[00:00:00] Speaker 01:
Our next case this morning is 23-24-37, Resentive Analytics, Inc.

[00:00:07] Speaker 01:
versus Fox Corporation.

[00:00:10] Speaker 01:
Okay, Mr. Fredrickson.

[00:00:12] Speaker 03:
Good morning, Your Honors, and may it please the Court.

[00:00:14] Speaker 03:
Robert Fredrickson on behalf of Resentive Analytics.

[00:00:18] Speaker 03:
This case involves a practical and concrete application of a technology that's changing our world today, artificial intelligence and machine learning.

[00:00:30] Speaker 03:
The Recentive patents, and Recentive was the first to discover and recognize.

[00:00:35] Speaker 01:
Why isn't this a patent just to use the use of machine learning generically in this particular area, network schedules or whatever?

[00:00:44] Speaker 03:
Two answers to that question, Your Honor.

[00:00:46] Speaker 03:
The first is the field of the generation of network maps and event schedules was a crude and unsophisticated field.

[00:00:54] Speaker 03:
So no one before Resentive had thought to apply this specific technical solution in this specific field of endeavor.

[00:01:02] Speaker 01:
which is something that the court has looked at.

[00:01:04] Speaker 01:
We've held repeatedly that the fact that you use computers in a new area doesn't create something that's patentable under 101.

[00:01:12] Speaker 01:
Exactly.

[00:01:13] Speaker 01:
So what's different about the machine learning here?

[00:01:18] Speaker 01:
The way it's described, it sounds like conventional use of machine learning, the specification.

[00:01:24] Speaker 01:
talks about various kinds of machine learning, various kinds of AI that can be used.

[00:01:31] Speaker 01:
It doesn't specify a particular one.

[00:01:34] Speaker 03:
So with respect to the machine learning training patents, it does specify and specifically claim particular machine learning models.

[00:01:42] Speaker 03:
It also specifically claims and identifies

[00:01:45] Speaker 03:
the input data, and the target features that are going to be used to train this model.

[00:01:50] Speaker 03:
But to answer the fundamental question, what this court's jurisprudence has really focused in on when it comes to the Section 101 inquiry, and something Your Honor has noted in previous decisions,

[00:02:01] Speaker 03:
is that are these patents that claim just the result, just the optimized and dynamic network maps, or do they claim a new process for obtaining those network maps?

[00:02:14] Speaker 03:
And what machine learning is... So what's the new process?

[00:02:17] Speaker 03:
The new process is the application of machine learning to what was this crude and unsophisticated process that was being used in the prior art.

[00:02:25] Speaker 01:
But that's just applying a conventional process to a new field, which is exactly what we've held

[00:02:31] Speaker 01:
is not paneligible.

[00:02:33] Speaker 03:
But the court and the Supreme Court has also recognized that inventions often rely on building blocks that are drawn from other endeavors in other fields of inquiry.

[00:02:45] Speaker 02:
Yeah, they've talked about that in the obviousness context.

[00:02:48] Speaker 02:
Yes, that's a sound principle, but how does it apply here?

[00:02:53] Speaker 03:
Well, it goes directly to Judge Steick's question about whether you're talking about what is inventive in a particular field.

[00:02:58] Speaker 03:
You're looking to what that industry, and this is what the court said in contact instruction when we're talking about the step two industry, what is that particular field being done?

[00:03:08] Speaker 01:
So just to be clear, your assertion is that this was conventional machine learning, but what makes it patentable is it was being applied to a new field.

[00:03:19] Speaker 03:
Among other things, the machine learning patents also identify particular input data that was not being used by these conventional subjective processes.

[00:03:30] Speaker 01:
What do you mean input data?

[00:03:31] Speaker 01:
You mean input data relating to network maps?

[00:03:34] Speaker 03:
Exactly.

[00:03:34] Speaker 03:
The event parameters that are being identified as being what is the training data that's being used to achieve these optimized maps, and what are the outputs?

[00:03:43] Speaker 03:
What are we going to be looking for?

[00:03:45] Speaker 03:
And what machine learning does, it's not just do what people have been doing in a computer,

[00:03:49] Speaker 03:
using general processors, but it's effectively a special purpose computer.

