[00:00:00] Speaker 00:
The first one is number 241725, Rensselaer Polytechnic Institute versus Amazon.

[00:00:07] Speaker 00:
Mr. Zhu.

[00:00:09] Speaker 01:
Good morning, Your Honors.

[00:00:11] Speaker 01:
Li Zhu for Appellants, Rensselaer Polytechnic Institute and CF Dynamic Advances.

[00:00:17] Speaker 01:
This case asks whether a district court may set aside its own claim constructions and reduce a specific computerized method to mirror storing and searching information under Section 101.

[00:00:31] Speaker 01:
Through two detailed markman orders and six years of litigation, the court found that the claims require, among other things, a constrained search based solely on the natural language input, a specific metadata database, the use of four categories of metadata, all four including case information, and connections among those categories.

[00:00:55] Speaker 00:
How is this different from Rosenta?

[00:00:58] Speaker 01:
Well, Your Honor, respectfully, Recintive Analytics analyzed four patents that attempted to claim the very nature of machine learning.

[00:01:06] Speaker 01:
The claims didn't specify any new way for the computer to operate, only a goal which was within event scheduling.

[00:01:14] Speaker 01:
The court also found that Recintiv had, quote, conceded the lack of a technological improvement.

[00:01:20] Speaker 01:
And the claims and the specification didn't have that improvement as well.

[00:01:25] Speaker 01:
By contrast, here, the claims are directed to a specific technological improvement over existing natural language processing systems.

[00:01:33] Speaker 01:
And that is reflected in the claim constructions.

[00:01:35] Speaker 00:
I was using a hack, right?

[00:01:38] Speaker 01:
That's one way, Your Honor.

[00:01:39] Speaker 01:
I would say that artificial intelligence, that includes a broad swath of different methods and implementations.

[00:01:46] Speaker 00:
But it can't be non-abstract to use AI for natural language processing, right?

[00:01:54] Speaker 01:
Well, I think that's a fairly loaded question, Your Honor, respectfully.

[00:01:57] Speaker 00:
There's a specific... But what's the answer to it?

[00:02:00] Speaker 01:
The answer to that question is that the claims here require a lot more.

[00:02:04] Speaker 01:
We know from the claim construction... And that would answer my question.

[00:02:06] Speaker 00:
If it were just the use of natural AI in connection with natural language processing would be abstract, right?

[00:02:14] Speaker 01:
If that's all the claims said, that's correct, Your Honor.

[00:02:16] Speaker 00:
It would be an abstract.

[00:02:17] Speaker 00:
Okay.

[00:02:17] Speaker 00:
So what else is there here?

[00:02:18] Speaker 01:
Sorry?

[00:02:19] Speaker 00:
What else is there here?

[00:02:21] Speaker 01:
Well, there is a connected enterprise metadata database, and we know that from the construction for the term information models, which were construed at appendix 2613 as, quote, the webs of concepts for enterprise databases.

[00:02:35] Speaker 01:
And these are connections that must exist between the four claim metadata categories, and these are structural connections in the database.

[00:02:43] Speaker 01:
These connections are also captured through claim 17s

[00:02:47] Speaker 01:
a claim for in 14-17's reference dictionary, which was construed as, quote, the set of database entries and their relations.

[00:02:57] Speaker 01:
And the specification repeatedly describes this interconnected nature of the metadata database.

[00:03:03] Speaker 01:
For example, in column 17, line 6-24, and that's the section entitled Case-Based Reasoning, Your Honor, it describes how... I'm not sure that I understand what you're saying.

[00:03:12] Speaker 00:
You're saying that this uses a different kind of database?

[00:03:16] Speaker 01:
That's correct, Your Honor.

[00:03:17] Speaker 01:
A never-before-seen database.

[00:03:19] Speaker 00:
So I'm not sure that you raised that issue below.

[00:03:24] Speaker 00:
Where did you raise that issue below?

