Latest / The Adults In The AI Room / Incentivizing Technology - Juan Sequeda - Earley AI Podcast with Seth Earley & Chris Featherstone - Episode # 024

Transcript

Incentivizing Technology - Juan Sequeda - Earley AI Podcast with Seth Earley & Chris Featherstone - Episode # 024 · The Adults In The AI Room

0:07
SPEAKER_00: Here we talk with top leaders and practitioners on what's possible and practical with AI and the enterprise as we move from the current state to the future of cognitive assistants and an end of AI in all aspects of our personal and professional lives.
0:22
SPEAKER_00: Thanks so much for listening.
0:23
SPEAKER_00: If this is your first time, we invite you to subscribe wherever you listen to podcasts.
0:28
SPEAKER_00: And now here are your hosts, Seth Early and Chris Featherstone.
0:36
SPEAKER_02: Good morning, good afternoon, good evening, depending on your time zone.
0:39
SPEAKER_02: Um welcome to today's podcast.
0:41
SPEAKER_02: I'm Seth Early.
0:43
SPEAKER_02: And I'm Chris Featherstone.
0:44
SPEAKER_02: It's good to be with you, Seth.
0:45
SPEAKER_02: Yeah, nice to see you.
0:47
SPEAKER_02: Um I uh I think uh uh winter is certainly setting uh up here in the northeast.
0:53
SPEAKER_02: I know you have what 10, 12, 15 feet of snow now.
0:57
SPEAKER_01: Actually, all of these crazy storms that are blasting, you know, Middle Earth or you know, Middle America kind of around us, you know.
1:06
SPEAKER_01: So yeah.
1:06
SPEAKER_01: So while while eastern Montana and the rest of you know, all the way over to you are getting blasted, we uh we just got a few inches, is all.
1:15
SPEAKER_02: No, that's not too bad.
1:16
SPEAKER_02: Well, let's jump in.
1:17
SPEAKER_02: Our guest today has dedicated himself to reliably creating knowledge from inscrutable data and industry work has been on designing and building knowledge graphs for enterprise data and metadata management.
1:29
SPEAKER_02: When he's not writing books, papers, or serving on boards and committees, he's on the salsa dance floor.
1:35
SPEAKER_02: Please welcome the principal scientist at data.world, Juan Cicada.
1:40
SPEAKER_05: Hello, Seth.
1:41
SPEAKER_05: Hello, Chris.
1:42
SPEAKER_05: Uh, thank you very much.
1:43
SPEAKER_05: How are you guys doing?
1:44
SPEAKER_05: We're doing great.
1:45
SPEAKER_05: So, first of all, I want to congratulate you on the new baby.
1:49
SPEAKER_05: Thank you very much.
1:49
SPEAKER_05: Yeah, it's uh new adventure, uh some something more more knowledge being produced in a graphic.
1:56
SPEAKER_05: Is this this is your first?
1:58
SPEAKER_05: Yes.
1:58
SPEAKER_05: Number one, yeah.
1:59
SPEAKER_05: Number one.
1:59
SPEAKER_05: Just what's what's uh what's the baby's name?
2:02
SPEAKER_05: A B girl Sophia, actually.
2:04
SPEAKER_05: Right now it's uh uh two months old.
2:06
SPEAKER_02: So two months, wow.
2:07
SPEAKER_05: Wow fast and and so much fun.
2:10
SPEAKER_01: So yeah, so the borough part though, Juan, is that babies, new babies don't come with knowledge graphs, so you can't do any of the governance or discovery that your normal, you know, well what's interesting is my well, my wife and I were both scientists, so we we are very diligent on all the data that we're taking about everything.
2:29
SPEAKER_05: Uh we we have uh a friend of ours uh gave us a book on uh experiments you can do on your baby.
2:35
SPEAKER_05: So we're just having so much fun seeing how that knowledge is being uh created little by little.
2:42
SPEAKER_02: So you're not gonna put her in a skin skinner box.
2:44
SPEAKER_02: Is that the BF Skinner?
2:46
SPEAKER_02: Remember, he did something.
2:47
SPEAKER_05: Actually, my wife or her PhD is in a behavior analysis.
2:51
SPEAKER_05: So she is, I have a lot of BF Skinner books and pictures and everything in my house.
2:56
SPEAKER_05: So no comment I'll say for now.
3:02
SPEAKER_02: So give us give us a background, tell us uh about the world according to Juan, you know, your background, you know, how you kind of got into the space and uh just give us uh the lay of the land in terms of where you're from and how you uh how your education kind of evolved and and how you got to where you are.
3:18
SPEAKER_05: Yeah, and so um if I go start, I um I I got into I did my undergrad in my then eventually my PhD in computer science.
3:28
SPEAKER_05: Um and so I've always been interested.
3:31
SPEAKER_05: I got interested early on, I think around 2004, 2005, I was exposed uh to this thing called the semantic web.
3:40
SPEAKER_05: And it was actually I was living in Colombia, my parents are Colombian, and I've lived in Colombia like from the ages of 10 to 20.
3:46
SPEAKER_05: Um, and a professor came by and gave a seminar one day, and I just walked into the room, and that professor is Oscar Korch, who's a professor in Madrid in the in UPM in Madrid, and just introducing this whole concept of the semantic web, and I'll never forget that the the example was um say you say Paris Hilton.
4:04
SPEAKER_05: What do you mean by Paris Hilton?
4:06
SPEAKER_05: Do you mean the person or you mean the hotel in a city Paris and which city Paris and so forth?
4:11
SPEAKER_05: And that for me was eye-opening.
4:13
SPEAKER_05: I mean, I don't know, I was I was young, I was nine, nineteen years old, eighteen years old, whatever.
4:18
SPEAKER_05: Uh, and it's like, oh, there's this thing about semantics, about meaning.
4:21
SPEAKER_05: So very early on at that time, I got into like, hey, how do you describe knowledge and learn about ontologies and using tools back in 2004, 2005 of protege and stuff like that?
4:32
SPEAKER_05: Um, I ended up uh going to to UT.
4:35
SPEAKER_05: I I transferred my my parents are academics too, so they they they they moved, they they go back between Columbia and the US.
4:41
SPEAKER_05: So I ended up at UT around 2006 as a doing my undergrad.
4:46
SPEAKER_05: Uh, and then I I I met uh a professor who became my advisor, uh Dan Maranker, and he was getting under learning a lot about all this semantic web, all these standards that were coming out.
4:57
SPEAKER_05: And but he's a database guy.
4:59
SPEAKER_05: Uh, and he was like, Well, what is this relationship between relational databases and all this semantic web stuff that I'm hearing about OWL and RDF and graphs?
5:08
SPEAKER_05: And that was around 2006.
5:10
SPEAKER_05: And that question, that very simple question, what's the relationship between relational databases and these semantic web technologies?
5:15
SPEAKER_05: That question changed my life.
