Sunday, November 12, 2017

Do Machines, and Women, Need a Different Obviousness Standard?

This blog post addresses two different articles that might at first blush seem to be very different. The first is Ryan Abbott's new article Everything Is Obvious, which explores the implications of machine-generated IP for the nonobviousness standard of patentability. Abbott argues the inventiveness standard should be adjusted to take into account the new reality that inventors are frequently assisted by machines or, in some cases, are machines. The second article is Dan Burk's Diversity Levers, published in 2015 in the Duke Journal of Gender Law & Policy. In the article, Burk argues the standard for nonobviousness should be adjusted to take into account the unique mindset and institutional situation of female inventors. (To be clear, Burk is not coming at this issue out of the blue. He has previously written about feminism in collision with copyright, arguing that copyright can be used to suppress feminist discourse).

Abbott's thesis is that, in comparison to machines, humans are all a little less skilled, so a human-based obviousness standard will necessarily lead to too many patents if machines are commonly employed. Burk's point is that, in comparison to men, women are typically more risk-adverse, so a male-based obviousness standard will necessarily lead to too few female-invented patents.

Ryan Abbott – Everything Is Obvious

As Abbott observed in an earlier article, I Think, Therefore I Invent: Creative Computers and the Future of Patent Law, machines are now generating patentable inventions. Abbott thinks this is a potential problem for the law of nonobviousness. If the machine inventor is held to the human obviousness standard, then too many machine-invented patents will be valid. For example, if IBM's Watson discovers a new drug by crunching numbers and combinations at an inhumanly fast rate (see p. 20), should this new drug be judged as "obvious" or "nonobvious" based on what humans alone can do? Presumably a human-based standard would require finding the invention is nonobvious, since no human could have invented the drug without a machine or at least not within a reasonable timeframe. On aggregate, using a human-based PHOSITA standard in this type of situation would lead to far too many patents that are in fact obvious to machines (p. 26).

The implications of advanced artificial intelligence for intellectual property law is an emerging area that several scholars have begun to tackle, including Liza VertinskyShlomit Yanisky-RavidXiaoqiong (Jackie) LiuSamuel MoorheadAmanda Levandowski, and Mark Lemley, among others. Clearly, the question of whether machines can or should be able obtain patents or other IP, and how machine invention challenges patentability rules, is a very importance threshold issue. Vertinsky's forthcoming chapter, in particular, identifies precisely the same problem explored by Abbott:
[To achieve a patent,] the invention must be non-obvious in the sense that "the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious ...to a person having ordinary skill in the art [the PHOSITA] to which the claimed invention pertains." ... In its current form, the PHOSITA is clearly a human of ordinary skill in the art, not a machine. With thinking machines in the equation, however, policymakers might have to consider whether the PHOSITA should be modified to include thinking machines – perhaps some kind of machine/person combination or "M/PHOSITA." But if thinking machines were to be included in the analysis as part of the PHOSITA category, it would be difficult if not almost impossible to limit the expansiveness of non-obviousness determinations. ... Little if anything might be considered patent eligible with this expansive approach to obviousness.
Abbott's solution to the rise of machine inventors is to hold machine inventors, and machine-assisted human inventors, to a different - in some cases much higher - standard. Abbott provocatively concludes that in such cases the "person having ordinary skill in the art" (PHOSITA) should instead be seen an "inventive machine," at least when the typical approach in the field, or to this type of problem, is to use a machine (p. 33). (The second-best option, he says, is to modify the current PHOSITA standard to take into account the "technologies used by active workers" in the relevant field. But Abbott thinks an "inventive machine" standard is better because it "emphasizes that the machine is engaging in inventive activity" (p. 5), which, in turn, would allow courts to use a more "objective" standard for judging obviousness than they currently do. More on this point below.)

So how exactly would this inventive machine standard work? Conducing the typical Graham inquiry would be near-impossible because it would be so difficult to imagine in the abstract what a machine would find obvious (p. 37). Abbott recognizes this difficulty and concludes that, rather than asking whether a (human) PHOSITA would find the invention obvious in light of all available prior art based on their level of skill in the art, see 35 U.S.C.§ 103, the "inventive machine" standard instead asks: would the average machine used in the field or with respect to respect to this problem have been able to reproduce the invention that the machine/inventor came up with? (pp. 37-39).

For example, if the question is whether IBM's Watson's invention of a particular drug was obvious at the time the patent was filed, the test would be whether the drug would have been "obvious" to typical machines used in drug discovery and invention. The difficult part of this inquiry is to select the machines for setting the standard (pp. 36-37). According to Abbott, the courts would have to either choose a "most commonly used computer in the field" (in this case, Google's DeepMind), or alternatively base the standard on several computers used in the field (for instance, IBM's Watson, Google's DeepMind, and any others of comparable quality).

Implications of an "M/PHOSITA" Standard 

This a highly thought-provoking paper that deepens our still-nascent understanding of the implications of advanced artificial intelligence for intellectual property law. I am still left with several questions.