[00:03:27] Speaker 01:
Specifically with respect to the claim constructions for information model, where we describe it as an interconnected database and in the way that the floor packs information.

[00:03:35] Speaker 00:
Where did you argue that the nature of the database made it non-abstract?

[00:03:40] Speaker 01:
That's in our argument with respect to infish, Your Honor.

[00:03:43] Speaker 00:
Would we compare the image?

[00:03:43] Speaker 00:
Could you show me where you argued that?

[00:03:46] Speaker 01:
Yes, Your Honor.

[00:04:17] Speaker 01:
So in our reply, Your Honor, in the argument and reply, page one, we specifically describe how the claimed invention is directed to applying.

[00:04:25] Speaker 00:
You're talking about the reply brief here?

[00:04:28] Speaker 01:
That's correct, Your Honor.

[00:04:29] Speaker 00:
No, I'm not asking you that.

[00:04:30] Speaker 00:
I'm asking where you raised this below.

[00:04:32] Speaker 01:
Oh, apologies, Your Honor.

[00:05:17] Speaker 00:
Well, if you can't find it, you can try giving it on revival.

[00:05:21] Speaker 01:
I apologize, Your Honor.

[00:05:22] Speaker 01:
It's at appendix 4406.

[00:05:23] Speaker 00:
4406?

[00:05:25] Speaker 01:
Where we describe the exemplary claim one as being directed to a specific metadata database with an interconnected architecture of four key types of metadata.

[00:05:35] Speaker 01:
And we specifically mentioned the four enumerated categories of metadata.

[00:05:39] Speaker 00:
And where does it argue that that makes it non-abstract?

[00:05:42] Speaker 01:
uh... well your honor if that's not the only thing in the claims that makes uh... claim one not abstract but where do you argue that the specific nature of the database renders it non-abstract it's the interconnected nature of the database where do you argue this apart from appendix 4406 we also describe where is it on i don't see where you argue that on 4406 so that's

[00:06:10] Speaker 01:
Page 13 of our summary judgment opposition, where we're arguing against the high level of abstraction that was argued by appellees below.

[00:06:22] Speaker 01:
And we specifically say, five lines above the bottom of that page, exemplary claim one is directed to a specific metadata database with an interconnected architecture of four types of data.

[00:06:34] Speaker 00:
OK, but where does that say that makes it non-abstract?

[00:06:38] Speaker 01:
It's within the context of our argument that claims are directed to a new method for processing natural language input by storing the four types of metadata and then using that metadata.

[00:06:48] Speaker 01:
And the claims and the specification describe how that information is used and why that metadata database has to be interconnected.

[00:06:57] Speaker 01:
That's in column 17 of the specification specifically, where it describes, quote, keywords and cases are connected to information models and database values through particular meta-relationships.

[00:07:09] Speaker 03:
If I assume that you've preserved that argument, help me with recentive, though, because I'm sure you're aware of the very last line of recentive.

[00:07:17] Speaker 03:
We say we hold.

[00:07:18] Speaker 03:
that patents that do no more than claim the application of generic machine learning to new data environments without disclosing improvements to machine learning models are patent ineligible, obviously, therefore abstract.

[00:07:33] Speaker 03:
Isn't your answer that you've just given us precisely within the scope of that holding, you're saying just do this in a new data environment, aren't you?

[00:07:43] Speaker 01:
But it's not just that, Your Honor.

[00:07:45] Speaker 01:
It's looking at how the specification describes choir art systems.

[00:07:49] Speaker 01:
Natural language processing systems had never used case information before, and it was never architected in the way as shown in the claims here.

[00:07:57] Speaker 00:
But that's exactly what Rescindive says is not sufficient to render it non-abstract.

[00:08:04] Speaker 00:
The use of AI in natural language processing you agreed earlier doesn't render it non-abstract.

[00:08:12] Speaker 01:
I agree again, Your Honor, that the use of artificial intelligence itself, if that's all the claims say, that wouldn't render an invention non-abstract.