5:17
SPEAKER_05: Uh, and that has been basically the the the quest that I've been on and trying to not just understand it, but to kind of take it to the next level.
5:24
SPEAKER_05: I think the the the first talk I ever gave in my life, uh, was it was in 2007.
5:29
SPEAKER_05: It was uh giving a talk on kind of our first kind of thoughts about integrating relational databases with all these semantic web technologies.
5:35
SPEAKER_05: And that was at the a W3C meeting at MIT and front row was Tim Berners-Lee.
5:40
SPEAKER_05: And I got I got I met Tim, I met a lot of the kind of the early kind of pioneers in the semantic web space, and and since then I've been able to kind of just keep in contact and build my my network and and ended up uh working in research here.
5:53
SPEAKER_05: So I I decided I I've always liked startups, and I had another startup at that time, but I was starting to go do research.
5:59
SPEAKER_05: Uh, and for me, it's like, well, I could I really like this topic, I really like Austin, I really like UT, I really like my advisor.
6:06
SPEAKER_05: And I said, well, to keep working on that, I decided to was to keep working on that, keep doing do a PhD.
6:12
SPEAKER_05: So I did my PhD at UT.
6:14
SPEAKER_05: And and the title of my PhD certation was literally is literally integrating relational databases with the semantic web.
6:19
SPEAKER_05: So a lot of the work that we did there was about um understanding um defining the mappings uh of how to map relational to RDF.
6:27
SPEAKER_05: So the standards that came out in 2012 are are based on a lot of the work that we did.
6:31
SPEAKER_05: We did a lot of the query engines to do query translation and so forth.
6:35
SPEAKER_05: But all of this was to say is we got all this data, which I use the you the use the word inscrutable because it's like so complicated.
6:43
SPEAKER_05: It's like what does this mean?
6:44
SPEAKER_05: Like if you look if you look at a uh a real enterprise database, you look at an SAP system, right?
6:51
SPEAKER_05: Uh uh an Oracle EBS system, you're talking about database schemas that have thousands, thousands of tables and tens of thousands of columns that the namings are inscrutable, like you don't know what this stuff means.
7:06
SPEAKER_05: And but there is meaning in there, and that meaning, and what is where is that meaning, and how do I expose that meaning?
7:12
SPEAKER_05: So kind of uh from there, we we take our our research and we said, hey, if if all this semantic stuff is gonna take off uh with these standards, people are gonna be saying, but my data is in Oracle, it is in SQL Server, and it's really complicated.
7:26
SPEAKER_05: Uh, how do I uh put this stuff together?
7:29
SPEAKER_05: Well, we had we had our approach and we had our technologies, our systems, and all that stuff.
7:33
SPEAKER_05: And we said, Hey, I guess our hypothesis that people were gonna be knocking on our door.
7:37
SPEAKER_05: And effectively, people did start knocking on our door.
7:39
SPEAKER_05: I think around 2012, we were starting seeing this.
7:41
SPEAKER_05: And we started a company called CapCenta around 2014, which was about commercializing all this technology and um yeah, creating what we now call semantic layers, what is now being called the knowledge graphs and stuff.
7:52
SPEAKER_05: So we're doing that for working in in oil and gas, uh, working in pharma and finance, right?
7:58
SPEAKER_05: And e-commerce companies who who were understanding this vision of being able to create your ontologies, creating your knowledge and map that to that inscrutable uh data.
8:07
SPEAKER_05: Uh and one of my one of my customers uh actually was data.world and uh data.world was using the IP, wanted the IP and stuff, and just realized that we were saying on the same mission and vision, and uh we joined some forces and I uh sold the company to Data.world like three years ago and keep working on the same vision.
8:24
SPEAKER_05: Here we are.
8:25
SPEAKER_05: Very I lived long, but yeah, that's a summary of like last 15 plus years.
8:30
SPEAKER_01: So one uh, you know, fast forwarding to um, which thanks for all that background, it's awesome because uh, you know, went through a lot of the similar scenarios, as you can imagine, from a commercial product perspective and building the products, you know.
8:44
SPEAKER_01: And um I'd love to get a take on on, you know, now that you're in the middle of a lot of these scenarios with customers and stuff, you know, it does I don't even think it matters in the size.
8:58
SPEAKER_01: What do you find is there is the biggest barrier to entry for them?
9:01
SPEAKER_05: One of my things is like incentives, right?
9:03
SPEAKER_05: It goes back to like talking about behaviors.
9:05
SPEAKER_05: Yeah.
9:05
SPEAKER_05: Um, what do people really want?
9:07
SPEAKER_05: I think what what the market is, I mean, to be very honest here, I mean, honest no BS is the brand that I have all the time, is I think the market is pretty immature on when it comes to like what they want, what or what they think they want, right?
9:18
SPEAKER_05: I think um what do the what does the market want?
9:21
SPEAKER_05: I need to go find data.
9:23
SPEAKER_05: And that's why a data catalog exists, right?
9:25
SPEAKER_05: And that's why we're we're a data we're a data catalog, because the majority of the use cases, I don't know what data I have.
9:30
SPEAKER_05: I need to go find my data.
9:31
SPEAKER_05: Um, so I think that that's where everybody is.
9:34
SPEAKER_05: That's one thing.
9:34
SPEAKER_05: Second, I think from a from uh from the perspective of the the data management world, we have focused so much on just technology.
9:44
SPEAKER_05: The problems that we've been solving 30 years ago continue to be the same problems we're trying to solve today.
9:49
SPEAKER_05: 30 years have passed, and the same problem we're solving today, something is wrong.
9:55
SPEAKER_05: And if we don't if we don't acknowledge something is wrong, I'm gonna, I'm sorry, you're part of the problem.
10:01
SPEAKER_05: So so what is so if we observe let's observe what's happened in the last 30 years, my observation is that the focus has always been just on technology.
10:11
SPEAKER_05: And we therefore our success metrics are just technology.
10:16
SPEAKER_05: We incentivize just for technology, and uh and and so what does success look like?
10:22
SPEAKER_05: Is it I created a data warehouse, a data lake, whatever?
10:24
SPEAKER_05: Well, that's how we've defined success and that's how we get paid, right?
10:28
SPEAKER_05: But success that's not what success means.
10:30
SPEAKER_05: Success is like, oh, wait, there's somebody in the business who is trying to solve a problem, whatever, because what is their goal?
10:36
SPEAKER_05: To make money and save money.
10:38
SPEAKER_05: That's it.
10:39
SPEAKER_05: Were they able to make money and save money because the of the technology work that you did?
10:44
SPEAKER_05: If you did all this stuff for the in the from a technology sake, but they still weren't able to go solve their question to make money, save money.
10:50
SPEAKER_05: I'm sorry, you were not successful.
10:52
SPEAKER_05: You failed, but we're not celebrating, we're not incentivizing that, we're incentivizing the technology.
10:56
SPEAKER_05: So, what has happened over the last uh 30 plus years is that we've only focused on the technology.