[00:03:54] Speaker 03:
It's a new form of programming what people in this industry had not been doing before.

[00:04:00] Speaker 03:
And they do that by first getting historical training data of what these event schedules and what these network maps look like.

[00:04:08] Speaker 03:
And they try to, through an iterative training process, input the input data.

[00:04:13] Speaker 03:
So these are process steps, Your Honor.

[00:04:15] Speaker 03:
and the machine learning algorithm.

[00:04:16] Speaker 01:
The typical process steps for conventional machine learning are not saying the specific data, but the idea of training.

[00:04:26] Speaker 01:
The AI is conventional, right?

[00:04:31] Speaker 03:
It was known.

[00:04:31] Speaker 03:
Absolutely, Your Honor.

[00:04:32] Speaker 03:
It was known.

[00:04:33] Speaker 03:
And this wasn't an invention of a new machine learning technique, because that would fall into another one of this court's Section 101 traps, is if the claim is improving the mathematical algorithm or making machine learning better, then we're claiming the natural law, the mathematical algorithm itself.

[00:04:51] Speaker 03:
And what these patents do are claim the amm- It's not claimed.

[00:04:54] Speaker 03:
It's not claiming that exactly it's claiming the application.

[00:04:58] Speaker 02:
This case seems quite close to me to the recent most a very recent 2023 case Trinity the other side sites repeatedly including charge point and electric power.

[00:05:11] Speaker 02:
So in Trinity as I recall we were dealing with the kind of the same general notion about taking all this polling data and playing around with it and doing it in real time.

[00:05:21] Speaker 02:
as you argue here, that that must make your invention special.

[00:05:24] Speaker 02:
And the court rejected all of those arguments under 101.

[00:05:28] Speaker 03:
Understandably, because what the court in Trinity found was lacking in those claims, and in particular the claims, not the specifications, was anything that tethered how to accomplish the result that was being claimed.

[00:05:40] Speaker 03:
And what the machine learning training patents do is they specify a process that say, we are not claiming just real time

[00:05:48] Speaker 03:
application or updating of maps that we would concede is abstract.

[00:05:52] Speaker 03:
We are not claiming automated network maps, optimized network maps, but how to achieve them.

[00:05:59] Speaker 03:
So we're not preempting all ways to use computers to achieve these optimized maps, but the specifically claimed use of the machine learning algorithms using the specifically claimed

[00:06:11] Speaker 03:
input data that's associated with training data to recognize features that people doing it in the prior art were unable to unable to identify.

[00:06:24] Speaker 03:
attendance, profitability, quantifying.

[00:06:27] Speaker 03:
It was not in the prior art because people lack the capacity to understand how something like weather or gambling or some of the inputs that are identified in the specification, how that activity actually translates to those output features.

[00:06:43] Speaker 03:
People were making these network maps and coming up with broadcast schedules based on their crude subjective thoughts about human behavior.

[00:06:51] Speaker 03:
And so this was effectively, and I submit Judge Prost, this case is much closer to the McRoe line of cases.

[00:06:58] Speaker 03:
Because McRoe was the subjective process.

[00:07:00] Speaker 02:
Let's stick with Trinity, at least in various places.

[00:07:03] Speaker 02:
But in Red Reef 24, your friend McRoe asserts that despite your assertion, the claims explain how to improve software functionality.

[00:07:13] Speaker 02:
acclaimed machine learning techniques and models are couched in purely results oriented language without any explanation of how the claim machine learning technique model carries out the steps other than machine learning algorithms doing what machine learning algorithms do.

[00:07:30] Speaker 02:
read it the same way she does or he does.

[00:07:33] Speaker 02:
So can you tell me why that's wrong?

[00:07:35] Speaker 03:
Because those steps matter.

[00:07:37] Speaker 03:
And when we talk about what is the abstract idea and how you characterize the claims, the court looks at what the result is.

[00:07:44] Speaker 03:
And so the result is the output of the map.