[00:08:22] Speaker 01:
But here we have the interconnected metadata database.

[00:08:24] Speaker 01:
Here we also have the constructions for case information in case.

[00:08:28] Speaker 01:
which were construed as specifically prior instances of use of the natural language processing method, and not just any past language data, and not just of any NLP system, but also it's reflected within the court's construction of the without augmentation limitation, Your Honor.

[00:08:44] Speaker 01:
And as construed, that claim search must be, quote, based solely on the natural language input.

[00:08:51] Speaker 03:
Where does the patent itself tell us how the technique applies case-based reasoning?

[00:08:56] Speaker 01:
That's a good question, your honor.

[00:08:58] Speaker 01:
The patent specifically describes case-based reasoning in numerous places.

[00:09:05] Speaker 01:
And I would begin by directing the court to column 17, where it describes the keywords in the cases being connected to the information models.

[00:09:17] Speaker 01:
But it's also reflected, for example,

[00:09:21] Speaker 01:
In column six, work describes, based on this analysis, a reference dictionary is used that integrates enterprise metadata, information models, and contextual knowledge with case-based reasoning.

[00:09:32] Speaker 01:
In fact, the summary of the invention describes the case-based interaction method.

[00:09:36] Speaker 01:
But more specifically, column 16 in the section entitled case-based reasoning details the structure of the cases and then the connections between the cases and the three other metadata types.

[00:09:47] Speaker 03:
Let me assume for the moment it's in columns 6, 16, and 17.

[00:09:51] Speaker 03:
What is in the claims other than just functional language?

[00:09:56] Speaker 01:
Respectfully, Your Honor, there's more than just functional language when the claims are interpreted in light of their claim constructions.

[00:10:04] Speaker 01:
So here, the district court recognized that case information in cases are prior instances of the use of the natural language processing method at Appendix 41.

[00:10:15] Speaker 00:
That information is AI, right?

[00:10:18] Speaker 01:
Well, taking a historical record of how this specific patented information was being used in prior uses, that's not something that ever exists within natural language processing.

[00:10:29] Speaker 01:
And I don't believe the court is saying that anything that uses an artificial intelligence would therefore not be patent eligible.

[00:10:36] Speaker 01:
And here in the claims, there's a specific implementation of that artificial intelligence of the specific form of case-based reasoning discussed within the specification and as construed.

[00:10:46] Speaker 03:
But even if case-based information is as construed and it's got that meaning in the claims, all the claims tell us to do is provide, perform, provide, identify, and determine.

[00:10:58] Speaker 03:
It doesn't really tell us anything other than those functions, does it?

[00:11:03] Speaker 01:
Well, the claims claim and the specification teaches.

[00:11:06] Speaker 01:
But I would also point to the court's claim constructions for exactly those two limitations.

[00:11:12] Speaker 01:
So the determining the combinations in claim one and the interpreting the permutations in claim nine, as construed at the markman order at appendix 27, specifically are used to, quote, determine candidate interpretations

[00:11:26] Speaker 01:
of the user's natural language input.

[00:11:28] Speaker 01:
And we know from column seven in the specification, lines 49 through 60, that it is this logical structure that is critical to how the invention operated.

[00:11:37] Speaker 01:
You take away any of those elements and you wouldn't have the improvement that's reflected in the claims and the specification.

[00:11:44] Speaker 03:
I got one other.

[00:11:45] Speaker 03:
The district court notes in both footnotes three and six that you oppose summary judgment in part based on a purported fact dispute.

[00:11:55] Speaker 03:
Can you tell me what that fact dispute was?

[00:11:57] Speaker 03:
And are you making that argument here, or is that not one we have to deal with?

[00:12:00] Speaker 01:
Well, there are two facts that are especially relevant here, Your Honor.

[00:12:05] Speaker 01:
The first is that it is undisputed that there was no case-based reasoning within natural language processing.