11:03
SPEAKER_05: And I think the shift that is is starting to occur, I would say.
11:08
SPEAKER_05: I'm hopeful here, uh optimistic in the sense, but the shift that we need to go have is having a social technical uh paradigm shift, which is we look at data management not just from a technical perspective, but from a social perspective.
11:23
SPEAKER_05: What does that mean?
11:24
SPEAKER_05: People and process, right?
11:26
SPEAKER_05: We just need to understand the people process.
11:28
SPEAKER_05: And the issue here is that as technologists, you don't like to talk to people and processes because of you.
11:33
SPEAKER_05: It's complicated.
11:34
SPEAKER_05: I just give you my technology, and we're incentivized to focus on our technology.
11:37
SPEAKER_05: So I think that that that that's what needs to change, and that's and that's been just the barrier to entry is that we focus too much on the technology and um not understanding the people.
11:46
SPEAKER_01: And I always I always find it you know, as defining irony as you're allowing your tactical operational direction dictate your strategic business objectives, right?
12:01
SPEAKER_05: It's just it's going like all bottoms up completely, right?
12:03
SPEAKER_05: Exactly.
12:04
SPEAKER_05: And it's we have to have it, it's always the it the answer to most questions is it depends.
12:10
SPEAKER_05: It's hybrid bottom up and top down.
12:12
SPEAKER_05: That's what it is.
12:12
SPEAKER_05: It's an right, because but all for a lot of these things, it is it it's not it depends, it's oh, it's technology first, and then it's always bottoms up broken.
12:21
SPEAKER_02: Technology-centric in terms of the thought process as opposed to the business process and the business objective, and that's kind of the nature of uh when there's emerging technologies or newer technologies, the executives don't understand it, they're gonna pass it on to their technical team.
12:36
SPEAKER_02: The technical team is gonna look at it through a certain lens.
12:39
SPEAKER_02: And we were with a recent client and we're building out you know ontologies and knowledge graphs.
12:44
SPEAKER_02: And I said, Well, what are the use cases?
12:46
SPEAKER_02: And they go, Well, we really don't care about the use cases, we're just a huge red flag, and that's a sign for failure.
12:52
SPEAKER_02: I know, isn't that crazy?
12:54
SPEAKER_02: And it's like, you know, what's your what's your business case?
12:56
SPEAKER_02: Well, it's to use the ontology.
12:58
SPEAKER_02: It's like, no, that's not a business case, right?
13:00
SPEAKER_02: So it really does it, and and we found one uh organization that spent a half a million dollars on knowledge graphs, and they did not have an outcome, they did not have an output.
13:11
SPEAKER_02: It was the science experiment, but they said we just had to get started.
13:14
SPEAKER_02: Okay, well, if you have that kind of budget and that kind of money to throw around to get started and do a science project, that's one thing.
13:21
SPEAKER_02: But I'd like to see real business value.
13:23
SPEAKER_02: Where are people seeing the low-hanging fruit in terms of the business value?
13:27
SPEAKER_02: Is it really a reduction in IT spend?
13:30
SPEAKER_02: Is it efficiencies and the on on the technology side, or is it really make having that impact on you know those business processes and doing things like personalization or doing things like you know, what what what are what are you seeing as kind of the common low-hanging fruit?
13:48
SPEAKER_02: And I imagine that depends on industry and sector and and so forth.
13:51
SPEAKER_05: I guess you answer your question yourself.
13:53
SPEAKER_05: It depends.
13:54
SPEAKER_05: I think a good a good strat a good kind of tactic I've I've seen around is like understand where you're if you're in the IT to the tech side, where do you report to, right?
14:04
SPEAKER_05: So if you report to the end up to the CFO, then you're considered a cost center.
14:08
SPEAKER_05: So how are you gonna provide value?
14:09
SPEAKER_05: It's figuring how to reduce costs, right?
14:11
SPEAKER_05: So if you report end up to the COO, right, then that's kind of more efficiency.
14:15
SPEAKER_05: So that's more about productivity.
14:17
SPEAKER_05: So then talk about productivity language.
14:19
SPEAKER_05: If you end up uh in a digital transformation office, whatever, right?
14:24
SPEAKER_05: That uh then that did then you're trying to about how I need to go migrate to the cloud as fast as possible, right?
14:28
SPEAKER_05: I need to go right though, those those types of things you go do.
14:32
SPEAKER_05: And I want to get people if you want to go do data democratization, all that stuff.
14:36
SPEAKER_05: Well, then have the have have the metrics to show how people are actually learning more about they're getting on board on data and so forth.
14:43
SPEAKER_05: If you're then if you end up reporting directly to the CEO, uh through the C up to the CEO chain, then you're about strategy innovation.
14:50
SPEAKER_05: And then it's like, okay, how are we making more money with this data that we're doing?
14:54
SPEAKER_05: So I think what I would answer the question, I would answer the question is under is figure out or understand where you report to, and then you speak that language, right?
15:04
SPEAKER_02: Right.
15:04
SPEAKER_02: So if you're reporting up to a CMO, then it's going to be about you know customer identity and it's going to be about first party data, and it's going to be about personalization, contextualization, dynamic product catalogs, et cetera.
15:16
SPEAKER_02: So yeah, totally, totally get that.
15:17
SPEAKER_02: That's a great way to look at it.
15:18
SPEAKER_02: You know, rather than, I mean, you really have to get to that finer degree of granularity in an organization because you can say, well, this industry may be doing this, but in your organization, I think it's a great way to say where you're reporting up to, right?
15:32
SPEAKER_02: Because that's the metric that's they're going to care about.
15:34
SPEAKER_05: That's a great uh and and another thing I want to add there is we we we all talk about uh data literacy for so long, right?
15:41
SPEAKER_05: Now, and the the the the drum I'm banging so hard right now is we need to have this this shift on what I'm calling now business literacy.
15:50
SPEAKER_05: So we're telling all the business people you need to be more data literate, which actually I kind of don't that doesn't even sound good because you know what you're telling them.
15:58
SPEAKER_05: You're like, oh, you are illiterate in data.
16:01
SPEAKER_05: It sounds kind of it's offensive, right?
16:03
SPEAKER_05: Guess what?
16:04
SPEAKER_05: Those people know so much stuff of the things that you don't know from the data.
16:07
SPEAKER_05: Like, how about you learn about those things?
16:09
SPEAKER_05: You learn about how the marketing works.
16:12
SPEAKER_05: And I always tell people you need to understand how the business, how the organization that you work in makes money.
16:16
SPEAKER_05: Where do you pour money in?
16:17
SPEAKER_05: What does that flow through and how does that generate more money?
16:19
SPEAKER_05: And then what that's one second, understand what are the operational goals and what are the strategic goals of the company.
16:25
SPEAKER_05: You're in you report to what department?
16:27
SPEAKER_05: Okay, what are the OKRs in those departments?
16:29
SPEAKER_05: What are the KPIs?