[00:07:46] Speaker 03:
But the steps of how you get there, taking specified data, feeding it into one of these specified models, tailoring it to a particular output parameter,

[00:07:56] Speaker 02:
Aren't though, don't we have, isn't the bulk of our 101 jurisprudence where we found ineligibility dealing with precisely those things.

[00:08:06] Speaker 02:
All it does is collect the information and do this with the information, et cetera, et cetera.

[00:08:11] Speaker 02:
Those are kind of standard terms that we use in cases where we find ineligibility.

[00:08:16] Speaker 03:
Exactly.

[00:08:17] Speaker 03:
But this court has never addressed the application, a practical and concrete application of machine learning in a field that never thought to use it before.

[00:08:25] Speaker 01:
OK, so that suggests to me that you're saying that any time people use machine learning, applying it to a field where it had never been used before, they could cut a patent on it.

[00:08:35] Speaker 03:
Not any time, Your Honor, and certainly I don't want to take on more than I need to in terms of all the possible iterations and how fast... Lots of times?

[00:08:43] Speaker 03:
Frankly, sure, Your Honor, there should be lots of times.

[00:08:46] Speaker 03:
But fast doesn't matter.

[00:08:48] Speaker 03:
That's part of the problem because I'm receiving questions about machine learning, doing what machine learning does.

[00:08:55] Speaker 03:
Machine learning has only been around and practically... But you didn't invent machine learning.

[00:08:59] Speaker 03:
Correct.

[00:09:00] Speaker 03:
The concepts have existed in academia and in research and in literature, but people couldn't figure out how to bring those tools to bear, how to make practical applications.

[00:09:10] Speaker 03:
And we're only seeing that today in different areas, and people are excited about and talking about all the things that machine learning and artificial intelligence can do.

[00:09:19] Speaker 03:
And so to foreclose the notion that people recognizing a new application of this very powerful tool to all these potentially future areas of endeavor is, I submit, a scary proposition for people working in this field.

[00:09:33] Speaker 03:
So I don't want to say that every possible application of machine learning in every possible field is going to be subject to patent eligibility.

[00:09:41] Speaker 03:
And there very well will be, obviously, a zero-limiting principle.

[00:09:45] Speaker 02:
I mean, you just did say that the fact that it's machine learning and it's introducing it, that's sufficient to claim eligibility.

[00:09:54] Speaker 02:
But then you said, I don't mean to say that it applies in every field.

[00:09:57] Speaker 02:
So it's limiting principle.

[00:09:59] Speaker 02:
Why would it apply in this case and not in every other case?

[00:10:02] Speaker 03:
Well, there have been few limiting principles when it comes to Section 1 jurisprudence.

[00:10:08] Speaker 03:
And so I would submit that when it comes to these types of inventions that do claim a how, and you may think that it's just the use of machine learning, but it is a how, and it is different than what was being done in the industry, that you take up those cases on a case-by-case basis.

[00:10:22] Speaker 03:
Here we have a very unsophisticated prior art system that Fox admits was less sophisticated than what the output is.

[00:10:31] Speaker 03:
There is a concrete improvement.

[00:10:33] Speaker 03:
The application of these machine learning models allows the network map makers to react in real time.

[00:10:41] Speaker 03:
We're not just saying react in real time and tell you how.

[00:10:44] Speaker 01:
Yeah, but your problem is that with respect to computers, the same thing could be said.

[00:10:49] Speaker 01:
Computers were applied to new fields.

[00:10:51] Speaker 01:
People sought

[00:10:52] Speaker 01:
patents on computerizing various kinds of data processing, and we've said that merely applying a computer to a new field, a conventional computer to a new field, is not patent eligible.

[00:11:06] Speaker 01:
Why is this different?

[00:11:08] Speaker 03:
Because where's the line drawing on that side of the calculus?

[00:11:11] Speaker 03:
Computers, I agree, general purpose computers, receiving data, analyzing data, outputting data, use a computer to do it, that falls.

[00:11:20] Speaker 03:
That's Judge Prost.

[00:11:21] Speaker 03:
The long line of cases that this court has held is ineligible.

[00:11:24] Speaker 03:
This series of claims

[00:11:26] Speaker 03:
focuses on a particular use of those computers.