[00:12:11] Speaker 01:
And the second is, with respect to- That's not fact dispute, right?

[00:12:14] Speaker 01:
You say that's conceited.

[00:12:16] Speaker 01:
Right.

[00:12:17] Speaker 01:
And with respect to the mental process argument that was made below, there is a factual, at worst a factual dispute at best, a conceited fact from the other side that a person of ordinary skill or a human wouldn't be able to implement the claim process in their minds.

[00:12:35] Speaker 03:
But did you argue below summary judgment on 101 should not be granted to them?

[00:12:39] Speaker 03:
because of that second fact dispute, or is that not what the district court's talking about in these two footnotes?

[00:12:45] Speaker 01:
So that wouldn't be what the district court is referring to in those two footnotes, but we would say that those concessions and those material facts are relevant to the 101 assessment.

[00:12:54] Speaker 03:
Is there any fact dispute here that you're now arguing to us should cause us to reverse and remand, or is that not the case?

[00:13:01] Speaker 01:
No more fact dispute, Your Honor.

[00:13:05] Speaker 00:
Okay.

[00:13:05] Speaker 00:
Do you want to save your rebuttal time?

[00:13:07] Speaker 01:
Yes, Your Honor.

[00:13:08] Speaker 01:
Thank you very much.

[00:13:23] Speaker 02:
may please the court.

[00:13:25] Speaker 02:
The problem with plaintiffs' appellants' position here and the arguments we just heard from Mr. Zhu is that they rely on material that is not claimed.

[00:13:36] Speaker 02:
This interconnected architecture is not in the claims.

[00:13:39] Speaker 02:
Any specific way of using case-based reasoning is not in the claims.

[00:13:44] Speaker 02:
Any improvement

[00:13:46] Speaker 02:
purportedly disclosed in the specification is not in the claims.

[00:13:50] Speaker 03:
Now, I just want to make sure I understand your argument.

[00:13:52] Speaker 03:
When you say it's not in the claims, you mean it's also not in the claim constructions adopted by the district court, which are unchallenged here.

[00:14:01] Speaker 02:
That's correct, Your Honor.

[00:14:03] Speaker 02:
And I can go through those claim constructions specifically.

[00:14:06] Speaker 02:
So Mr. Zhu mentioned information models.

[00:14:12] Speaker 02:
The construction there is webs of concepts for enterprise databases.

[00:14:17] Speaker 02:
Every database has webs of concepts in it.

[00:14:20] Speaker 02:
It might represent orders.

[00:14:22] Speaker 02:
It might represent customers.

[00:14:24] Speaker 02:
It might represent items that are purchased.

[00:14:27] Speaker 02:
And all of those concepts are linked together in any database.

[00:14:30] Speaker 02:
In fact, if you look at appendix 78, which is column 12 of the patent, lines 12 to 24, it explains

[00:14:40] Speaker 02:
that information models are webs of concepts for enterprise databases.

[00:14:48] Speaker 02:
And it's clear from the context that those enterprise databases are conventional.

[00:14:54] Speaker 01:
This is column 12?

[00:14:55] Speaker 00:
Yes, Your Honor.

[00:14:59] Speaker 02:
What line?

[00:15:00] Speaker 02:
Is it 12 to 24?

[00:15:00] Speaker 02:
Is that right?

[00:15:01] Speaker 02:
Yes, this is where the construction comes from.

[00:15:10] Speaker 02:
says, in the database field, information models are webs of concepts for enterprise databases.

[00:15:15] Speaker 02:
So this was already the preexisting notion of information models that was known in the field.

[00:15:21] Speaker 03:
Just to make sure I'm following your argument, that means, I think, webs of content, webs of interconnected content, are in the claims.

[00:15:30] Speaker 03:
It's just it's nothing novel.

[00:15:32] Speaker 02:
Right, that's just describing how normal databases work.

[00:15:35] Speaker 02:
They have an information model.