16:30
SPEAKER_05: How do you know that the work that you're doing is contributing to that?
16:34
SPEAKER_05: So I think so.
16:35
SPEAKER_05: We just lack a lot of kind of curiosity.
16:37
SPEAKER_05: Uh, and I think that that's a lot of stuff that we need to go push more.
16:40
SPEAKER_05: So actually, the low-hanging fruit here for is tell everybody who's listening is like, you do you know what the strategic goals of your companies are?
16:48
SPEAKER_05: Do you know the department that you're in, what those OKRs are and how you and and how they contribute to the strategic goals?
16:53
SPEAKER_05: Do you know what that work that you did today, how that is contributing to that?
16:57
SPEAKER_05: If you don't know, if that's unclear, go figure that out.
17:02
SPEAKER_01: That's a little hanging through.
17:03
SPEAKER_01: That's a huge leadership gap, though, too, in in what you're describing, because we allow so many times for those below the budget line to um you know to keep them because it's an it's either an insecurity or it's an information gateway, or it's your you know, your classic IT, like, hey, listen, you know, don't change what I've put in place because, you know, this is my world and I don't like change.
17:28
SPEAKER_01: So therefore, listen, I'm gonna stay in my bucket and I'm gonna create all this kind of stuff and be a gateway to either enablement or disablement.
17:37
SPEAKER_01: And then the business doesn't know how to speak that speak.
17:41
SPEAKER_01: And what you have is this this huge, you know, above the budget line, below the budget line discrepancy between what the strategic objectives are and what actually needs to get done.
17:52
SPEAKER_01: And they need to come together to say, listen, let's do a working backwards session.
17:55
SPEAKER_01: Let's drive what the business outcome is, going back to the previous thought, and let's map all this out because guess what?
18:02
SPEAKER_01: You know, um, IT data get I don't, you know, for all purposes, care what you think about this, that, and the other.
18:09
SPEAKER_01: What I need is to get that information out so I can actually make insightful decisions and automation on this other site to actually drive a customer experience, to drive efficiency, or to drive revenue making opportunities with all the stuff that you're doing.
18:24
SPEAKER_01: Otherwise, you know, who cares?
18:26
SPEAKER_01: You're just doing it for the sake of doing it.
18:27
SPEAKER_02: Yeah, I think it is a great, a great objective to encourage the technology organization to understand that role.
18:35
SPEAKER_02: You know, a lot of times when we do metrics programs, we talk about the data at the lowest level, but data is supporting some process.
18:42
SPEAKER_02: That process has has KPIs, that process is in support of a strategic initiative or some type of business outcome.
18:48
SPEAKER_02: And that business outcome is supporting the the strategic uh direction of the organization.
18:53
SPEAKER_02: We need to have that linkage, but it's so super important for people who are doing those day-to-day jobs to understand that linkage, as well as the executives to understand that linkage.
19:03
SPEAKER_02: But I agree it's a critical missing piece.
19:05
SPEAKER_02: Not changed topics much, but I do want to hear uh more about your book.
19:09
SPEAKER_02: Like what caused you to write the book?
19:11
SPEAKER_02: Um, how long did it take?
19:13
SPEAKER_02: And tell tell us a little bit about it and who it who it's for and and what it uh covers.
19:17
SPEAKER_05: Yeah, great.
19:18
SPEAKER_05: So so the book.
19:19
SPEAKER_05: So, first of all, I wrote the book in conjunction with uh Aura uh Aura LaCilla.
19:23
SPEAKER_05: So Aura is is one of the the original authors of the Semantic Web uh article in Scientific American, and he's like one of the authors of the original RDS specification 1998.
19:33
SPEAKER_05: So I've known over for a long time.
19:34
SPEAKER_05: And and for me, the this book has been about how how you actually build knowledge graphs from relational databases, but then it's gonna be abstracted from for any types of sources from structured sources.
19:46
SPEAKER_05: And what I what I've always struggled seeing people getting kind of excited about knowledge graphs is that they think it's a oh, I just dumped turned data into a graph and then it's all there.
19:55
SPEAKER_05: And I'm like, wait, no, no, there's all this issue about the knowledge, like understanding what this stuff means.
20:00
SPEAKER_05: And even though this is specific to knowledge graphs, and in the book, we don't make it specific to RDF.
20:07
SPEAKER_05: We actually are very clear that this is for any type of graph.
20:10
SPEAKER_05: I mean, we are the notation that we use is not specific RDF.
20:14
SPEAKER_05: We show a very generic notation where you can use it for a property graph or an RDF graph, whatever.
20:19
SPEAKER_05: So for me, just a quick parentheses the discussion of RDF property graph is a waste of time.
20:24
SPEAKER_05: Stop it.
20:24
SPEAKER_05: If you're making that discussion, you're doing a service to the industry, and I want you to shut up.
20:28
SPEAKER_05: All right, before this is ended.
20:30
SPEAKER_02: So um dig into that a little bit more.
20:32
SPEAKER_05: Tell us tell us the flaw with that because it's it's it's a technology problem, and I don't care about technology.
20:37
SPEAKER_05: Let's understand what the problem they're gonna solve, right?
20:39
SPEAKER_05: And it's a thick, and yes, there are just technology discrepancies, but they're gonna be merged together.
20:43
SPEAKER_05: It's inevitable, it's happening right now.
20:45
SPEAKER_05: Neo folks at Neo4j are fantastic, and they're have ways of incorporating RDF and they're making it seamlessly.
20:51
SPEAKER_05: The Amazon Neptune team are making some really cool things with their one graph.
20:54
SPEAKER_05: It's gonna happen, it's all gonna be merged, all these standards coming together.
20:57
SPEAKER_05: It just takes time, right?
20:58
SPEAKER_05: The the market is maturing and we're getting there.
21:01
SPEAKER_05: So it's just in a couple of years' time, it's just a moot discussion.
21:03
SPEAKER_05: We're gonna be laughing about that on how much time we wasted on it.
21:07
SPEAKER_05: I'm exactly we're not wasting the time.
21:09
SPEAKER_05: Every discussion is we're learning more about it, but we're converging.
21:12
SPEAKER_05: So, but the point here is the I I define kind of what is the knowledge graph.
21:17
SPEAKER_05: Like, I have my my clear definition you're representing real-world concepts and the relationships between those real world concepts, and they happen to form a graph, right?
21:27
SPEAKER_05: Nodes and edges.
21:28
SPEAKER_05: By the way, a really well third normal form database is also a graph, too, right?
21:32
SPEAKER_05: Every table is is is a node, is a concept, and and you have foreign keys, which can be relationships.
21:38
SPEAKER_05: Some relationships can become many to many.
21:40
SPEAKER_05: So you can't a table can represent itself as a relationship, but effectively you're representing a graph there.
21:45
SPEAKER_05: Now, the reason why the graph model itself is really valuable is because it's very nimble and you can now integrate data coming from so many diverse sources.