1. How Would an Inventive Machine Standard for Obviousness Affect Disclosure?

My first question is how an inventive machine standard would affect other patent law doctrines, such as the level of disclosure required to satisfy Section 112? In a book and an article (both co-authored with Lemley), Dan Burk has explored the issue in depth, observing that the PHOSITA working in areas such as software development is quite different from the PHOSITA working in an area such as biotechnology. In Policy Levers in Patent LawBurk and Lemley argued that obviousness law, which requires assessing obviousness from the perspective of the PHOSITA, constitutes one of several doctrinal "policy levers" that courts use to affect innovation incentives (p. 1649-1651). One of their most intriguing insights was that how the PHOSITA is calibrated for an issue like obviousness affects how the court holds on other issues, such as enablement. A dimmer PHOSITA may make an invention more likely to be found nonobvious, but also more likely to fail the disclosure requirements of Section 112. For example, "if the court concludes that an art is uncertain and its practitioners not particularly skilled, it will be inclined to find even relatively modest improvements nonobvious to the PHOSITA." At the same time, however, the court will also "be inclined to require greater disclosure to satisfy the requirements of Section 112, and correspondingly to narrow the scope of claims permissible from any given disclosure. If the art is predictable and the PHOSITA quite skilled, the reverse is true. ..." (p. 1650).

Applying these insights here, an "inventive machine" standard might mean patents will be more likely to be found obvious, but also that less disclosure would be required to implement the invention. Thus, we should predict fewer patents, but less disclosure required to achieve broad claim scope, given how easy it is for the smart machine to understand how to reproduce the invention. In other words, we might see machine-produced inventions follow the same pattern as software inventions: few patents with very broad claims and limited disclosure of specific information about the invention. This makes a funny sort of sense. After all, software patents typically cover software programs to be implemented by or with the use of computers. Machine-invented patents can cover any kind of invention, but are presumably directed at machines. So in both cases, the PHOSITA for purposes of disclosure is conceptualized with non-human machines in mind. If Abbott's inventive machine standard is adopted more broadly, will patent disclosures begin to look different? What does a machine need to implement an invention, anyway? It will presumably be distinct from what a human requires. Will patents contain fewer words and pictures, and more zeros and ones?

The point here is that changing the perspective of PHOSITA for purposes of judging obviousness will likely produce ripple effects across patent law that may alter what patents disclose, what scope they cover, and how they function in the economy.

2. What Will Happen to Incentives to Innovate If No Inventions Are "Nonobvious" Anymore

A closely related question relates to the problem stated above: if we accept Abbott's "inventive machine" standard for fields in which machine inventors or machine-assisted inventors predominate, how would this effect innovation incentives? If all patents, or at least all patents in fields where machines are used, are held to an inventive machine standard, then won't most merely human-invented patents be found obvious?

Abbott addresses this possibility briefly, and indeed his title ("Everything Is Obvious") accepts its sweeping potential (pp. 40-41). I am not sure I agree that a machine-inventor standard will render everything obvious. Machines may be better at some things, but they aren't better at everything, and may lack the skills necessary for innovation. On the other hand, if Abbott is right that incorporating machines into the PHOSITA standard will make everything obvious and nothing patentable, I find his response somewhat unsatisfying. His main response is simply that patents are not the only means of promoting innovation, and that "business ventures may be successful without patents" (p. 40). This is a typical response when a legal change seems to reduce inventors' likelihood of filing for patents. For example, the Supreme Court in Kewanee essentially stated that it doesn't care that much if fewer people file for patents, and decide instead to keep their inventions secret, since other people will just invent and disclose things anyway. This just can't be the solution we break out any time a change makes filing patents less likely. Otherwise, why are we bothering to support a costly patent system in the first place?

3. Is a Sui Generis Standard Necessary?

Abbott's general goal is to address the problem of accurately calibrating the PHOSITA standard in a world where machines invent or where human inventors often use machines to help them invent. But I am not certain a sui generis "inventive machine" standard is strictly necessary to reach that outcome. If courts more consistently incorporate machines into calculating the PHOSITA standard or into the secondary considerations for obviousness (p. 4, 16-17), I think we would get to the same result. As Abbott himself notes, KSR held that, "[t]he diversity of inventive pursuits and of modern technology counsels against confining the obviousness analysis ... by overemphasizing the importance of published articles and the explicit content of issued patents." The Court expressly stated that, in conducting the obviousnesss inquiry, "a court can take account of the inferences and creative steps that a person of ordinary skill in the art would employ."

So I think under a proper interpretation of modern obviousness law, the focus would (or should) still be on whether the invention would have been obvious to the ordinary person in the art, who may well use a certain machine. Abbott's response to this point is that an inventive machine standard would provide a more "objective" standard for judging obviousness. Rather than asking in the abstract "would an inventor in this field, using a typical machine, have found this obvious based on the scope and content of the prior art?", courts could simply cut to the chase and ask, "would an average machine in the art have been able to reproduce this invention at the time?" (p. 5). We could literally just insert the prior art into the average machine and see what happens.

