BY Matthew T. Kitces

Artificial Intelligence (AI) is quickly taking the forefront in many of today’s innovations and new works. However, understanding how the idiosyncrasies of AI do—or should—fit into the legal systems for protecting IP can be daunting.

To better understand how AI should be treated in the U.S. legal system, the United States Patent and Trademark Office (USPTO) has been taking input from the public, including individuals, practitioners, companies, bar associations, trade associations, law firms, academics, and foreign patent offices. The USPTO recently released a report on its findings, titled “Public Views on Artificial Intelligence and Intellectual Property Policy” (the Report). The full report is available online.

The Report covers IP generally, as well as specifics related to patents, copyrights, trademarks, trade secrets, and other issues. Some of the Report’s major findings include:

• No AI definition—Most believe there is no universally recognized definition for AI. Because AI is dynamic, most believe it would be prudent to avoid specifically defining AI in IP policymaking.
• No inventorship—Most believe AI itself cannot and should not qualify as an inventor or author, leaving inventorship to humans. However, many suggested that if artificial general intelligence arises (i.e., an intelligence akin to that possessed by humankind), such an AI can be the inventor or author.
• Patentability—Most believe that AI is a subset of computer-implemented inventions, patentable under the same standards as other such inventions. However, there is some concern that certain AI innovations may be difficult to describe with sufficient detail and understanding to meet the enablement requirement of 35 U.S.C. § 112(a).
• IP examination—Most believe AI would be useful to the USPTO in examining patent and trademark applications, but that AI should be used only as tools for an examiner, and should not entirely supplant human examination.
• Copyrighted training data—Many believe that the use of copyrighted materials to train AI may be a violation of the reproduction right of a copyright owner. Nevertheless, such use may be defensible as a non-infringing fair use.
• Trade secrets—The general consensus was that current trade secret laws, policies, and practices are sufficient to protect trade secret aspects of AI. Some noted that AI might be used as a tool to discover trade secrets without breaking security measures, but there was no consensus about whether this issue should be addressed with a change in the law.
• Current laws and new classes of IP—Most believe the U.S.’s current IP laws are sufficient to address AI. Nevertheless, some suggest it may be prudent to look into new potential classes of IP specific to AI, such as classes of IP protection specific to data used to train AI and the trained models themselves. Otherwise, many felt that any gaps left by IP law can be filled by existing commercial law, such as contract law.

While not official policy, this Report may be a good barometer as to how AI policy will be shaped in the future, which can be useful to any entity making use of AI.