21:53
SPEAKER_05: Let it be relational sources, let it be tab uh other types of tabular CSV spreadsheets.
22:00
SPEAKER_05: Resources, XML, JSON, you can put the heck you do NLP over text, the output of that is triple.
22:05
SPEAKER_05: So graph is just as big, this great medium.
22:07
SPEAKER_05: I think one of the issues is that we believe now, oh, just because of graph, we we're gonna put tables away and we're gonna move to graphs.
22:12
SPEAKER_05: Like, no, graph is just that intermediate layer, and then you'll have a graph view over that, you'll have a Tabler view over that, you'll have a an API view, a GraphQL, whatever you want.
22:22
SPEAKER_05: That graph is an intermediate layer to be able to go integrate things.
22:25
SPEAKER_05: Users, you go talk to the end users, you end like tell them what does this stuff mean?
22:30
SPEAKER_05: Let's go describe our business, right?
22:32
SPEAKER_05: Well, you have a customer, a customer places an order, an order has a bunch of order lines, and you draw that on the whiteboard, it's a graph.
22:39
SPEAKER_05: So that's why the graph comes in here.
22:41
SPEAKER_05: Now, um, it's all about integrating data and knowledge at at scale.
22:46
SPEAKER_05: That's so that's that that's what it is.
22:48
SPEAKER_05: Now, when once you start creating that knowledge layer, I'm gonna use this all interchangeably, the knowledge layer, the ontology, the semantic layer, the the the knowledge graph schema.
22:58
SPEAKER_05: It's all the same thing, right?
22:59
SPEAKER_05: It it's it's that layer that you'll you show this to the end user, a business user, and they'll put it on the whiteboard, they're gonna look at that and they're like, Yeah, that makes sense to me completely.
23:11
SPEAKER_05: I those words make sense, those relationships make sense.
23:14
SPEAKER_05: They may have they may disagree with something.
23:16
SPEAKER_05: And why do they disagree?
23:17
SPEAKER_05: Because they understood what they saw and they said, No, that's incorrect, right?
23:21
SPEAKER_05: You're having that discussion right there at that strategic level, right?
23:25
SPEAKER_02: Right.
23:26
SPEAKER_02: You don't want silence, you want questions, you want pushback, you want people to say, you know, yeah, that doesn't make sense, but they understand the framework.
23:33
SPEAKER_05: And for this reason, right?
23:34
SPEAKER_05: For that reason.
23:35
SPEAKER_05: So then, so so understanding how to define that ontology semantic layer, right?
23:41
SPEAKER_05: All these things are the same.
23:42
SPEAKER_05: That's one thing.
23:43
SPEAKER_05: And then you have your inscrutable data, right?
23:45
SPEAKER_05: All your different resources.
23:47
SPEAKER_05: So that's your source.
23:48
SPEAKER_05: And then what I need to do is I need to ground the semantics, ground the meaning of those bubbles and lines, because people are saying this or saying words.
23:56
SPEAKER_05: I'm like, okay, so show me the data.
23:59
SPEAKER_05: I'm gonna say, well, that thing, where does that what does that thing actually mean?
24:03
SPEAKER_05: I need to go connect it to the source of the data.
24:06
SPEAKER_05: And that connection is what we call the mappings.
24:09
SPEAKER_05: So this book is all about how to start.
24:11
SPEAKER_05: I mean, first of all, understand that process and how to go create this, how to design them, how to go, how am I going to design these, the semantic layer, the ontology, and so forth?
24:20
SPEAKER_05: How am I gonna design the mappings?
24:23
SPEAKER_05: And uh, so we go through a bunch of mapping patterns because they're these are just things that you see very repeatable.
24:28
SPEAKER_05: You see the structure in the source, you have the structure in the target.
24:31
SPEAKER_05: How do I create these connections?
24:33
SPEAKER_05: But then also importantly is the process.
24:35
SPEAKER_05: It is how do you avoid boiling the ocean?
24:37
SPEAKER_05: Because this is the problem I've seen continuously.
24:40
SPEAKER_05: So we all get focused, no, we get focused on creating that semantic layer of the ontology.
24:44
SPEAKER_05: Like that's just a means to an end.
24:46
SPEAKER_05: Focusing so much on that doesn't provide any value.
24:48
SPEAKER_05: I need to be able to go understand why I'm creating that and and where's the data.
24:52
SPEAKER_05: So we go through a whole uh uh what we call a pay as-you go methodology about where you start with what are the business questions you want to go answer, and then you start saying, I'm gonna go with literally select one number one.
25:05
SPEAKER_05: You go through a first phase that we call the knowledge capture phase, which is trying to understand what that question means.
25:10
SPEAKER_05: There are words in that question.
25:11
SPEAKER_05: What do these words mean?
25:12
SPEAKER_05: To whom does that mean?
25:13
SPEAKER_05: I'm getting different answers from different people.
25:15
SPEAKER_05: Who are those people?
25:16
SPEAKER_05: Like you go through that whole process, understand what that means, and figure out where it is in the data.
25:20
SPEAKER_05: And you generate what I call the knowledge report.
25:22
SPEAKER_05: That's that first phase.
25:23
SPEAKER_05: Then you go off and implement that.
25:24
SPEAKER_05: All right, now we're actually gonna go put it in.
25:26
SPEAKER_05: We're gonna create those mappings, it's SQL, we're mapping these things.
25:29
SPEAKER_05: At the end, you're gonna start generating the data in the form of the knowledge graph.
25:33
SPEAKER_05: But also, you need to understand how do people want to consume this.
25:36
SPEAKER_05: I want to consume it at a table, I want to consume it as an API, whatever.
25:39
SPEAKER_05: Give that to the folks so they can actually review it.
25:41
SPEAKER_05: And then they're actually gonna go test it and saying, was I able to go answer that question or not?
25:45
SPEAKER_05: If they're able to go answer that question, success.
25:47
SPEAKER_05: If they're not, not success, and you figure out where you where where to go.
25:51
SPEAKER_05: So then then you continue with the next question and you start iterating.
25:54
SPEAKER_05: So you start something really small, and then it keeps growing little by little.
25:58
SPEAKER_05: At some point, you realize, hey, I have enough that I can start asking a bunch of other questions, right?
26:03
SPEAKER_05: I I answered question one, two, and three.
26:04
SPEAKER_05: And then, like, by the time I finish question three, like question 17, uh 22, and like I can start asking those, like, oh wow, great.
26:13
SPEAKER_05: So then you kind of reshuffle priorities.
26:15
SPEAKER_05: So that's a lot of the process that we explained through the through uh through the book.
26:18
SPEAKER_02: Because of those relationships that you're building, you're able to do more inference and be able to uh uh uh answer questions that you didn't have uh that you didn't think of initially.
26:28
SPEAKER_05: I so what I like, I tell knowledge graphs are knowledge graphs enable you to scale, and I don't mean scale from a technical point of view, I mean scale from a social point of view from the people, the process.
