the virus, and (3) tests used in the management of patients with COVID-19" and ancillary products (such as swabs, etc.).  Therapeutic entities likewise include the drugs and delivery vehicles (syringes, etc.) for administering them.  This portion of the Summary is devoid of any recognition or discussion of vaccines, in view of the mandate from the Trade Representative and the existing vaccine waiver.  The Report provides schematic representations of how diagnostics and therapeutics were characterized:

With regard to supply chain issues (which in some ways were even more critical post-pandemic), the Report recognizes the multinational, interdependent nature of the pharmaceutical industry (see Cott et al., "How Pfizer Makes Its Covid-19 Vaccine," New York Times, April 28, 2021), and the complexities created by this situation.  As of June/July 2023, the Report estimates there were 900 diagnostics manufacturers and 221 therapeutics manufacturers.

This Summary of the Report also addresses the source of COVID-related IP, which include R&D agreements, manufacturing partnerships, and licensing agreements (only R&D agreements were informative for diagnostics technology).  While small- and medium-sized entities (SME) were responsible for development of new COVID therapeutics, larger pharmaceutical companies and the Federal government supplied economic and development support (and, according to the Report, was a source of calls for the resulting products and IP to be broadly and more cheaply available and those contracts be more transparent).  Voluntary licenses provide one source of IP, which would disclose not only the patented invention but also technical information, know-how, and trade secrets the disclosure of which would undoubtedly raise resistance to this disclosure.  While geographical restrictions are discussed, the Report also somewhat unrealistically posits royalty-free licenses as a norm.  One advantage of such licenses noted in the Report was somewhat more fulsome disclosure than was obtained from compulsory licenses.  Also mentioned with approval was disclosure though the Medicines Patent Pool ("a United Nations-backed public health organization").  So-called "upper-middle-income countries," defined by the World Bank as those economies with incomes per capita between $4,256 and $13,205, were not included in voluntary licenses for four COVID-19 therapeutics.

Turning to compulsory licenses, the Report notes that while the subject and provisions for and under which compulsory licenses for COVID-related IP was the subject of the 2022 Ministerial decision, no compulsory licenses to patents related to COVID-19 vaccines have been granted (although such licenses to IP relating to COVID-19 therapeutics have been granted in Hungary, Israel, and Russia).  The Report briefly sets forth purported advantages and real challenges that arise with using compulsory licenses, including for advantages reduced price, development of domestic manufacturing capacity, and "leverage" for negotiating with rightsholders.  The Report recognizes the limitation (for now) that compulsory licenses involve solely patents and do not grant access to know-how and trade secrets (which the Report notes can be particularly important for biologic drugs).  The Report acknowledges that compulsory licenses can be granted in the context of lack of availability of a qualified manufacturer and result in economic and political pressure and legal challenges under domestic law or TRIPS (although the latter may be something of a paper tiger in the current political climate).

The Report particularly acknowledges issues in "least developed countries" (LDC) as classified by the United Nations.  Bangladesh, Laos, Myanmar, and Paraguay are provided as exemplars for taking advantage of the exceptions to TRIPS requirements related to IP, which in the COVID context has been used as a pretext for producing generic versions of nirmatrelvir (+ ritonavir), remdesivir, and baricitinib and exporting these drugs to countries without patent protection for them.

With regard to the availability and consumption of COVID-19 diagnostics and therapeutics, the Report assesses both parameters through the lens of price, actual purchase, donations, and consumption of diagnostics and therapeutics, as illustrated in this schematic for therapeutics:

For diagnostics, the data reviewed (albeit subject to inconsistencies in different countries, from different manufacturers, and different times) indicate government procurement and distribution of test kits (in the U.S. pledged to provide $1 billion in kits free of charge).  Multilateral organizations made tests broadly available to middle- and low-income countries during the pandemic, which have reduced substantially by mid-2023, as the pandemic seems to have waned and WHO has ended the public health emergency.