[00:11:30] Speaker 03:
And in McRoe and in Conoclica, the court didn't grapple with how sophisticated or how unsophisticated the rules were in those cases.

[00:11:39] Speaker 03:
In McRoe, the rule that got the patentee over the subject 101 hump was basically associating the facial expressions with time.

[00:11:50] Speaker 03:
And the defendant in that case said, well, you didn't tell us any specific rules.

[00:11:55] Speaker 03:
You didn't claim any specific rules.

[00:11:57] Speaker 03:
And what the court said was, that's OK, because it was a genus of rules that were being created.

[00:12:02] Speaker 03:
That claim didn't preempt all ways of using computer.

[00:12:06] Speaker 03:
It didn't preempt all rules to doing 3D animations.

[00:12:10] Speaker 03:
The same is absolutely true here.

[00:12:12] Speaker 03:
These claims do not present all uses of machine learning.

[00:12:15] Speaker 03:
They do not preempt all uses of generating optimized network maps.

[00:12:19] Speaker 03:
And they don't generate and they don't preempt all uses of computers.

[00:12:23] Speaker 03:
I see I'm into my rebuttal time.

[00:12:25] Speaker 01:
OK, you want to save it?

[00:12:30] Speaker 01:
Mrs. Acara?

[00:12:42] Speaker 00:
May it please the court, Ranjani Acharya for the Appellee Sports Corporation.

[00:12:46] Speaker 00:
Sorry.

[00:12:47] Speaker 00:
No problem.

[00:12:50] Speaker 00:
Unless there are questions, I'd like to pick up where my colleague left off, which is the idea that Resentive here has applied machine learning to what was they previously characterized as a previously unsophisticated process.

[00:13:05] Speaker 00:
My friend said people couldn't figure out how to bring these tools to bear until Resentive came along.

[00:13:10] Speaker 00:
Now, I think Resentive should be justly proud of its commercial success.

[00:13:16] Speaker 00:
They've been the first to offer a technology that the market has reacted to.

[00:13:20] Speaker 00:
That's not the question in a one-on-one analysis.

[00:13:24] Speaker 02:
Well, how is this different than McGrow?

[00:13:26] Speaker 02:
He seems to rest heavily on McGrow.

[00:13:29] Speaker 00:
Well, with respect to McGrow, the claims at issue in McGrow really dealt with taking something that was done using subjective human inputs

[00:13:40] Speaker 00:
and using the computer, the functionality of a computer in order to change the process and to improve the display in some way.

[00:13:52] Speaker 00:
The claims that issue here don't actually, if you look at the language of the claims, there's no improvement to computer functionality that's recited.

[00:14:02] Speaker 00:
What's recited is taking known types of data, training a machine learning model,

[00:14:07] Speaker 00:
to recognize what the claims describe as relationships, identify relationships in that data, and then create a network map or event schedule that's optimized to prioritize attendance, maybe viewership in a certain demographic, et cetera.

[00:14:26] Speaker 00:
So things that were known to routine processes that were done before but now applying machine learning techniques to them.

[00:14:35] Speaker 00:
And with respect to the machine learning training patterns, certainly the claims do specify two particular types of machine learning models.

[00:14:45] Speaker 00:
But they don't recite any particular improvement on those machine learning models.

[00:14:49] Speaker 00:
The specification makes it clear that the models in question are known techniques that were conventional in the art.

[00:15:00] Speaker 00:
And Rescent have told the Patent Office

[00:15:02] Speaker 00:
during the prosecution of the machine learning training patents, that this idea of iteratively training to improve the model's accuracy is doing what machine learning does.

[00:15:13] Speaker 00:
And that's at Appendix 355 in the prosecution history.

[00:15:18] Speaker 00:
So unlike McCrow, I think this case is a lot closer to the SAP decision of this court, where this court did distinguish McCrow.

[00:15:28] Speaker 00:
In SAP, this court looked at,

[00:15:31] Speaker 00:
methods for performing statistical analyses of investment information.

[00:15:36] Speaker 00:
And the court said there it was selecting information, analyzing it using mathematical techniques, reporting, displaying the results.