26:42
SPEAKER_05: I want to be able to go deliver a body of data and knowledge, which is called a knowledge graph, such that I can satisfy the known use cases that I'm giving right now, and also the unknown use cases that I don't know.
26:56
SPEAKER_05: I mean the other the the unknowns that's gonna come tomorrow.
26:59
SPEAKER_05: That is the value, that's how I'm scaling.
27:02
SPEAKER_05: Otherwise, I'm just gonna go do, I'm gonna be very efficient.
27:05
SPEAKER_05: Oh, I can here answer this question, here answer this question, answer this question.
27:08
SPEAKER_05: But answering that question, there's a lot of work, and that work there's a lot of knowledge, and that's not that doesn't get that gets repeated, it doesn't get shared, it's not the same.
27:16
SPEAKER_05: So knowledge graphs are a way to scale and to be resilient.
27:21
SPEAKER_05: And I think that's that's a differentiation.
27:23
SPEAKER_05: That's that's that's what that's the value of how they differentiate from just doing things like, oh, I can just do this and write some quick SQL query and go do that.
27:30
SPEAKER_05: Like, of course you can.
27:31
SPEAKER_05: Now, the issue is about incentives.
27:33
SPEAKER_05: Are you being incentivized to go do something efficient now?
27:36
SPEAKER_05: Or are you doing something are you being incentivized to put amount of work that is going to stand the test of time?
27:42
SPEAKER_02: Right, right, future-proof things.
27:44
SPEAKER_02: I just want to take a moment, uh, we're about halfway through our time.
27:47
SPEAKER_02: Today, wanted to remind our listeners that we're talking with knowledge graph expert Juan Sakeda.
27:53
SPEAKER_01: Hey, I also I also wanted to make sure, because I mean, this is this is mostly a podcast folks on AI and stuff.
28:00
SPEAKER_01: And this is absolutely critical because to remind our listeners that as we talk talk about, you know, I don't know if you want to even talk about like the infrastructure modernization or the data generation into the data prep, the data um, you know, adherence and governance, which I want to go back to that, you know, and ask you some specific questions about that one, that then gives us the ability to create, you know, like Seth had said, the inference, right?
28:27
SPEAKER_01: And the application of artificial intelligence to actually even ask those questions and then put it into application to then drive insights and outcomes.
28:35
SPEAKER_01: So, and you know, I love when Seth brings up this uh notion of you can't have good AI without a good information architecture, right?
28:44
SPEAKER_01: So let's talk about that for just a minute, Juan.
28:46
SPEAKER_01: I'd love to get your perspective on how you advise and what your what your thoughts are when you talk about when you're pulling all this data in, the evolution of it will change over time because you're also you know asking questions, getting more insights, questions, and then those answers come back, and then that can update your data graph or your knowledge in ontology.
29:06
SPEAKER_01: How do you deal with the governance and with the discovery to evolve into you know more?
29:12
SPEAKER_05: Yeah, so my answer.
29:13
SPEAKER_05: So I I came to this realization uh when I joined Data.world.
29:19
SPEAKER_05: And uh so Brian Jacob, who's the technical the CTO founder of Data.world, I've known him for a long time.
29:25
SPEAKER_05: He actually was did the the in uh the the tech due diligence for one of the investors of of my previous company, CapCenta.
29:32
SPEAKER_05: And that's how we've really kind of gotten together and known realized that we're so aligned.
29:36
SPEAKER_05: Yeah, he told me back in when when he acquired the company, he said, Juan, the secret here is that we're gonna get the world to get into knowledge graphs through a data catalog because we're gonna insert knowledge graphs through a data catalog.
29:49
SPEAKER_05: And and the data catalog is just a way to go inventory manage all the the kind of your data resources.
29:55
SPEAKER_05: It is fancy word, it's metadata management.
29:58
SPEAKER_05: That's it, right?
29:59
SPEAKER_05: So what does that mean?
30:01
SPEAKER_05: What that meant was your first knowledge graph should be of your metadata, and I think this was this was an aha moment I had fairly recent, fairly recent, three years ago.
30:12
SPEAKER_05: I have been working on creating knowledge graphs, large-scale knowledge graphs about your customer 360, your patient 360, your drug 360.
30:20
SPEAKER_05: And I'm like, well, you you run into kind of a lot of the the friction there because people is like, well, we were we've been doing that already.
30:28
SPEAKER_05: We've been doing that 360 with MDM tools, right?
30:30
SPEAKER_05: And so then they go back to but but our old ways, right?
30:34
SPEAKER_05: And then then you have these arguments about well, graphs are better.
30:37
SPEAKER_05: Then what is the arguments?
30:38
SPEAKER_05: They go back into the technology.
30:40
SPEAKER_05: Um, I always say one of my favorite quotes we have in the book is the limits of my language are the limits of my world.
30:47
SPEAKER_05: Ludwig Wittgenstein.
30:48
SPEAKER_05: That means if everything in your world is tables and SQL, then you believe everything can be done with tables and SQL, right?
30:53
SPEAKER_05: So that I mean that's one of the reasons why having the knowledge graph is something kind of hard for people to grasp.
30:57
SPEAKER_05: But what we really, what we realized was that what um the metadata is that first part that you is is that entry point to to your data management.
31:08
SPEAKER_05: If we if we go back in time, all of data management has been just on data.
31:11
SPEAKER_05: You do ETL, I put it under a warehouse.
31:13
SPEAKER_05: The missing element before is the lack of metadata.
31:16
SPEAKER_05: And what metadata gives me all that context.
31:19
SPEAKER_05: Now, understanding the semantics, what metadata means, is actually easier to go do.
31:24
SPEAKER_05: I think we're not going to have that much.
31:25
SPEAKER_05: We're not, we get people in the room.
31:27
SPEAKER_05: I think everybody's gonna agree that a database consists of tables.
31:30
SPEAKER_05: Tables have columns, right?
31:32
SPEAKER_05: Pretty straight, the ontologies behind this is very straightforward, which they already exist.
31:36
SPEAKER_05: Like we use open source, open standards on on vocabularies like DCAT and Dublin Core and Provo and all this stuff.
31:42
SPEAKER_05: Um, so you start building your knowledge graph about your metadata, and people use that to say, I want to be able to search and find the data that I have today.
31:52
SPEAKER_05: That's my baseline around that stuff.
31:55
SPEAKER_05: Uh, I want to understand what are the what are the what are people searching for, right?
31:59
SPEAKER_05: What are the things that they are, what are the things they agree the most on that that that's the most important data and so forth.
32:05
SPEAKER_05: That's the metadata layer we go first.
32:07
SPEAKER_05: And at the end of the day, if you think about it as an X360, the X360 is basically my my table 360, my my business concept 360.
32:16
SPEAKER_05: So even before getting into like richer ontologies, let's just start with a list of words, a very basic business glossary.