I like this point and agree with Abbott that the focus on reproducibility-by-the-standard-machine is a fascinating benefit of conceiving of the machine itself as the PHOSITA. But I think we would still end up doing a similar analysis in practice, without actually raising the PHOSITA standard to the level of the machine. For instance, in the Watson example above, I think we'd still be asking whether the PHOSITA, using ordinary machines in the art such as Watson or DeepMind, would have found the invention obvious. Experts could be introduced by one or both parties to do the analysis.

4. Obviousness - It's All Relative

My final point is that I am not certain that we should, as a matter of policy, start tailoring obviousness to what kind of entity - or person - is doing the inventing.

This brings me to Dan Burk's Diversity Levers. 

Dan Burk – Diversity Levers 

Abbott and others like Vertinsky and Yanisky-Ravid are tackling the looming problem that the average inventor may no longer be a human being. In his 2015 article, Burk tackled a different problem in the same bucket: that the average inventor is not necessarily always a male. Both issues arise from the fact that as society becomes more diverse, so does the pool of potential inventors.

A few selectively chosen quotes will help to get across Burk's thesis, and where he is coming from.

"A substantial and growing body of empirical literature demonstrates that women patent and commercialize new inventions at only a fraction of the rate of similarly situated men..." (p. 26)

"[While] [t]he failure to attract women into science and engineering fields undoubtedly contributes to the smaller total number of patents involving women[,] ... the gender gap in patenting can only partially be explained by the lower number of women entering patent-intensive sectors involving engineering or the physical sciences." (p. 32).

"The causes of the patent gender gap are likely complex, arising from an intricate milieu of deeply-seated social factors. Women may have been socialized to take fewer risks, to push their projects less aggressively, and to think about commercialization of their work less often than their male counterparts." (p. 33)

"[T]he woman of ordinary skill in biotechnology experiences innovation and patenting very differently from her male counterpart. Statistically, she is significantly less likely to develop discoveries into innovations, and is far less likely to seek a patent for patentable innovations. The result is an effective shift in the contextual standard for non-obviousness; where the male biotechnology innovator will recognize and develop an innovation, the female biotechnology innovator, hampered by social and cultural impediments, may not." (p. 37).

Burk's solution to the unstated problem - that women patent less because of their natures - is to use the obviousness standard as a policy lever in order to encourage more patenting by women.
The uncertainty of success is assessed from the point of view of the person having ordinary skill in the art, but that uncertainty is higher for a woman of ordinary skill. We know that female innovators are far less likely to develop a promising technology, meaning that the ostensibly neutral obviousness standard incorporates a de facto gendered assumption about risk-taking. Recognizing that the woman of ordinary skill faces a gender gap avoids the assumption that the PHOSITA makes a masculine uncertainty assessment and assists in overcoming the impediments faced by female innovators. (p. 37-38).
In other words, Burk proposes the opposite of what Abbott proposes. Abbott explores whether it might be sound policy to shift the non-obviousness standard (generally) upwards in order to ensure that not too many patents are granted to smart machines or people assisted by smart machines. Burk explores whether it might be sound policy to shift the non-obviousness standard (generally) downwards in order to assist women, who are more "risk adverse" than men, in obtaining more patents than they otherwise could.

Is an "Inventive Woman" Standard a Good Idea?

There is not necessarily anything wrong with adjusting inventiveness standards to better comport with reality, especially if the usual standard would end up discounting the intellectual creations of particular minority group. For example, in the copyright sphere, Rebecca Tushnet has discussed that issue.

But to me the idea of an inventive woman standard seems totally contrary to patent policy. From the economic perspective, the point of the obviousness standard is to make sure that innovations get out into the public that would not otherwise be produced, and avoid needless patents. If we use the inventive woman standard whenever the inventor is a woman, it seems there would then be a lot of patents that are not "obvious" to women, but nonetheless are "obvious" to men. Imagine a terrifying new world in which female patent-holders set out to sue businesses run by men to whom the female-patented inventions were obvious. I can see The Wall Street Journal headline.

There are also various other problems to point out. If Burk's goal is to get more women patenting, then altering the obviousness standard does not seem an ideal fix. How would women in the field feel about being evaluated under a standard that counts them as "more risk adverse" or "less likely to invent" than a man? It seems that evaluating women under an easier standard would in fact devalue their work, and consequently decrease their likelihood of inventing, or at least of seeking patents for their inventions. It would also likely increase negative stereotypes about women, not to mention trigger concerns about institutional-level gender bias by the United States government.

Burk is right to call out the gender disparities in patenting and in many areas of science. But solutions are more likely to include measures like unconscious bias training for people who are in a position to mentor young women in science.

No comments:

Post a Comment