[Guest post] USPTO’s report “Public Views on AI and IP Policy”. What does the industry think about AI inventions?

The IPKat is very happy to host the following guest post by former GuestKat Bertrand Sautier (Clarivate Analytics), who tackles a recently released USPTO report on AI and IP policy.

Here's what Bertrand writes:

USPTO’s report “Public Views on AI and IP Policy”. What does the industry think about AI inventions? 

AI Kat

Recently published by the USPTO, the report entitled “Public Views on Artificial Intelligence and
Intellectual Property Policy” discusses various aspects of the AI/IP policy relationship, following the initial conference held in January 2019 and the following requests for comments (RFC) made in August and November 2019 by the American agency.

Primarily patent driven, the report covers many aspects related to the AI and IP discussion, including AI’s impact on patenting as well as inventorship, ownership, eligibility, disclosure and the level of the ordinary skill in the art and others.

Close to 200 comments were received between the two RFC. The USPTO has tried to provide us with a balanced summary of the views expressed throughout the year, whilst reaffirming the current state of US law.

Similar efforts have been conducted by other major IP offices, including WIPO, EPO, JPO and, more recently, the UKIPO.

Notable points are selected here based on the most contentious:

On what constitutes an AI invention: There is a confusion that makes the report difficult to read on the definition of an AI invention. While the question initially asked by the USPTO aims at defining this notion, it unfortunately groups together two distinct ideas. On one hand, inventions that embody an advance in the field of AI or that apply AI to a field other than ITC, and on the other hand, inventions that are produced directly by the use of AI, regardless of the field and invention itself. In other word it’s the difference between AI being the output captured in the invention or AI being the input to create the invention.

This is more than a taxonomy issue and there should be a clear distinction between protecting AI itself via patent, and protecting an invention created by or with the use of an AI, to avoid confusing the challenges at hand.

On the inventor determination when an invention is embedding AI elements: The majority of commenters concluded that the current US law is equipped to deal with any type of contributors to the conception of an invention, the driving criteria to determine who is deemed an inventor. Therefore, for a data scientist performing data training or model building to be considered an inventor, proof should be made that she or he contributes to the conception and is not simply reducing the invention to practice.

On the question of inventorship generated by: The USPTO states that US law is replete with language indicating that the inventor of a patent application must be a natural person. This is consistent with other IP offices' views that have expressed their opinion about the DABUS application (see IPKAT posts here and here).

Most commenters consider that the current and near-future state of AI technology does not indicate that AI is capable of inventing autonomously. However, as noted by the USPTO, this opinion is not shared unanimously, as several large companies (amongst others) have already voiced that inventions can be automatically generated by AI.

On ownership of patents generated by AI: No change of law is required. No surprise here: only a natural person or company should remain entitled to the property. However, arguments are made in favour of the data trainer, or the person that control/owns the AI system to be the owner of the AI-generated invention by default.

On AI technology inventions disclosure and enablement: Current US law contains three separate disclosure requirements: written description, enablement, and best mode. For most commenters, disclosure requirements for inventions in the AI field should not be considered differently than computer implemented inventions, for which the USPTO has issued recent examination guidelines. The disclosure should be detailed enough that one of ordinary skill can reasonably conclude that the inventor possessed the claimed subject matter.

However, the enablement requirement might be harder to fulfil for AI technology inventions. Based on the Wands Factors, it represents the guidance given by the patent specification to the man skilled in the art to make and use the full scope of the claimed invention without undue experimentation. The variability in predictability of AI systems has been a topic of debate and some suggest that the enablement factors and written description requirements might be difficult to fulfil.

On AI technology inventions and the level of the person of ordinary skill in the art: When it comes to non-obviousness requirement and the skill level of the person of ordinary skill in the art, commenters argue that AI should logically affects that fictional skill level, in that it is necessarily impacted by the evolution of various technologies becoming widely available, including AI systems.

Also, most commenters point out that there is no specificity to prior art when it comes to AI tech inventions. Some pointed that prior art generation by AI could become an issue, for all patenting fields (see here for an example of automatic prior art generation system started in 2016).

Copyright protection given to a work produced by AI: In a similar debate as for inventorship, but surely more of an agreed reality than inventions generated by AI, the office reaffirms the obligation for an author to be a natural person in order to be able to obtain copyright protection. A non-human cannot be an author (remember the Naruto selfie case?) This view is shared by most commenters. Here again, those arguing in favour of copyright protection for a work produced by AI usually insist that the controller/owner of the AI should obtain ownership of the work created.

It will be interesting to see if policy recommendations emerge from all these collaborative efforts around the world and if it follows AI driven changes in other fields like tax law, social law or tort law. There will surely be opportunities to read many more reports and analyse real life cases on this complex relationship.