Is the USPTO Automating Art Unit Assignment? Here’s the Evidence.
When a patent application is filed with the USPTO, it’s assigned to an art unit and a patent examiner in that art unit based on the technology involved.
The USPTO “classifies” an application by assigning it codes to indicate the technological subject matter of the patent. Then an art unit and examiner are selected based on the codes. Only then does the examiner leverage their subject matter expertise to review the application, do a prior art search, and determine whether to grant a patent.
In the past, the assignment process was manual. One group of USPTO employees reviewed applications and assigned codes and another group assigned the application to an art unit and examiner based on the codes.
More recently, we believe that the USPTO has begun automating the assignment process using machine learning.
How do we know the USPTO is automating this process? We’ve verified it using our proprietary Art Unit Predictor.
USPTO Automation? The Proof is in the Predictions.
The Art Unit Predictor by Patent Bots reviews patent applications and predicts the art units to which they will be assigned. As patent practitioners, we can use this tool to revise our applications and steer them toward favorable art units.
Over the last two years, we’ve seen the Art Unit Predictor’s accuracy rate improve dramatically.
When the process was fully manual, human inconsistency impacted the assignment process. Two different USPTO employees could classify the same patent with two different codes. Even the same USPTO employee could classify the same patent differently depending on the day.
By contrast, machines are incredibly consistent. When machines classify applications, the outcome will always be the same. Machines are capable of learning even complex decision-making processes as long as the process is consistent.
The increasing consistency in USPTO art unit assignment – and increasing alignment with Patent Bots predictions – indicates greater use of automation in the process over the last two years.
The following graph illustrates this increasing alignment between (presumably automated) USPTO assignments and Patent Bots predictions over the last two years:
How Automation Benefits All Patent System Stakeholders.
The Art Unit Predictor predicts which of five art units an application will be assigned to. As the chart illustrates, our prediction accuracy has steadily increased from 80% in the first half of 2020 to over 82% in the second half of 2021.
During this period, the implementation of our Art Unit Predictor has not changed. The only change is the USPTO data used to train and test the Art Unit Predictor.
The USPTO has stated publicly that it is introducing more automation into the patent assignment process.
Our increased accuracy indicates that the USPTO has become more predictable. It provides third-party validation that, yes, the USPTO is replacing manual human processes with automation.
We expect our accuracy to continue increasing as the USPTO automates more of the process. When the process becomes fully automated, we may even see 100% accuracy from our Art Unit Predictor.
Predictability through automation is good for the patent system! Automation lowers the USPTO’s costs and speeds up processing. It leads similar applications to the same art units and those art units will have (or develop) relevant expertise in those technology areas.
And, most importantly, automation promotes greater consistency and thus fairer outcomes for patent applicants.
