US patent office wants an AI to scan for prior art, but doesn't want to pay for it
'The selected vendor must be willing to receive consideration that is primarily non-monetary,' says the USPTO
The US Patent and Trademark Office is exploring a plan for using AI at the agency to speed up the process of granting patents - but its initial request for information says that vendors should expect to be paid in exposure rather than cold, hard cash.
A request for information [PDF] sees the patent agency turning to commercial AI providers to help it "reduce patent and trademark pendency." The primary objective [PDF] of the Office's plan centers on using AI tools to plumb the depths of the ever-growing catalog of domestic US patents, foreign patents, and non-patent literature to figure out if an application conflicts with something that's already been registered.
According to the USPTO, AI bots scanning mountains of prior art should be able to return information such as the relevance of prior art to a claimed invention and provide notations indicating where relevant portions can be found in a prior patent filing. They should even be capable of reasoning how prior patents may relate to new claims in novel, non-obvious ways.
This, the USPTO believes, will "increase the overall efficiency of the patent examination process," and ensure "consistency of the patent work product."
There is some irony in using AI bots, which are often trained on copyrighted material for which AI firms have shown little regard, to assess the validity of new patents.
It may not be the panacea the USPTO is hoping for. Lawyers have been embracing AI for something very similar - scanning particular, formal documentation for specific details related to a new analysis - and it's sometimes backfired as the AI has gotten certain details wrong. The Register has reported on numerous instances of legal professionals practically begging to be sanctioned for not bothering to do their legwork, as judges caught them using AI, which borked citations to other legal cases.
The risk of hallucinating patents that don't exist, or getting patent numbers or other details wrong, means that there'll have to be at least some human oversight. The USPTO had no comment on how this might be accomplished.
Interested AI firms have until June 24 to submit their proposals to the USPTO, which made clear in the announcement that it's just gathering some ideas right now, not looking for actual bids; thank you very much.
- Google's AI vision clouded by business model hallucinations
- AI hallucinates software packages and devs download them – even if potentially poisoned with malware
- AI models routinely lie when honesty conflicts with their goals
- Cursor AI's own support bot hallucinated its usage policy
That said, once the Office does decide to approach select firms to get things rolling, they ought not to expect much in the way of compensation for their contributions.
"The selected vendor must be willing to receive consideration that is primarily non-monetary," the USPTO said in its RFI. "Benefits to the vendor include its ability to display and market its capabilities, and the ability to fulfill a critical US Government technology gap on the world stage."
In short, they're really hoping AI companies will be willing to work for exposure, not monetary compensation for their contributions. Speaking of what selected AI firms will be responsible for, the RFI includes the expectation that vendors "provide the infrastructure for the models, to include the compute, storage, and networking assets/capabilities." All of that hardware, the RFI noted, will have to exist within the USPTO's secure cloud boundary, too.
The timeline from requesting information to the deployment of the USPTO's desired AI system wasn't specified, but it appears to be something the patent boffins want as soon as possible, with the RFI mentioning plans to acquire "significant new AI capabilities in the coming months." ®