How high will it go? An update on the allowance rate for applications relating to artificial intelligence at the USPTO

It has been about a year since my last post on allowance rates for applications relating to artificial intelligence at the USPTO, and it seemed like a good time for an update.

In short, the allowance rate is still around 90% for applications relating to artificial intelligence, but, notably, the allowance rate in 2019 actually surpassed 90% for this technology. This may be partially due to the updated guidelines distributed by the USPTO at the beginning of the year, which "aim to improve the clarity, consistency, and predictability of actions across the USPTO." Actions are sent out by examiners to prevent the office from granting patent rights over technologies that were already publicly accessible (described exactly or as an obvious variant) at the time of filing. The USPTO could attain perfect clarity, consistency, and predictability of actions with a 100% allowance rate, but at what cost to the public?



METHODOLOGY

In the study, to find what I consider "artificial intelligence" cases, I searched for any of the following terms in the title, abstract, and claims of applications and patents with priority dates reaching back to 2011:

fuzzy logic
classifier model
artificial neural network
support vector machine
feature vector
training data
learned model
k-means clustering
naive bayes
decision tree
reinforcement learning
supervised learning
unsupervised learning
semi-supervised learning
artificial intelligence
machine learning

PRIORITY 2011 AI CASES

There were 2936 finally disposed AI cases with priority in 2011, of which 437 were abandoned and 2499 were granted for an allowance rate of 85.1%. Notably, there are only 219 such cases still pending at the USPTO. Even if all of those cases are ultimately abandoned, the allowance rate would still be 79.2%. I treated 2011 separately as a priority date because the law has changed quite a bit since then, and because these cases have had more than enough time to run their course at the USPTO.

Of the 219 remaining cases,
12.8% are in art units 2121-2124 ("data processing", including "data processing: artificial intelligence") with a weighted average art unit allowance rate of 71.2%;
9.1% are in art units 3622-3628  ("data processing" relating to finance or business methods) with a weighted average art unit allowance rate of 28.0%;
5.5% are in art units 3681-3689 ("data processing" relating to finance or business methods) with a weighted average art unit allowance rate of 45.2%; and
5.0% are in art unit 1631 ("data processing: measuring, calibrating, or testing") with an allowance rate of 36.1%.




PRIORITY 2012-TODAY AI CASES

There were 13,605 finally disposed AI cases with priority in 2012 or later, of which 1765 were abandoned and 11,840 were granted for an allowance rate of 87.0%.

Of those, there were 7,019 finally disposed AI cases with priority in 2014 or later, of which 732 were abandoned and 6287 were granted for an allowance rate of 89.6%.


EFFECT OF THE 2019 USPTO GUIDANCE

From October to December 2018 (inclusive), there were 1211 finally disposed cases (181 abandoned; 1030 granted) for an allowance rate of 85.1%.

The guidance was announced on Jan. 4th and took effect on Jan. 7th.

In 2019, there were 809 finally disposed cases (76 abandoned; 733 granted) for an allowance rate of 90.6%.  That is a big jump from a high starting point.

USEFUL TERMINOLOGY IS STILL USEFUL

In past studies, I concluded that comparison/iterative claim terms were generally helpful to the applicant to obtain an allowance. That wisdom still holds true in the sample for priority 2012 and later. The chart below shows the number of times these terms appeared in the claims (N(Claims)) for the AI cases priority 2012 and later. In the next column, the chart shows the allowance rate when the corresponding term appeared (AR(Claims)).

COMPARISON/
ITERATIVE  CLAIM TERMS
N(Claims)
AR (Claims)
differ*
5408
90.0%
less
1533
90.9%
frequen*
2018
89.7%
threshold
4192
90.5%
after
2115
89.3%

BAD TERMINOLOGY IS STILL BAD

In past studies, I concluded that business or broad field of use terms were generally harmful to the applicant in pursuit of allowance. That wisdom still holds true in the sample for priority 2012 and later. The chart below shows the number of times these terms appeared in the claims (N(Claims)) for the AI cases priority 2012 and later. In the next column, the chart shows the allowance rate when the corresponding term appeared (AR(Claims)). While the noted allowance rates are still high (60-70%), they are significantly below average for the sample (87.0%).
BUSINESS OR BROAD FIELD OF USE CLAIM TERM
N(Claims)
AR (Claims)
payment
144
68.8%
health
271
66.8%
busi*
330
73.9%
transact*
355
78.3%


HOW HIGH IS TOO HIGH OR HIGH ENOUGH?

With the allowance rate for AI cases surpassing 90% in 2019, it begs the question of where the ceiling for allowance rates might be. How high is too high, or how high is high enough? It depends on whether you ask a person trying to enforce patents, a person trying to use public knowledge, or a patent attorney trying to make a living. Will we see a push even higher by the USPTO in 2019?

What happens to all of these patents granted by the USPTO? For most of them, nothing happens other than the USPTO collecting fees. For some, they get asserted against operating companies. When they do, 63-85% have all claims invalidated at the district court (https://www.rpxcorp.com/intelligence/blog/q2inreview/), and about 80% that reach the Fed. Cir. level have all claims invalidated (http://www.patnotechnic.com/2018/04/reconciling-alice-review-between-patent.html). In other words, it was a mistake to grant a high percentage of litigated patents, but the grant rate continues to increase.

Comments