[00:15:36] Speaker 03:
I thought your argument was nothing that the appellant says is in the claims is actually in the claims.

[00:15:40] Speaker 03:
But at least this part you concede is in the claims.

[00:15:44] Speaker 02:
Right, there's no technological improvement in the claims, because that's just referring to a normal database, Your Honor.

[00:15:52] Speaker 02:
Also, interconnections in databases, this court has repeatedly held, for example, in Erie, SAP, and Resentive,

[00:16:01] Speaker 02:
don't make a claim patent eligible.

[00:16:03] Speaker 02:
So the database in ERI had relationships.

[00:16:06] Speaker 02:
That's on page 1326 of ERI.

[00:16:10] Speaker 02:
The database in SAP had correlation between different information stored in the database.

[00:16:14] Speaker 02:
That's page 1169 of the SAP America case that we cited.

[00:16:20] Speaker 02:
And recrative on page 3 includes a step of identifying relationships between information in the machine learning model.

[00:16:29] Speaker 02:
So nothing about having conceptual relationships between information makes a claim patent eligible.

[00:16:36] Speaker 02:
What's important is that there's no specific relationships claimed here.

[00:16:41] Speaker 02:
The general notion that information in a database is related doesn't improve computer technology.

[00:16:46] Speaker 02:
And there's no specific relationships that are in the claim.

[00:16:49] Speaker 03:
But as you know, they're saying what their patent does is allow a user to use natural language.

[00:16:55] Speaker 03:
and learn from it, and the model learns from itself.

[00:17:01] Speaker 03:
That's what this case-based information is.

[00:17:06] Speaker 03:
What's your argument as to why that is not patent eligible?

[00:17:12] Speaker 02:
So the idea of learning from natural language is something that humans have done for thousands of years.

[00:17:18] Speaker 02:
So it's conventional mental steps.

[00:17:20] Speaker 02:
That's one reason.

[00:17:21] Speaker 02:
It's conventional maybe for us.

[00:17:23] Speaker 03:
Is it conventional for computers?

[00:17:27] Speaker 02:
Well, it was conventional for computers, but our argument doesn't depend on that, Your Honor.

[00:17:32] Speaker 02:
The general idea of learning, having a computer learn, doesn't make something patent-eligible, even if... Well, that's AI, right?

[00:17:41] Speaker 02:
Yes.

[00:17:42] Speaker 02:
So AI is a field, and it's a field of trying to make computers learn, and that is an abstract idea.

[00:17:50] Speaker 02:
So your position is all AI is not patent-eligible?

[00:17:56] Speaker 02:
No, Your Honor.

[00:17:57] Speaker 02:
the general idea of having a computer learn is not patent eligible.

[00:18:02] Speaker 02:
And that's for several reasons.

[00:18:03] Speaker 02:
Because humans learn.

[00:18:06] Speaker 02:
Because if you don't say how to learn, it's functional.

[00:18:12] Speaker 02:
And this court held in recente, for example, that applying machine learning to a specific environment doesn't make a claim patent eligible.

[00:18:22] Speaker 02:
And so by the same logic, applying case-based reasoning

[00:18:27] Speaker 02:
to a specific field doesn't make a claim patent eligible.

[00:18:32] Speaker 03:
I want to make sure I didn't misunderstand your negative answer to my probably negative question.

[00:18:37] Speaker 03:
Your contention is some artificial intelligence could be patentable under current law.

[00:18:46] Speaker 03:
It's just that the patent in front of us is not.

[00:18:49] Speaker 03:
Is that your position?

[00:18:51] Speaker 02:
Yes, Your Honor.

[00:18:51] Speaker 02:
Some specific implementations of artificial intelligence

[00:18:55] Speaker 02:
could be patent eligible.

[00:18:55] Speaker 02:
And it says on the last page of the recrative opinion that the court's holding in that case that merely applying conventional machine learning to a computer environment wasn't patent eligible there.