For therapeutics, the Report found very little price transparency but notes a tiered-pricing regime depending on income level, with low-income countries paying substantially less.  Except for high income countries like the U.S. and Europe COVID therapeutics are obtained/procured through donations or purchase from governments (79% of 98 million treatment courses) and multinational organizations.  Pfizer, Eli Lilly, and Merck have donated COVID therapeutics to low- and middle-income countries.  Consistent with these data the Report states that "many consider the prices of COVID-19 therapeutics to be a significant barrier to access for many" countries (other than high income countries, of course), illustrating the challenge of therapeutic cost with Pfizer's nirmatrelvir (+ ritonavir) treatment, which at $250 per treatment course is about half of the average per capita healthcare expenditure across all upper-middle income countries.  Coupled with this disincentive the waning of the pandemic has caused a deep reduction in demand for these drugs.  Even when demand was higher regulatory approvals in each country caused delays, and many countries have pre-existing public health demands (HIV/AIDS, malaria, and tuberculosis are given as examples) that limited availability of COVID-19 drugs.

The Report next sets forth synopses of the views of "interested persons" on "(1) IP protection, R&D, and jobs; (2) the TRIPS Agreement and access to medicine; and (3) the TRIPS Agreement and COVID-19 diagnostics and therapeutics" obtained for eight topics in each of these categories.  Not unexpectedly, the Report asserts that while some respondents attested to the beneficial effect of IP on development of technologies critical to addressing COVID-19 and future pandemics, others expressed the belief that extending the 2022 Ministerial Decision would not have a deleterious effect on the supply of diagnostics or therapeutics.  Likewise respondents could be split into those who believe the IP protections in TRIPS facilitated access to medicines (including COVID-related diagnostics and therapeutics by incentivizing investment) while others thought the purported increased cost had the opposite effect.  The pros and cons of compulsory licensing were also addressed and some respondents asserted that further clarification on the "flexibilities" in IP protection would be a useful pathway towards improving access to medicines worldwide, particularly in low- and middle-income countries.  Respondents were in agreement, however, that "waiving or weakening IP rights on COVID-19 diagnostics and therapeutics would reduce incentives to invest in the R&D of tests and treatments for future pandemics."  Nevertheless, respondents split reliably on lines wherein some advocated that IP protection was necessary to be prepare for future pandemic (providing incentives and financial rewards for innovation) while others stated that the 2022 Ministerial Decision should be extended to encompass future drug development.  The former respondent group warned that "an extension of the 2022 Ministerial Decision to COVID-19 diagnostics and therapeutics would cover an overly broad set of diagnostics and therapeutics and reduce incentives for companies to look for COVID-19 applications for their products."  Respondents also submitted statistics on the effect of the Decision on employment in 26 countries (including the U.S.) but these statistics included jobs unrelated to COVID-19 diagnostics and therapeutics and thus were of limited value.

At the request of the Trade Representative the Report contains a "literature review" on four topics: innovation in the health sector and access to medicines in relation to patent protection; the effects of compulsory licensing; and the effects of the Medicines Patent Pool (this topic having received limited academic study that focused in the main on HIV/AIDS drugs).  These reviews found little consensus in the current academic literature on these topics, the Report characterizing them as "limited" and that would "benefit from additional research on these topics," due in part to "limited data availability" and "difficulty identifying the causal effect" of IP on these topics.

The remainder of the Report comprises the following sections.

Chapter 1: Introduction: explaining that the Report was in response to a letter dated December 16, 2022 from the U.S. Trade Representative asking the ITC to perform the study in response, in part, to the June 17, 2022 Ministerial Decision providing a waiver of certain TRIPS protections for intellectual property in response to a joint communique from India and South Africa.  The scope and intention of the December 2022 letter indicated that the ITC's Report would be used by the U.S. before the WTO in its deliberations regarding whether to extend the IP waiver to include COVID-related diagnostics and therapeutics.  In making these determinations, economies for WTO members were set forth according to this schema:

(where HIC = high-income countries; UMIC = upper-middle-income countries; LMIC = lower-middle-income countries; LIC = low-income countries).  This section provides cumulative information that as of August 2023 the COVID pandemic had produced 769 million confirmed cases and almost 7 million deaths.