[00:15:44] Speaker 00:
That is all abstract.

[00:15:46] Speaker 00:
And the court distinguished McCrow, which was directed to the creation of something physical, how the physical display operated to produce better quality images.

[00:15:57] Speaker 00:
That's not what we're talking about here.

[00:15:59] Speaker 00:
Language of the claims doesn't say take these concrete steps in order to improve machine learning.

[00:16:06] Speaker 00:
It doesn't say take these concrete steps to identify patterns that were not available to humans before.

[00:16:15] Speaker 00:
All the claims say is input two kinds of data, iteratively train the machine learning model to identify relationships.

[00:16:25] Speaker 00:
Give it a weight for a certain parameter that you're interested in, and then have it generate a network map or event schedule that is optimized for that feature.

[00:16:35] Speaker 00:
So this is a lot closer to SAP than macro, and the reasoning of SAP applies equally here.

[00:16:43] Speaker 02:
And the district court's opinion dealt with a lot of arguments made with regard to McCrow, right?

[00:16:48] Speaker 02:
Correct.

[00:16:49] Speaker 02:
And the district court also relied heavily on Trinity.

[00:16:52] Speaker 02:
Yes.

[00:16:52] Speaker 02:
So are you familiar with that case?

[00:16:54] Speaker 02:
And do you think that case compels the result here?

[00:16:57] Speaker 00:
Yes, Your Honor.

[00:16:57] Speaker 00:
I think it certainly serves as a very useful guidepost.

[00:17:01] Speaker 00:
And the district court did discuss it extensively in the order as well, because the case had just come out.

[00:17:06] Speaker 00:
And the reason I think it applies very strongly here, or is a good guidepost here, is

[00:17:13] Speaker 00:
In Trinity, you had this idea of real-time polling.

[00:17:17] Speaker 00:
So this is technology that, as computers improve, as machine learning algorithms become more powerful, you're able to do these types of analyses in real time that perhaps were not available to prior art systems.

[00:17:34] Speaker 00:
And in Trinity, this court held correctly that doing something in real time is

[00:17:42] Speaker 00:
shorthand maybe for saying do it faster or more efficiently than what a human can do it and we've known since Alice that using a computer functionality that is running on basic generic computer hardware known computer technology with no improvements to that functionality to do something that is faster or more efficient quicker than what a human can do by looking at real-time data

[00:18:10] Speaker 00:
That's just not sufficient.

[00:18:12] Speaker 00:
And so this is not really an outlier case.

[00:18:14] Speaker 00:
I think this falls very neatly within the line of cases, including Trinity most recently.

[00:18:27] Speaker 00:
If I may continue, I also heard my colleague mention that the key technology point here is that

[00:18:39] Speaker 00:
The machine learning is being applied to special types of data, event parameters, target features, and the specification makes it very clear that these are in fact known types of data.

[00:18:53] Speaker 00:
When we talk about event parameters, we are looking at attendance, we're looking at the schedule of a particular performer.

[00:19:02] Speaker 00:
When we're talking about target features, it's things like profits,

[00:19:07] Speaker 00:
maximizing viewership in a particular market or a demographic.

[00:19:11] Speaker 00:
So there isn't anything in the claims that applies machine learning to a very specific type of data.

[00:19:19] Speaker 00:
What we're left with is exactly what Your Honor's pointed out.

[00:19:23] Speaker 00:
We have here generic off-the-shelf computers running known machine learning technologies.

[00:19:30] Speaker 00:
The claims say just have the machine learning

[00:19:33] Speaker 00:
do what machine learning does and apply it in this particular field.

[00:19:38] Speaker 00:
And that's just not enough.

[00:19:40] Speaker 00:
Since Alice, we've known that those kinds of field of use limitations are not enough to add an inventive concept here.

[00:19:52] Speaker 00:
I think we have spent some time today talking about the machine learning training patterns.