[00:19:09] Speaker 02:
It doesn't mean that no machine learning is patent eligible.

[00:19:11] Speaker 02:
If you have a specific improvement to the way that the computer does it, then it could be patent eligible.

[00:19:17] Speaker 00:
And the idea of using machine learning in a particular environment, like natural language,

[00:19:25] Speaker 00:
under recidivism is not patent eligible.

[00:19:28] Speaker 00:
That's correct, Your Honor.

[00:19:32] Speaker 02:
And that's what we have here.

[00:19:33] Speaker 02:
The argument is that they were the first to apply case-based reasoning, which is a way of making a computer learn, to natural language.

[00:19:42] Speaker 02:
Even if that were true, it wouldn't be patent eligible, because that's merely confining an abstract idea to a particular field of use.

[00:19:52] Speaker 02:
And importantly here,

[00:19:54] Speaker 02:
or additionally, they admitted to the district court that case-based reasoning has been well known since the 1970s.

[00:20:01] Speaker 02:
So it's an admittedly conventional learning technique, and they're merely applying it to a particular field of use.

[00:20:10] Speaker 03:
And they tell us at least today that they think they invented a specific improvement, that they did come up with a particular technological solution to a technological problem or a technical problem with computers, and they have a specific improvement that had not been done before.

[00:20:30] Speaker 03:
How do I evaluate that?

[00:20:32] Speaker 03:
That sounds like a fact question to me.

[00:20:36] Speaker 02:
Well, you evaluate that by looking at the claims, Your Honor, and seeing that any purported improvement in the claims is, sorry, any purported improvement described in the specification is not in the claims.

[00:20:48] Speaker 02:
So what the specification describes as the purported improvement is this search and learn approach.

[00:20:55] Speaker 02:
And specifically, this is also what appellants rely on on page three of their reply brief when they're saying that they have a computer-specific solution.

[00:21:04] Speaker 02:
They refer to the patent's disclosure

[00:21:07] Speaker 02:
at column 6, and they say that's where the improvement is.

[00:21:13] Speaker 02:
And that, if you look at it, Your Honor, this is on Appendix 75, column 6, starting at lines 39.

[00:21:25] Speaker 02:
This is what they cite as the source of their purported improvement in the spec.

[00:21:29] Speaker 02:
It says, the strategy uses a concept referred to herein as search and learn.

[00:21:34] Speaker 02:
And then it goes and explains a little bit about the search and learn.

[00:21:38] Speaker 02:
There's further disclosure of search and learn at column 15, for example, that goes through the steps of search and learn.

[00:21:48] Speaker 02:
They pointed to this in the district court as well.

[00:21:50] Speaker 02:
And Judge Sanis explained at appendix pages 19 to 20 that

[00:21:58] Speaker 02:
This purportedly improved search and learn approach is not in the claims.

[00:22:03] Speaker 02:
So Judge Sanders said, bottom of appendix 19, citing the specification, plaintiffs exalt a learning mechanism that allows richer keywords and cases to provide more accurate performance, an exemplary search and learn system that can learn from the users in a way that improves both effectiveness and efficiency,

[00:22:27] Speaker 02:
She quotes their brief, which was quoting 798 patent, column 6, line 35 to 39.

[00:22:32] Speaker 02:
And then the district court said, but the claims include only oblique references to cases and case information as information to be searched or stored.

[00:22:42] Speaker 02:
So she concluded that the search and learn approach that provides the purported benefit in the specification isn't in the claims.

[00:22:49] Speaker 02:
All right, but if it were in the claims, would it make a difference?

[00:22:52] Speaker 02:
It depends the level of detail that was in the claims.

[00:22:57] Speaker 03:
So where have we said in a 101 case, it depends on the level of detail?

[00:23:04] Speaker 03:
That is, for purposes of this question, if search and learn is in the claims, where have we said that it would make a difference at either step one or step two, how much detail there was with that technological improvement allegedly being in the claims?