Chapter 2: Background on IP and Regulations: providing background information on TRIPS and mandated IP protections required thereby.  Notably, this section discloses that the "flexibilities" regarding IP protection of TRIPS Articles 31 and 31bis and the Doha Declaration (2001) have resulted in "172 instances (122 compulsory licenses; 46 extended transition implementation period until 2033 for LDCs; 3 exceptions to patent rights under TRIPS Article 30; and 1 instance of parallel importation) in which governments used or attempted to use TRIPS Agreement flexibilities for public health purposes as of May 2023.  And modifications to the Doha declaration in 2003 permitted compulsory licenses to extend to production for export as well as domestic applications.  This Chapter also sets forth the particulars of the 2022 Ministerial Declaration (in particular mentioning that the original proposal included waiver of "undisclosed information" meaning trade secrets), in apparent contradiction to the provisions of TRIPS Article 39.3.  Finally in this regard, this Chapter sets forth information from both those in favor and against the 2022 Ministerial Declaration that IP rights did little or nothing to inhibit diagnostic testing for COVID-19 during the pandemic, due to expiration of most applicable IP.

The Report also discloses landscape searches performed by WIPO and the USPTO (824 COVID-19 diagnostic-specific published patent filings at the USPTO between December 2019 and the end of March 2023, in which 47% of relevant patent families were filed in China) directed to diagnostics.  On the other hand, WIPO reported 4,787 applications filed between January 2020 and September 2022 directed to COVID-19 therapeutics.

This Chapter concludes with an overview of the drug regulatory landscape worldwide (in each individual country and WHO) and the status of regulatory approval for COVID-19 related diagnostics and therapeutics.  Five COVID therapeutics have been granted prequalification by WHO:  tocilizumab, dexamethasone, molnupiravir, remdesivir, and nirmatrelvir (+ ritonavir), and FDA has authorized 278 molecular diagnostic tests and 64 antigen tests under EUAs as of July 2023.

Chapter 3: Definitions and the Universe or COVID-19 Diagnostics and Therapeutics (vaccines were not included in the Trade Representative's Request letter):  specifically sets forth "the range of definitions for diagnostics and therapeutics in the medical field; (2) identifies and defines the universe of existing COVID-19 diagnostics and therapeutics covered by patents, as well as diagnostics and therapeutics in development; and (3) provides a broad overview of relevant COVID-19 diagnostics and therapeutics," acknowledging that these definitions are more complex than identifying vaccines, at least with regard to functional properties.  These data are illustrated in several tables in the Chapter.

Chapter 4: COVID-19 Diagnostics and Therapeutics Manufacturing Supply Chain and Trade: provides information "describing the development, production, and trade of COVID-19 diagnostics and therapeutics; discussing the regulations needed to bring these goods to the global market; and providing an overview of the key producing countries and firms operating in the supply chain."  Diagnostics were limited to PCR and rapid antigen tests, while therapeutics were limited to virus-directed drugs on the market during the pandemic (January 2020 to July 2023), including both small molecule and biologic drugs.  The Report acknowledges the limitations in the data for a variety of reasons.

Chapter 5: Approaches to Access the Intellectual Property Associated with COVID-19 Diagnostics and Therapeutics: This Chapter is focused more on therapeutics than diagnostics and addresses ways to accesses COVID-related IP, including "research and development (R&D) agreements, manufacturing partnerships, bilateral license agreements (BLAs) between firms, and license agreements involving the Medicines Patent Pool (MPP), a United Nations (UN)-backed public health organization" established in 2010 (these agreements comprising voluntary licensing).  The consequences and benefits ("advantages and challenges") of compulsory licenses are also discussed.  Statistics provided include that "87 percent of COVID-19 therapeutic R&D programs in the United States, and 75 percent of products in the global clinical development pipeline, originate from U.S.-based small- and medium-sized enterprises (SMEs)," according to the Biotechnology Innovation Organization (BIO).  Also mentioned is the involvement of the Federal government in producing COVID-related therapeutics.  Specific BLA agreements (to the extent they are public) discussed include Gilead/remdesivir, Merck/molnupiravir, and Eli Lilly/baricitinib, particularly as a means to avoid compulsory licenses and regulatory delays.  Finally, the Chapter contains an extensive discussion of the advantages of using the MPP for voluntary licensing, stating:

Rather than negotiating individual license agreements directly with producers, licensors authorize the MPP to sublicense their technologies to third parties.  In turn, the MPP solicits applications for sublicense agreements, reviews prospective producers, and handles all aspects of implementation of the sublicense agreements with selected companies.

