AI patents depress innovation---laundry list.
Raid ’18 [Shlomit and Xiaoqiong Liu; August 2018; Professor of Law at Yale University, Visiting Professor at the Fordham University School of Law, Fellow at the Information Society Project; Engineer and Fellow at the Fordham Intellectual Property Institute; Cardozo Law Review, “When Artificial Intelligence Systems Produce Inventions: An Alternative Model for Patent Law at the 3A Era,” vol. 39]
VI. An Alternative Model for Patent Law at the 3A Era - Incentivizing Stakeholders Within the AI Multiplayer Model
A. Rethinking the Incentive Effect of the Current Patent Regime
The patent reward in general is questionable. Scholars argue that patent laws fail to reach their goal of maximizing benefit to society. Granting twenty years of exclusive rights to the inventor or the inventor's transferee may not significantly incentivize inventors. Surveys show that even CEOs in most industries see patent incentives as relatively unimportant. 192 If corporate leaders who are driven by a profit motive do not see the value of patent incentives, then certainly such an incentive is meaningless to the multiple players and cumulative patent models. When determining whether to grant patent rights in uncertain situations, therefore, such as in regard to inventions by AI systems and the multiplayer and cumulative patent models, we should not overemphasize the importance of the patent system compared to other alternatives.
What are the flaws in patent incentives that make them irrelevant to AI? First, patent law only rewards the first inventor, while second comers get nothing. The drive to be first - even by an hour or two 193 - can force competing owners of AI systems deep into diminishing returns. This is wasteful.
Second, patent law grants exactly the same length of protection for all inventions, 194 "regardless of [their] R&D costs and … other economically relevant factors." 195 Although this approach is simple, it "grossly overrewards some inventions and underrewards others." 196
Third, patents are superfluous for products that would be invented anyway. 197 For instance, "conventional software developed rapidly even before courts were willing to grant it patent protection." 198 Patent protection serves a limited purpose.
As a fourth point, patent law fails in the multiplayer and cumulative patent environment characteristic of AI systems. It is not flexible in allocating rewards and is thus economically inefficient. Indeed, patent law only offers four ways to allocate rewards among two parties: "0:0 (both patents invalid), 50:50 (both patents upheld), 100:0 (first patent valid, second invalid), [and] 0:100 (first patent invalid, second valid)." 199 Patent law may also impede future technological progress by making it harder for other AI systems to build on earlier inventions. 200
Fifth, in practice, patent law would pose difficulties in bringing patent infringement actions against or on behalf of inventions by an AI or against the copying of an AI's invention. Because of the unpredictable nature of AI, it is very difficult to identify the human that is responsible for the "actions" of an AI system. Sometimes the human involved in the process does not know how an AI system arrives at an invention. In other cases, the AI system can "break" data into electronic nanocomponents and rebuild it in different ways, thus rendering it impossible to establish proof of infringement. 201