Everyone is thrilled about the cost savings of AI-drafted patents. Litigators who will tear them apart in five years are thrilled as well.
LLM’s are good at mimicking the syntax of a patent. They have been trained on the shape of a ‘large’ number of them. It produces documents that pass the sniff test, read beautifully, and cost a fraction of the price. But AI lacks the single most important mechanism that an attorney has when they are drafting: fear.
A human patent attorney drafts defensively. Every word is chosen with the paranoid anticipation of a hostile opponent trying to invalidate it at an EPO Opposition or in litigation. The human attorney creates friction. They interrogate the inventor to extract the unwritten, tacit constraints of the technology, the messy reality of the lab bench that never makes it into the invention disclosure form.
Drafting a patent is an adversarial process. You write down the invention and then rip it apart yourself to see where the gaps are. Better to rip it apart yourself than have your pants pulled down in front of your clients at the EPO.
AI doesn’t feel fear. It optimizes for plausible text, not adversarial survival. An attorney is not optimizing for the next plausible word. They’re optimizing for someone like them tearing it apart line by line.
It will be interesting when this new generation of LLM drafted patents are attacked by motivated human attorneys. That’s when the rubber will meet the road and everyone will find out if it was an actual cost saving or if they have a bag of lemons.