(although conceding that academic research on the MPP is limited and what exists is focused on HIV/AIDS drugs).  Merck is noted as the first company to enter into an MPP license in October, 2021 for molnupiravir (having the advantage that sublicenses come with a regulatory waiver), followed by Pfizer for nirmatrelvir (+ ritonavir) in November 2021, and Shionogi for ensitrelvir in October 2022.  Examples of these MMP licenses are set forth in the following table:

And the geographic distribution of MPP licenses shown in this diagram:

The Report also describes compulsory licenses which have not been used with COVID vaccines or diagnostics but have been imposed on COVID therapeutics.  According to the Report, there have been approximately 121 attempts to issue a public health compulsory license (including 4 by the U.S.) since 2001, but only about 75% of threatened compulsory licenses were imposed, as illustrated by this diagram:

Advantages of such licenses include lowering drug prices, providing leverage in negotiations, and in some instances improving domestic manufacturing capacity.  Of course, compulsory licensing has also resulted in retaliation against countries imposing them, including reducing investment or withholding drug products from the national market.  Compulsory licenses are also imperfect in having the desired effect because they are limited to patents and cannot expropriate know-how necessary to produce a drug product.  Specific instances of compulsory licensing discussed in the Report include Zambia, India, Brazil, Ecuador, Malaysia, South Africa, Thailand, and Canada (in this instance for importing HIV/AIDS drugs to Rwanda).  And the effects of compulsory licenses on innovation are disputed by researchers according to the Report.

Chapter 6: Availability and Consumption of COVID-19 Diagnostics and Therapeutics: This Chapter provides the Trade Representative with information regarding the necessity (or lack of it) of a COVID waiver extension based on global needs.  This discussion is presented before the backdrop of fundamental disagreements between proponents and opponents of the waiver over the causes and even the existence of global deficiencies in providing COVID-19 diagnostics and therapeutics, arguing insufficient demand or insufficient (affordable) supply, with evidence on both sides of the debate.  Needs assessments are limited almost entirely to therapeutics, there being no analogous studies or reports for diagnostics.  The Report provides statistics relating to prices, purchases (including governmental purchases and multilateral programs and private sector purchases), donations, and consumption for diagnostics and therapeutics as well as factors affecting each of these statistics (such as "last mile delivery" effects).  Regulatory authorizations (and needs therefor) as well as the effects of competing healthcare priorities in many countries is also discussed in the Report.

Chapter 7: Views of Interested Persons: This Chapter provides synopses and examples of public input requested by the Trade Representative.  Providing a cautionary warning the Report states that "[t]he summaries of public views provided in this chapter do not attempt to assess, analyze, or draw conclusions about these views" and that "this chapter often uses language as provided by participants to avoid changing the meaning of their submitted views" and thus "some wording used in this chapter may not match wording used in other chapters of this report."  The structure of the Report in this chapter is as follows:

IP protection, R&D, and jobs

• The relationship between IP protection and corporate R&D expenditures, taking into account other expenditures, such as share buybacks, dividends, and marketing

• The location of jobs associated with the manufacturing of diagnostics and therapeutics, including in the United States

The TRIPS Agreement and access to medicine

• Whether and how existing TRIPS rules and flexibilities can be deployed to improve access to medicines

• Successes and challenges in using existing TRIPS flexibilities

• To what extent further clarifications of existing TRIPS flexibilities would be useful in improving access to medicines

The TRIPS Agreement and COVID-19 diagnostics and therapeutics

• How the TRIPS Agreement promotes innovation in and/or limits access to COVID-19 diagnostics and therapeutics

• The extent to which products not yet on the market, or new uses for existing products, could be affected by an extension of the Ministerial Decision to COVID-19 diagnostics and therapeutics

