Earlier this week, the U.S. Patent and Trademark Office (USPTO) issued a request for public comment regarding the impact of artificial intelligence (AI) on prior art and the knowledge of a person having ordinary skill in the art (PHOSITA).

Of the two issues, I believe the prior art question is the more difficult one to address. As the USPTO mentions, it is conceivable that AI tools may be used to create vast databases of disclosures on virtually any topic, even if no human directly contributes to or reviews those disclosures. In recognizing that a large majority of such publications may be nonsense or, in the very least, inoperable, the USPTO appears to be wondering whether these AI-generated references should still be considered prior-art “printed publications” for the purposes of assessing patentability for later-filed applications.   

Even if nonsense, a particular reference “is prior art for all that it teaches” and may still be cited for anticipation or obviousness. See MPEP 2121. Because, while prior art must be operable, that operability is presumed when the reference relied on “expressly anticipates or makes obvious all of the elements of the claimed invention.” Id. One potential avenue suggested by the USPTO is that AI-generated references, particularly those created without any human input or review, should not enjoy the same presumption of operability that human-generated references are currently given. Such an approach could potentially limit the number of AI-generated prior art references available to the patent office (and other challengers). But it would also introduce a distinction between the treatment of AI- and human-generated prior art that is not currently recognized by courts or the USPTO. Further, it would inevitably bring along a new threshold inquiry as to the provenance of prior art, and who should bear the burden of proving whether a reference was indeed AI-generated.