AI Disclosures
Written by Ben Esplin
What Inventors Give Me Now
The standard story about AI and patent practice is a story about speed. Drafting tools promise faster application drafting, faster response drafting, faster deep analysis of portfolio parameters. That story is true as far as it goes, but it overlooks a large part of the relationship. For attorneys working with inventors on intricate or highly technical inventions, the more consequential change has not been in how quickly the attorney can write. It has been in how much the attorney can absorb, in part because of the quality of the reference materials provided by inventors is so enhanced.
Invention disclosure has always been a bottleneck. It was the attorney's finite capacity to get up to speed on an unfamiliar technical domain quickly enough to ask the right questions. Over the course of my career, I found most inventors are incredibly generous in spending their limited time teaching the attorney the state of the art, in addition to explaining what is actually new. However, coming up to speed quickly on emerging and/or esoteric technical fields is time consuming even with coaching from an expert. In this kind of information context, it is natural that nuance and detail can be lost and/or inaccuracies may occur. Even if these slip ups are caught in time, they make the overall process much less efficient, at the very least.
Language model technology changes where that bandwidth constraint sits. An attorney can now use these tools to construct technical background independently and quickly, surveying a field, understanding its terminology, and identifying its state of the art before an inventor conversation ever starts. That removes the inventor's old incentive to pre-simplify. The inventor no longer needs to spend the first half of a disclosure meeting teaching the attorney the field, because the attorney has already caught up. What is left is more time, and more inventor attention, devoted to the only question that actually matters for a patent application: what is different here, and why.
The second half of this shift happens after the disclosure lands, not before it. Richer, more technically detailed disclosure materials would once have been a mixed blessing, since depth beyond a certain point simply overwhelmed the attorney's capacity to process it by hand. That constraint has loosened as well. An attorney working with language model technology can synthesize far more volume and technical density without losing the thread, which means the added depth inventors are now willing to provide translates into a more complete and more accurate record, rather than into an unmanageable pile of material that gets skimmed rather than understood.
Put together, these two changes reinforce each other. Because the attorney arrives at the conversation already fluent in the relevant technical ground, the inventor discloses more, and with more precision. Because the attorney can synthesize that additional depth without being overwhelmed by it, none of that additional precision is wasted. The result is a written disclosure that more faithfully reflects the actual invention, which is the raw material every subsequent step of prosecution depends on: claim scope, the strength of the written description, and the ability to distinguish the invention from the prior art that actually matters.
None of this is a story about AI drafting a better patent application. It is a story about AI changing what an attorney is capable of hearing, and what an inventor is therefore willing to say. The quality of a patent application has always started with the quality of the disclosure behind it. What has changed is how much of that quality was previously being lost in translation, and how much of it is now getting through.