• The relevance, if any, of the fact that diagnostic and therapeutic products used with respect to COVID-19 may also have application to other diseases

Appendix D of the Report contains the positions and policy statements of witnesses at the public hearings

The Report is a wealth of information and data (albeit wherein some of the data is incomplete) and abides by the request by the Trade Representative not to espouse any particular course of action or outcome.  Accordingly, the Report can be expected to be a useful tool during negotiations.  It is the reality, however, that whatever course the Administration takes will, in the absence of fact- or data-driven outcomes predominantly be a political decision based more of the concessions that can be obtained than any demonstrated need for any extension of the IP waiver from these data.  Should increasing the waiver come to pass the consequences are unlikely to be immediately felt, unless another virulent COVID-19 variant arises that cause higher morbidity and mortality than the current crop of variant COVID-19 species has developed.  The greater risk is in the precedent that a widened waiver would set and whether such a waiver would in the next pandemic reduce the capability for the type of rapid development of vaccines, diagnostics, and therapeutics that have (at least for now) stemmed the tide of the COVID-19 pandemic to manageable levels.  That is an outcome that will only become apparent through experience, and such an experience may be very costly indeed and completely unpredictably so.  Prudence suggests it would be wise to court such an outcome carefully.

For additional information regarding this topic, please see:

• "Top Four Stories of 2022," January 18, 2023
• "WTO TRIPS Council Recommends That General Council Extend Waiver Deadline," December 18, 2022
• "Nine Countries Seek Extension of WTO Waiver to COVID-19 Therapeutics and Diagnostics," December 11, 2022
• "Status of Proposed Extension of TRIPS Waiver in WTO," December 8, 2022
• "C4IP Presents Webinar on COVID Waiver Extension," December 5, 2022
• "Senators Send Letter to Commerce Secretary Regarding WTO Waiver Compromise," March 28, 2022
• "The Proposed WTO IP Waiver: Just What Good Can It Do? -- An Analysis," March 24, 2022
• "IP Associations "Concerned" by Reports of TRIPS Waiver Compromise," March 24, 2022
• "More on Leaked WTO COVID-19 Vaccine Patent Waiver Compromise," March 21, 2022
• "Compromise Reportedly Reached on COVID-19 Vaccine Patent Waiver," March 16, 2022
• "Sen. Tillis Writes to U.S. Trade Representative (Again) Regarding TRIPS Waiver," December 12, 2021
• "U.S. Trade Representative Responds to Letters from Senators Regarding TRIPS Waiver," November 14, 2021
• "U.S. Chamber of Commerce Urges Administration to 'Double Down' on Global Vaccine Distribution," November 3, 2021
• "Is This the WTO Waiver End Game?" July 25, 2021
• "BIO Declaration on Global Access to COVID-19 Vaccines and Treatments and Role of IP," June 24, 2021
• "GOP Legislators Write in Opposition to Proposed TRIPS Waiver," May 16, 2021
• "Population of Patents at Risk from Proposed WTO Patent Waiver," May 12, 2021
• "Sen. Daines Urges Biden Administration to Withdraw Support for COVID-19 IP Waiver," May 12, 2021
• "Pfizer CEO Pens Open Letter on COVID-19 Vaccine IP Waiver," May 10, 2021
• "If the Devil of the WTO IP Waiver Is in the Details, What Are the Details?" May 9, 2021
• "The Road to Hell Is Paved with What Everybody Knows," May 6, 2021
• "BIO & IPO Issue Statements on Biden Administration's Support for Proposed WTO Waiver," May 6, 2021
• "Biden Administration Supports Waiver of IP Protection for COVID-19 Vaccines," May 5, 2021
• "Suspending IP Protection: A Bad Idea (That Won't Achieve Its Desired Goals)," April 26, 2021
• "Sen. Tillis Asks Biden Administration to Oppose WTO Waiver Proposal," April 21, 2021
• "IP Organizations Support Continued Opposition to Waiver Proposal," April 5, 2021
• "Industry Coalition Supports Continued Efforts to Oppose Waiver Proposal," March 29, 2021
• "BIO and PhRMA Urge Biden Administration to Oppose Proposed WTO TRIPS Waiver," March 11, 2021
• "IPO Sends Letter on IP Law and Policy to President-Elect and Vice President-Elect," January 4, 2021