[00:19:58] Speaker 00:
I do want to emphasize for the report

[00:20:01] Speaker 00:
Network map patterns which are the other two patterns that issue in this case like even the narrowing sort of limitations of the machine learning training patterns they truly are take the NFL's network map for programming a football season and Apply machine learning to it so to the extent that the machine learning training patterns are abstract and add no inventive concept which we believe is the case and

[00:20:30] Speaker 00:
the network map patterns are even more vulnerable.

[00:20:34] Speaker 00:
As to the remaining arguments, I am happy to rest on the briefs, unless Your Honors have further questions for me.

[00:20:40] Speaker 01:
OK.

[00:20:41] Speaker 01:
Thank you, Ms.

[00:20:41] Speaker 01:
Itchariah.

[00:20:42] Speaker 00:
Thank you very much.

[00:20:44] Speaker 01:
Mr. Fredrickson, you have two minutes and 47 seconds.

[00:20:54] Speaker 03:
I will start with Trinity, because Judge Prost, you've asked about it a couple of times.

[00:20:58] Speaker 03:
I think the important thing to do is not to take just the buzzwords out of the opinions out of context, but to look at the actual claim language at issue in that case.

[00:21:06] Speaker 03:
The claims at issue in that case required receiving user information, providing polling questions, receiving answers, and then figuring out if those answers matched other answers.

[00:21:19] Speaker 03:
And what the court found in an analysis of those claims was those steps of matching questions to answers with others' questions to answers, those are steps that can be performed in the human mind.

[00:21:30] Speaker 03:
And so that was the first sort of linchpin of the court's analysis in that decision.

[00:21:35] Speaker 03:
That is not at issue here because it is pleaded and doesn't appear to be disputed that the steps of machine learning are not processes that are done in the human mind.

[00:21:45] Speaker 03:
The second important thing about the Trinity decision is the court repeatedly said that those very generic and human interaction steps of matching polling questions, the claims were not tethered

[00:21:58] Speaker 03:
to any specific technological solution or did not require specialized computing components.

[00:22:05] Speaker 03:
You may think that the use of machine learning is conventional, but it is a specialized computer solution and it's a specific technical solution.

[00:22:15] Speaker 03:
It's not just do something and achieve a result in a computer, it's use a computer to train itself using specified algorithms in order to identify patterns

[00:22:26] Speaker 03:
that people can't see.

[00:22:28] Speaker 03:
Now you contrast that with what my friend on the other side said about McRoe.

[00:22:32] Speaker 03:
And the way that she distinguished McRoe was she said, well, that was a case involved a subjective human input, the prior art.

[00:22:39] Speaker 03:
Well, that's the same true here.

[00:22:40] Speaker 03:
The prior art maps were subjective human inputs.

[00:22:43] Speaker 03:
and the functionality was improved of these animations by applying the rules.

[00:22:49] Speaker 03:
The same thing is true here.

[00:22:51] Speaker 03:
It is undisputed that the network maps that are generated by machine learning algorithms are markedly superior than crude

[00:23:00] Speaker 03:
generalized subjective maps that are being generated by humans.

[00:23:04] Speaker 03:
And that's actually reflected in the claims.

[00:23:07] Speaker 03:
The claims specifically require that the network maps be dynamic.

[00:23:12] Speaker 03:
That can't happen if a human is just guessing as to whether you want to see a Cleveland game in one geographic geography or a Dallas game.

[00:23:19] Speaker 03:
The network maps are automated and updated in real time.

[00:23:23] Speaker 03:
The only way that's accomplished is by having a trained model that can instantaneously recognize when there's changes in input data, run it through this model that's been changed using a specific method, and output the optimal network map.

[00:23:39] Speaker 03:
And the final point I'll just make is these SAP and Stanford cases.

[00:23:42] Speaker 03:
have nothing to do with the recensive patents, because those cases are all about improved mathematical algorithms.

[00:23:50] Speaker 03:
If you claim an improvement in math, if that's the inventive contribution, this court and the Supreme Court says, that falls in the realm of natural law and abstract idea.

[00:23:59] Speaker 03:
The recensive patents claim no improved math.

[00:24:02] Speaker 01:
Thank you very much.

[00:24:03] Speaker 01:
Thank you, Mr. Rickson.

[00:24:04] Speaker 01:
And both counsel, the case is submitted.