[00:23:19] Speaker 02:
Well, what I mean by the level of detail, Your Honor, is you have to say how it's done.

[00:23:23] Speaker 02:
So specific steps that implement

[00:23:26] Speaker 02:
the improvement are required, that's the distinction between claims that merely recite a desired result or a function.

[00:23:34] Speaker 02:
So if you merely recited the desired result of search and learn, which is reducing complexity, if you said providing a result in response to a natural language query with reduced complexity, that would be the function.

[00:23:48] Speaker 02:
And that wouldn't be sufficient detail.

[00:23:50] Speaker 02:
If you said how you do it using the steps of the search and learn approach, then we would have a question of, OK, are those steps conventional or not?

[00:24:00] Speaker 02:
But they haven't even recited those steps, and so they don't have a purported improvement in the claim.

[00:24:06] Speaker 02:
And that's what the district court held at appendix pages 19 to 20, and that holding was correct, Your Honor.

[00:24:23] Speaker 02:
So eerie, SAP, and rescintive are dispositive here, Your Honor.

[00:24:28] Speaker 02:
They show that a database with new information content doesn't make claims patent eligible.

[00:24:37] Speaker 02:
They show that merely applying machine learning generically to a new field of use doesn't make claims patent eligible.

[00:24:51] Speaker 02:
Having interconnected data in your database doesn't make claims patent eligible.

[00:24:57] Speaker 02:
They show that identifying combinations of information in the database doesn't make claims patent eligible.

[00:25:05] Speaker 02:
There's just no way for the claims here to overcome ERIE, SAP, and Resentive.

[00:25:15] Speaker 03:
Is it correct, I think, that in Resentive, the patent owner there did not claim

[00:25:21] Speaker 03:
to be improving how computers function?

[00:25:25] Speaker 03:
At least they conceded certain things that evidently are not conceded here.

[00:25:29] Speaker 03:
Do you agree with that?

[00:25:30] Speaker 03:
And if so, why would that not be a meaningful distinction?

[00:25:34] Speaker 02:
Well, I think the language that Mr. Zhu is pointing to in recitative is about whether there's an improvement in the claims or not.

[00:25:48] Speaker 02:
So that's a legal question.

[00:25:49] Speaker 02:
Your honors don't need

[00:25:51] Speaker 02:
the other side to concede that an improvement is not in the claims, to look at the claims and see that there's no improvement to computer technology there.

[00:26:04] Speaker 02:
And many times this court has held that claims were directed to an abstract idea at step one, even where the appellant argued that the claims were cited an improvement.

[00:26:18] Speaker 02:
This court has often held, for example,

[00:26:21] Speaker 02:
If your improvement is merely identifying information by content or source to go in your database, that that's, as a matter of law, is not enough.

[00:26:31] Speaker 02:
And that's what we have here.

[00:26:32] Speaker 02:
There's four categories of information in the database.

[00:26:36] Speaker 02:
Appellants are arguing that that makes their claims patent-eligible, that that's part of their improvement.

[00:26:41] Speaker 02:
But BSG, for example, at page 1288,

[00:26:46] Speaker 02:
explains that an improvement to the information stored in a database is not an improvement in database technology.

[00:26:54] Speaker 03:
Can I ask you, you wrote on page 34 of your red brief, because learning to understand language is a human mental activity, it is abstract as a matter of law.

[00:27:07] Speaker 03:
You didn't have recintive yet at the time you wrote the brief, so you cited Trinity Info Media and personal web text.

[00:27:14] Speaker 03:
I'm not sure that those stand for such a broad proposition.

[00:27:17] Speaker 03:
Do you see recintive as standing for that broad proposition, that learning to understand language?

[00:27:23] Speaker 03:
is abstract as a matter of law, and in particular because it's a human mental activity?

[00:27:34] Speaker 02:
In general, so one of Vercentiv's rationales for holding that the machine learning there was abstract is because it was analogous to what humans did before.