Early Responses to Proposed AI Regulation

By Michael Borella --

As a follow up to last week's article about the Biden executive order on artificial intelligence (AI), this is a brief overview of one of its provisions that has proven to be controversial -- namely, the additional scrutiny that the government plans for AI models and computing clusters over a certain size.  The order states that the Secretary of Commerce shall require that entities acquiring, developing, training, or using such models must report their activities to the government, including results of red-team testing intended to discover flaws and vulnerabilities of these systems.

The current size thresholds are:

(i)   any model that was trained using a quantity of computing power greater than 10²⁶ integer or floating-point operations, or using primarily biological sequence data and using a quantity of computing power greater than 10²³ integer or floating-point operations; and

(ii)  any computing cluster that has a set of machines physically co-located in a single datacenter, transitively connected by data center networking of over 100 Gbit/s, and having a theoretical maximum computing capacity of 10²⁰ integer or floating-point operations per second for training AI.

For sake of comparison, today's high-end PCs typically can maintain a few trillion floating-point operations per second (teraflops), which is around 10¹² floating-point operations.  The NVIDIA DGX H100 server, which is specifically designed for supporting AI uses, is rated at 32 petaflops, which is around 10¹⁶ floating-point operations.  Thus, to reach the 10²⁰ floating-point operation threshold, one would need about 10000 DGX H100s.  At a price of about $400,000 each, it would require exceptionally deep pockets to even consider building such a system, and this does not even take into account the cost of electrical power.[1]

According to some reports, OpenAI's GPT-3 was trained on a cluster that has about 800 petaflops of computing power (just under 10¹⁸ floating-point operations per second).[2]  Hitting the regulatory threshold would require a system with about 100 times more capacity.  OpenAI did not release the technical details of its training cluster for GPT-4 but did indicate that the training cost the company over $100 million.[3]

It is safe to say that only a handful of non-governmental entities currently have the ability to train models that would trigger the reporting required by the order.  But, if anything, we have learned over the last several decades that the amount of computing power available per dollar increases by an order of magnitude every few years.  Thus, the 10²⁰ floating-point operation threshold may be within reach of many entities in the near future.

Despite the call for regulation coming from OpenAI and other leaders in generative AI, there has been some pushback.  So far, this is largely originating with the open source community and venture capitalists.

Researcher Andrew Ng is concerned that regulations based on model size are an overreaction to the dangers of AI, would stifle innovation, and would advantage the incumbent players.[4]  Researcher Yann LeCun, with a number of other tech executives and venture capitalists, sent a letter to President Biden arguing that the order would result in AI being dominated by a few well-heeled companies.[5]  The letter also urges the president to consider the benefits that open source AI models can bring to the public by increasing marketplace competition while opening development and use of the technology to academia and individuals.

The argument for taking a lighter regulatory touch on open source AI models is that they foster collaboration, transparency, and security by democratizing the use of AI.  An open source mantra is that "given enough eyeballs, all bugs are shallow."  This can be interpreted to mean that the world's programming community being able to inspect these models would increase their reliability and usefulness while reducing harm.  Yet, by their very nature, open source initiatives are unlikely to have the resources to engage with regulators.  Furthermore, the concern over the most powerful AI in the world being black boxes understood and controlled by only a handful of people is very real.

While open source has proven to be an incredibly powerful tool for advancing technology and bringing it to the masses -- note the rise of Linux over the last 30 years -- a blind adherence to its principles ignores some obvious risks.  Bad actors of all stripes could obtain and customize generative AI models at a low cost, potentially removing any guardrails and using them to produce an unlimited amount of misinformation, deepfakes, and cyberattacks, or even to design novel bioweapons.  There is a big difference between open sourcing an operating system or web server versus open sourcing a large language model with even a fraction of GPT-4's capabilities.

At this juncture, we appear to be between the devil and the deep blue sea.  The question comes down to who can be trusted more -- a handful of wealthy tech firms, the governments of the world, or the public.