32:22
SPEAKER_05: And then oh, they say, oh, wait, but this table is actually represents this concept.
32:28
SPEAKER_05: All right, let's create that relationship.
32:29
SPEAKER_05: They started to create a very simple mapping around these things, and then little by little, they kind of get that that that muscle.
32:35
SPEAKER_05: Oh, okay.
32:36
SPEAKER_05: I'm starting to understand what data I have.
32:38
SPEAKER_05: I'm trying to understand what the what are the the concept people are using.
32:42
SPEAKER_05: By the way, the next thing that comes into play is connecting the people.
32:46
SPEAKER_05: You know what?
32:47
SPEAKER_05: The catalog should not just be of my tables and columns and my dashboards and my business glossary terms.
32:52
SPEAKER_05: It should also include the people, their team structures, who do they report to, right?
32:57
SPEAKER_05: Because then I want to say, oh, this table or this database is actually being governed as a steward is this person.
33:05
SPEAKER_05: This person is in this department.
33:07
SPEAKER_05: What happens if this person leaves?
33:08
SPEAKER_05: I know what the department suggests.
33:10
SPEAKER_05: So you start building that knowledge graph of your metadata, which is actually the other thing is that it's a smaller amount of data first.
33:16
SPEAKER_05: People get get they get it.
33:17
SPEAKER_05: They're like, oh, I want more things in there.
33:19
SPEAKER_05: Can you connect this thing?
33:20
SPEAKER_05: Can you connect this thing?
33:21
SPEAKER_05: And then suddenly your knowledge graphs turn off your metadata and it just starts growing and growing and growing and growing.
33:27
SPEAKER_05: So that's that's how that's what I advise.
33:30
SPEAKER_05: So I tell everybody you want to go build the knowledge graph, and and and that's running.
33:36
SPEAKER_05: Crawling is we focus first on your metadata.
33:39
SPEAKER_05: That's why you start with the data catalog, and that's why your data catalog should better be down built on a knowledge graph.
33:45
SPEAKER_01: Yeah.
33:46
SPEAKER_01: Yeah, that's good.
33:47
SPEAKER_01: Because otherwise, you're gonna, you know, the the the biggest issue I see a lot of times too is to your point, and I use more of the Wayne Gretzky style, like you miss 100% of the opportunities you don't take, right?
33:57
SPEAKER_01: Same thing.
33:58
SPEAKER_01: You miss 100% of the things you don't know about.
34:00
SPEAKER_01: So part of that too is um, but if you always stay within that confined and predefined box, then you're gonna screw yourself out of so many other opportunities.
34:11
SPEAKER_01: And that's what I see a lot of times, especially within within these data environments.
34:14
SPEAKER_01: I had a customer, you know, just last week that was so excited that they were getting all their data into Redshift.
34:21
SPEAKER_01: I'm like, that's awesome.
34:22
SPEAKER_01: That's you know, and that but they believe that this is the end all be all to your point on the technology solving all problems, is that this is gonna be the um the silver bullet for everything that they do.
34:32
SPEAKER_01: And I'm like, well, that's a bummer because um I get it that it's a columnar system and it's great and it's fast and it's gonna do all this great stuff, but it's predefined.
34:43
SPEAKER_01: And you have nothing that's also outside of that.
34:46
SPEAKER_01: Great, you're pumping everything into it, but you always again, you're pre-defining everything.
34:50
SPEAKER_01: So, how are you gonna evolve when you know you go from red to green to purple instead of blue?
34:56
SPEAKER_01: What's gonna, you know, how's that gonna change and how you're gonna evolve that?
35:00
SPEAKER_01: And they didn't have an answer, of course, because it was technologists building this, convincing the business that this is gonna be the end-all be-all for all of their reporting and dashboards and BI and you name it, you know, down the road to make insightful decisions.
35:13
SPEAKER_01: Yet it doesn't give them the application that they need to actually answer the business and what they need in terms of that solution.
35:20
SPEAKER_01: So, yeah, I'm with you.
35:21
SPEAKER_01: It's it's it's super critical.
35:22
SPEAKER_01: And I get, you know, again, going back to how important that prep step is to, you know, with the crawl and understand, hey, you don't know everything, but if you put some flexible systems in place to be to take that in and the all the metadata, guess what?
35:39
SPEAKER_01: You can actually then, you know, prepare for the future of unknown.
35:43
SPEAKER_05: If there's one takeaway that one everybody to have is are you focused on being efficient and doing something quickly now?
35:50
SPEAKER_05: Or are you setting yourself up to be able to are you setting yourself up to be resilient to focus on the known cases of today and the unknown cases of tomorrow?
36:00
SPEAKER_05: Ask yourself where you are.
36:02
SPEAKER_05: And guess what?
36:03
SPEAKER_05: You're probably being incentivized to be efficient and not being resilient, right?
36:06
SPEAKER_05: And also, let's be very honest, as technologists, people like to jump around, get it, go jump from job to job.
36:11
SPEAKER_05: So actually, are not even incentivized to go put in more effort to go do something that's gonna survive more time because they're not gonna be around.
36:20
SPEAKER_05: So that I think so this all goes back to to Skinner and incentives.
36:26
SPEAKER_02: What are you saying?
36:27
SPEAKER_02: Uh, so there's a lot of AI vendors out there that are still claiming that you know you don't need reference data, that uh they just point their systems to you know the content and they ingest it and they you know build questions and answers from that.
36:42
SPEAKER_02: And you know, and and what they're much of what they're doing behind the scenes is building ontologies to recognize things like intents, right?
36:50
SPEAKER_02: And and look at utterances and mapping those to intents and then kind of related concepts so they can understand, you know, the semantic worlds that they're they're addressing.
36:59
SPEAKER_02: But what do you say to those organizations who are hearing that from vendors and hearing that they don't need things like knowledge architecture, they don't think you need things like a knowledge graph.
37:10
SPEAKER_02: Uh, their system does uh all of it, and uh I I always find that uh hard build to swallow.
37:17
SPEAKER_02: And and and you know, they show you nice canned use cases and examples and things that demo well in PowerPoint.
37:23
SPEAKER_05: But so so a couple things.
37:26
SPEAKER_05: I host this podcast called Catalog and Cocktails, the honest no BS non-salesy data podcast.
37:31
SPEAKER_05: Uh, and in this last season, we had two really good AI episodes.
37:36
SPEAKER_05: Uh, one was with Andrew I, who's the CEO of Closed Looped AI, and the other one was with Patrick Bangert, who is the VP of AI at Samsung.
37:44
SPEAKER_05: And uh, so what I'm gonna say is kind of a lot of stuff that I learned from them.
37:48
SPEAKER_05: Um, first of all, something from Patrick, uh, whatever he said is we're all focused on autonomous vehicles, right?
37:56
SPEAKER_05: That's kind of the main thing of AI, and like, oh, that's so cool.
37:59
SPEAKER_05: And he said something which surprised me, which was uh, autonomous vehicles is actually an easy problem because you just need a lot of data.
