Patent eligibility and generative AI
Allen E. Hoover
Intellectual Property Lawyer and Partner at Fitch, Even, Tabin & Flannery LLP (since 2009)
The Federal Circuit court of appeals today issued an interesting decision under the law of patent eligibility. The case, AI Visualize, Inc. v. Nuance Communications, Inc., has potential implications for generative artificial intelligence technologies.
AI Visualize holds several patents relating to visualization of medical scans at a remote computer, such as over the Internet. Remote viewing of complex medical scans can be challenging because of bandwidth limitations and the large amount of data generated in a medical scan. The patented solution involved transmission of a portion of the data and generation of a “virtual view” on the remote end.
Nuance Communications brought an Alice motion to dismiss, which the district court granted. On appeal, the Federal Circuit affirmed. The court applied the familiar two-step Alice framework, concluding in step 1 that the claims were simply directed towards the manipulation and display of data, which was an abstract idea, and that the patentee had not shown anything further under Alice step two.
The patentee argued that the requirement for creation of “virtual views” meant that the invention was not an abstract idea. But the Federal Circuit disagreed, because the claim specified that the creation of the virtual views was achieved by manipulation of the underlying pre-existing data. In the court’s words:
领英推荐
AI Visualize argues that the claims are not directed to an abstract idea because the claims require the creation of “on the fly” virtual views at a client computer. . . . But the claim language makes clear that virtual view “creation” is achieved by the manipulation of a portion of the existing VVD [the data set]. For example, Claim 1 of the ’609 patent requires “accepting at a remote location at least one user request for a series of virtual views of the volume visualization dataset” and “creating the requested frames of the requested views from the volume visualization dataset.” . . . [T]his “creation” of a virtual view from the existing VVD, recited in general terms, is abstract data manipulation. (Cleaned up.)
So, another set of patents falls to Alice.
Could one describe the AI Visualize patents as being in the field of “generative” artificial intelligence? The patented technology apparently did indeed generate virtual views, albeit perhaps by a simple algorithmic manipulation of preexisting data.
But how does this technology differ in kind from the latest generative artificial intelligence technology (ChatGPT, Microsoft Copilot. etc.)? These latest systems also begin with an underlying, and very large, data set. Using some algorithm and a user query, these systems can generate realistic text responses, photographs, document summaries, and so on. ?At their core, though, these systems are likewise using algorithms to manipulate preexisting data to generate a response. ?The algorithms used by ChatGPT and the underlying data sources are undoubtedly vastly more complex than those of the AI Visualize patents, but to be sure, these technologies still involve algorithms manipulating data.
Put another way, generative artificial intelligence can produce some profoundly amazing results, but ultimately, this technology is at its core the manipulation of data to generate an interesting or useful output. The AI Visualize patents fell because “the claim language makes clear that virtual view ‘creation’ is achieved by the manipulation of a portion of the existing [data set].” What does this imply for patents directed to generative artificial intelligence?