[00:27:44] Speaker 02:
And so I think it's informative, but as we discussed earlier, it depends on whether there's

[00:27:52] Speaker 02:
what I refer to it as a level of detail.

[00:27:54] Speaker 02:
But what I meant was if the how is in the claim.

[00:27:58] Speaker 02:
So you need a how that distinguishes the claim from human activity.

[00:28:03] Speaker 02:
And what Mr. Zhu has pointed to is case information.

[00:28:08] Speaker 02:
And all case information requires, even under the district court's interpretation and claim construction, is case-based reasoning generically.

[00:28:17] Speaker 02:
So if it's generic learning, then

[00:28:20] Speaker 02:
Yes, that is abstract, because humans learn, and it cannot supply a technological solution.

[00:28:33] Speaker 00:
OK, thank you.

[00:28:35] Speaker 00:
Mr. Zhu.

[00:28:43] Speaker 01:
Logan, your honors.

[00:28:46] Speaker 01:
Humans learn, computers compute.

[00:28:49] Speaker 01:
Now, we heard a lot about the recintive decision from arguing counsel on the other side.

[00:28:54] Speaker 01:
And they mentioned that their recintive was talking about the invention being devoid from the claims.

[00:29:00] Speaker 01:
But specifically, the court's decision at 1213 talks about how both the claims and the specification did not describe how any improvement could be accomplished.

[00:29:11] Speaker 01:
And that factual record and the factual record below here matters, Your Honor.

[00:29:17] Speaker 01:
Judge Stark asked a question about what are the material fact disputes here?

[00:29:22] Speaker 01:
And I guess our fact dispute is that the court below ignored the material facts and they did they did this by ignoring the stipulation of the parties which material specifically your honor the fact that

[00:29:37] Speaker 01:
A person could not conduct the claimed invention in their mind.

[00:29:41] Speaker 01:
That's at appendix 3551 to 52, appendix 3233.

[00:29:45] Speaker 00:
OK, let's assume that that's true.

[00:29:47] Speaker 00:
But is there any other material fact that the district court ignored?

[00:29:52] Speaker 01:
There is, Your Honor.

[00:29:53] Speaker 01:
Just to round out the record, there's also appendix 4069 to 70.

[00:29:57] Speaker 01:
But the other material fact was whether case-based reasoning was meaningfully reflected in the claims.

[00:30:04] Speaker 01:
Now both experts agreed to this in their expert report.

[00:30:07] Speaker 00:
Well, the concept of case-based reasoning may be in the claims, but that mere concept of case-based reasoning doesn't render it non-abstract.

[00:30:15] Speaker 01:
The concept itself does not, Your Honor, but the fact that the district court ignored that case-based reasoning was at all reflected in the claims does matter.

[00:30:25] Speaker 01:
And that's stipulated at 3229 and at Appendix 3156.

[00:30:29] Speaker 01:
I also want to direct the court to specifically the specification at Column 7

[00:30:35] Speaker 01:
and columns 16 through 17 below, which was discussed extensively in our briefing, both at the district court level and in our briefs here, where it discussed the search and learn in the claims, how to implement that with specific computational algorithms, and how to enact the determination and the implementation in the claims.

[00:30:56] Speaker 01:
What does search and learn mean?

[00:30:56] Speaker 01:
Sorry, Your Honor?

[00:30:57] Speaker 00:
What does search and learn mean?

[00:31:00] Speaker 01:
searching a database, using the information within the database, and within this context, the four different types of metadata categories and the connections between those metadata, and then conducting a constrained search, conducting case-based reasoning, and adding new cases to the reference dictionary.

[00:31:19] Speaker 00:
OK.

[00:31:19] Speaker 00:
I think we're out of time.

[00:31:20] Speaker 00:
Thank you.

[00:31:20] Speaker 01:
Thank you, Your Honors.

[00:31:22] Speaker 00:
May I go to counsel?

[00:31:22] Speaker 00:
The case is submitted.