The order draws some lines in the sand, but is still a thoughtful response to an emerging technology that has the potential to be simultaneously useful and dangerous.  And unlike many regulatory efforts, the Biden administration is not waiting for a disaster to happen before putting a framework into place.  One thing that we do know about the generative AI space is that size matters -- scaling up models can produce emergent capabilities.  Given this, focusing initial scrutiny on models above a certain size is not a bad idea.

[1] This is just a back of the envelope calculation based on currently available information.  The costs of the NVIDIA and other hardware is currently in flux due to market demand. 

[2] https://www.linkedin.com/pulse/mind-boggling-processing-power-cost-behind-chat-gpt-what-thakur/

[3] https://www.wired.com/story/openai-ceo-sam-altman-the-age-of-giant-ai-models-is-already-over/

[4] https://www.cnbc.com/2023/11/02/biden-ai-executive-order-industry-civil-rights-labor-groups-react.html

[5] https://twitter.com/ylecun/status/1720547401075675613

FDA Signals Willingness to Approve CRISPR-Based Treatment for Sickle Cell Anemia

By Kevin E. Noonan --

Ever since Watson and Crick ended their seminal Nature paper in 1953 by saying that:  "It has not escaped our notice that the specific pairing we have postulated immediately suggests a possible copying mechanism for the genetic material," James Watson & Francis Crick, "Molecular Structure of Nucleic Acids; A Structure for Deoxyribose Nucleic Acid," Nature 171: 737-38 (April 25, 1953), it has been envisioned that identification of genetic mutations causing disease would someday provide a way to repair them to achieve a cure.  Those hopes have only increased over the years since then, particularly in light of the ever-increasing genetic information that arose from cloning specific human genes in the late 1970's and the 1980's and even more so with the crowning achievement of the Human Genome Project at the turn of the century.  Unfortunately, these hopes have not been fulfilled, in large part due to failures in successful introduction of unmutated genes into appropriate cells in patients suffering from genetic diseases.  The causes for these failures are many, ranging from delivery of intact genetic constructs into these cells to stable insertion of the genes into the genome at a chromosomal location capable of efficient transcription under circumstances that did not result in inappropriate or disease-causing) expression of endogenous genes.

On October 31st, the U.S. Food and Drug Administration's Cellular, Tissue, and Gene Therapies Advisory Committee announced its recommendation that exagamglogene autotemcel (exa-cel) is safe for clinical use.  This drug is the subject of biologics license application (BLA) 125787 from Vertex Pharmaceuticals, Inc. in collaboration with CRISPR Therapeutics (Zug, Switzerland) and provides for the first time gene therapy for sickle cell anemia.  Sickle cell anemia was the first known human disease associated with a genetic polymorphism (an A→T single nucleotide polymorphism, or SNP), resulting in the substitution of a valine residue (encoded by a GTG codon) for a glutamic acid residue (encoded by a GAG codon) in the beta chain of human hemoglobin.*  The Advisory Committee's announcement followed the FDA's earlier announcement on October 27th that exagamglogene autotemcel is effective, having 29 of 30 patients stay pain-free for 18 months.

The CRISPR-mediated effect on sickle cell disease from exa-cel is mediated by targeting the binding site of an inhibitor (BCL11a) of the gamma-globulin gene, which results in fetal globin expression that alleviates the symptoms of the disease.  While there remains the possibility in some patients for "off-target" effects (some of which have been predicted by scanning human genomic DNA for such targets), the expected frequencies are low (1.6% according to David Altshuler MD, Vertex's Chief Science Officer).  Scot Wolfe, PhD (UMass Chan Medical School), a member of the FDA advisory committee, voiced the sentiment that:

We don't want to let perfect be the enemy of the good.  At some point we have to try things out in patients.  There is a huge unmet need for sickle cell disease.  It's important to advance therapies.  I certainly think that this is one of them.

The FDA is scheduled to decide whether to approve the drug at a meeting scheduled for December 8th.

* The work, by Linus Pauling and colleagues at Cal Tech was published in the November 25, 1949 issue of the journal Science, about three and a half years before Watson and Crick deciphered the structure of DNA.