“To conclude that AI serving as an aid to human thinking is necessarily better than possibly replacing some aspects of human decision making, when we simply don’t yet have the technological capability to test one over the other, would fall into the logical fallacy of equating a presumption with a conclusion.”
In response to our recent article on artificial intelligence (AI) reducing transactional costs to help determine infringement and invalidity determinations, a commenter made an interesting counterpoint, paraphrased as the following: AI provides useful tools that should be used as an aid to human thinkers, not as a replacement to human thinking. Moreover, when it comes to AI making subjective determinations, such as obviousness or novelty, we should be skeptical of relying on AI, either legally or practically.
We appreciate the counterpoint and we wanted to address it in this follow-up article.
How do we determine the best role for AI in our patent system? If we have a choice between AI serving as an aid to human thinking, or possibly replacing some aspects of human decision making, what is the correct choice? What would better serve to improve our patent system?
What Constitutes ‘Improvement’?
To have a productive discussion re AI’s proper place in our patent system, we need to first understand what “improvement” to our patent system means. When we have common ground as to how to assess improvement, only then we can discuss which role of AI would better implement that improvement.
To frame our understanding of such improvement, let us look at the patent system as it stands today.
The judiciary sits in the middle of the patent-transaction ecosystem. When presented with a case, the judiciary performs a two-step process. It takes on the role as (1) the arbiter of factual and legal contentions between parties, and (2) the enforcer of its ultimate decision. In its role as arbiter, the court determines informational attributes relating to patent validity, scope, and infringement. After this determination is made, it then enforces that decision.
In this system, the presumption is that the basic informational attributes of a patent are either unknown or at best contested, and we need the court system to make this determination.
Herein lies the problem with our patent system. Relying on the court to determine basic informational attributes of a patent is both costly and inefficient. It costs millions of dollars and takes several years to determine whether the patent is valid, infringed, and the damages.
Because the court is so inefficient at making these informational determinations regarding a patent, enforcement costs in turn are extremely high.
Further, these information and enforcement costs are intermingled; meaning, you cannot enforce a patent unless the same court system first determines the basic informational attributes of a patent, resulting in a costly self-perpetuating cycle of inefficiency.
Why is this important to understand when framing a discussion about the patent system and patent transactions within that system?
As Douglass C. North pointed out in his 1992 paper “Transaction Costs, Institutions, and Economic Performance,” the framework of our patent system creates the actors that operate within it.
The constraints imposed by the institutional framework (together with the other standard constraints of economics) define the opportunity set and therefore the kind of organizations that will come into existence.
North gave a very powerful example:
If the highest rates of return in a society are from piracy, then organizations will invest in knowledge and skills that will make them better pirates; if the payoffs are highest from increasing productivity, then firms and other organizations will invest in skills and knowledge that achieve that objective.
In our present-day patent system, extremely high informational costs create the economic driver to reduce enforcement and bargaining costs between parties in a patent transaction.
Put another way, the court system’s high informational cost structure creates a driver to minimize enforcement costs, which manifests in today’s patent litigation as early settlements that are below the cost of determining the informational attributes of a patent.
Put yet another way, the court system’s high informational cost structure creates the economic driver for low-value and “nuisance” patent litigation (see part I and part II of an analysis relating to how we have historically misdirected patent policy to deter such “nuisance” patent litigation).
North recognized the central role of informational costs: “[t]he cost of transacting arises because information is costly and held asymmetrically by the parties to exchange.”
What is the Ideal System?
In a perfect patent system, the informational attributes of a patent are efficient to determine and known to both parties. When informational costs are low and informational attributes are known to both parties, the following occurs:
North describes this as the “zero-cost transaction.” This is a perfect system in which there are no transactional costs between a patent holder and an alleged infringer reaching an agreement on a patent transaction. The only money spent, if any, is for the value of a patent license.
So, when we are thinking about patent reform, the discussion should be centered on “how do we approach a zero-cost transaction for patent transactions?” This sets the standard for “improvement.”
Bringing AI into the Equation
Assuming we are on the same page regarding what it means to improve our patent system, this frames the next question: between (1) AI serving as an aid to human thinking, or (2) possibly replacing some aspects of human decision making, which of the two better serves to improve our patent system?
At this point, I don’t believe we can actually answer that question, because we don’t live in a world where AI can reliably replace aspects of human thinking with respect to our patent system.
But to conclude one is necessarily better than the other, when we simply don’t have the technological capability to test one over the other, would fall into the logical fallacy of equating a presumption with a conclusion.
Instead, North would offer a different approach. He described characteristics of successful institutions. Namely, institutions that allow for decentralized decision-making and trial and error see greater success over time.
Therefore, institutions should encourage trials and eliminate errors. A logical corollary is decentralized decision making that will a society to explore many alternative ways to solve problems.
Applying North’s teachings to our patent system, he would recommend we test different methodologies to determine informational attributes of a patent and learn through trial and error which methodology best reduces informational costs. Only when we have the opportunity to apply and test different methodologies to determine informational attributes of a patent will we truly learn which method is best.
Don’t Put the Cart Before the Horse
North certainly factored in the use of technology and technology’s role in an institution:
Institutions, together with the technology employed, affect economic performance by determination transaction and transformation (production) costs.
Relying on the teachings of North, we should actively test AI in different applications and scenarios and determine which would allow us to approach a zero-cost transaction, particularly zero costs to determine the informational attributes of a patent.
But to enable us to test AI effectively, we cannot foreclose ourselves to the possibility that AI’s proper place could be to actually replace some aspects of human thinking.
If AI replacing human decision making in certain circumstances would enable a zero-cost patent transaction, then this may be the proper place for AI in the patent system. But if using AI as a mere tool to aid human thinking enables us to approach this zero-cost transaction, then this may instead be the best role for AI.
In essence, let’s not put the cart before the horse when making determinations regarding AI’s proper role in our patent system. To improve our patent system, we need to come to common understanding on the key problem it faces, namely, its unsound economic underpinnings. And we need to allow ourselves greater flexibility to test different methods and technology to improve the patent system by helping us to eliminate, or at least significantly reduce, the high costs and inefficiencies of determining the informational attributes of a patent.