38:08
SPEAKER_05: And so a lot and and all these AI models need a lot of data, and a lot of that data we're focusing on is a lot of unstructured data, images and text, right?
38:17
SPEAKER_05: So, um, if you look at autonomous vehicles, you need a lot of the images and videos and that stuff, and that's how they can build these amazing uh AI systems to do autonomous vehicles.
38:25
SPEAKER_05: But then we're like, if you were to go, what are what is the knowledge that you need, the knowledge you need to learn uh to go drive a car?
38:35
SPEAKER_05: I mean, you're gonna go take a driving, the driving test, the booklet around what you have to go read is pretty small.
38:44
SPEAKER_05: The amount of knowledge there is a small thing.
38:47
SPEAKER_05: And therefore, autonomous vehicles and that type is just a bunch of skills.
38:52
SPEAKER_05: And you get skills is by a lot of practice by looking.
38:55
SPEAKER_05: So I think there are these AI applications where if it's focused on low body of knowledge and high skill, and and I can get I can get that skill by looking at a lot of things which is unstructured, that's fine.
39:07
SPEAKER_05: Perfect.
39:07
SPEAKER_05: I think those would be those areas of AI where you don't need to have extra structured knowledge and have knowledge graphs.
39:13
SPEAKER_05: I mean, this autonomous vehicles are shown us that.
39:16
SPEAKER_05: So that wasn't a surprising to think is like, oh, autonomous vehicles is actually an easy thing.
39:21
SPEAKER_05: Um, but then you go to other areas where if I have a uh I've I've I mean now you start to go very, very specific, very narrow.
39:30
SPEAKER_05: So in healthcare, if you go very, very narrow in in x-rays and stuff, and you've been able to go annotate so many things, then it is a very, very narrow aspect.
39:41
SPEAKER_05: I mean, there's you just had the humans putting the knowledge in that stuff, and doing the all the labeling.
39:46
SPEAKER_05: So, where did the where did the knowledge aspect come in there is a lot of work on doing annotations, but you're doing something very narrow.
39:53
SPEAKER_05: Now, if you look at the AI for just detecting things on radiology, that's even better than doctors.
39:58
SPEAKER_05: So if you're if you're actually getting a x-ray chat your x-ray of your chest or whatever, you want the AI to tell you what to do, not the doctor, because the doctor has a higher error rate than the AI at this moment, but because it's so narrow right there.
40:11
SPEAKER_05: Now, the thing is, but because it's also then the other thing is it's all unstructured data, right?
40:16
SPEAKER_05: Images.
40:17
SPEAKER_05: So for me, it's if you're starting to go deal with a lot of structured data, that's where you need to be able to go.
40:24
SPEAKER_05: Uh, which a little bit, it's a lot of structured data, and the the knowledge kind of surface is pretty high and it's pretty and it's it there, it's not there's not agreements because of humans.
40:36
SPEAKER_05: That's where you need to have the knowledge.
40:37
SPEAKER_05: So I think those are the two types of things we need to go look into.
40:40
SPEAKER_05: So structured data, and where you put humans in the room and they don't agree, that's already indicated that we need to have put that effort into the knowledge part.
40:49
SPEAKER_05: Because otherwise, what I tell people is not even the humans will agree.
40:53
SPEAKER_05: Why the heck do you believe a machine is gonna come up and give you give you the right answer?
40:56
SPEAKER_05: And even if it gives you an answer, why would you actually trust that?
41:00
SPEAKER_05: Yeah, somebody has to go into that.
41:02
SPEAKER_05: So I think what we're gonna go start seeing is this whole human, not I mean, people talk a lot about human in the loop.
41:07
SPEAKER_05: I actually think it's more about machine in the loop around it.
41:10
SPEAKER_05: Um, but then all this needs to be really well uh coordinated on figuring out something much more vertical.
41:16
SPEAKER_05: So I think the complete uh general AI, the other company who say, I have I have trained my model on so much stuff.
41:23
SPEAKER_05: It's like, yeah, you've trained your model on all the stuff, but you haven't trained on my data.
41:27
SPEAKER_05: And how do I know that my data is relevant to what you go do?
41:29
SPEAKER_05: So uh so in both of these two episodes we had our podcast, independently they come to the same conclusion saying, Yeah, you can have all these great AI models, but it really depends the the devil's in the details, and you need to have your details, which is your data.
41:43
SPEAKER_05: So that's why just using whatever random uh open generic AI model is not the way to go.
41:50
SPEAKER_05: And then again, if you have the not the knowledge part is is is crucial when you want to be able to have explainability, want to be able to give trust about what this stuff comes from.
42:01
SPEAKER_05: Um, so very clear that the knowledge graphic is is what comes after.
42:06
SPEAKER_05: Uh everybody gets so excited about what you can go do.
42:08
SPEAKER_05: I mean, everybody's doing GPT GPT 3, all that stuff, but there really is no explainability around that.
42:14
SPEAKER_05: Now we're all having fun with it, but once we translate that to the real business use cases, we're gonna realize oh, we need the governance behind that, right?
42:24
SPEAKER_05: What does that stuff come from?
42:26
SPEAKER_05: That's where the knowledge comes in.
42:28
SPEAKER_02: And I and I also think if you look at some of the pure machine learning applications like uh uh machine vision to say identify defective parts coming off of uh an assembly line, you still need to contextualize that that observation.
42:44
SPEAKER_05: What does defective mean?
42:45
SPEAKER_02: And to whom defective mean what manufacturing line, what product, what are the source suppliers, what are the processes?
42:52
SPEAKER_02: Who's gonna be responsible for remediation?
42:55
SPEAKER_02: What is that remediation?
42:56
SPEAKER_02: Is it change to a procedure?
42:58
SPEAKER_02: Is it change to a vendor?
42:59
SPEAKER_02: Is it change to a parameter?
43:00
SPEAKER_02: All of that is the knowledge uh context that needs to uh be put into place or that needs to be surrounding that observation of the effective product, right?
43:12
SPEAKER_02: So uh so that's kind of a big when you start thinking about knowledge architectures, they they you know, it's contextualizing the answers that you get from uh from uh AI as well.
43:22
SPEAKER_02: So what are your, you know, I knew you had some takeaways uh about looking at the business problem um and uh and really focusing on understanding your role.
43:31
SPEAKER_02: Uh any other uh big picture takeaways that you'd like uh executives to walk away with or programmers to walk away with?
43:38
SPEAKER_02: I mean, you've covered uh a lot of ground.
43:40
SPEAKER_05: So I want to wrap up with the story from one of our customers who's like very they're very forward thinking on on and what they're doing aligns with this vision that we have of what I call we need to move from a data first world to a knowledge first world.
43:54
SPEAKER_05: Um, so Prologis is uh one of the world's largest kind of uh uh um warehouse logistics.
44:00
SPEAKER_05: Company, right?
44:00
SPEAKER_05: If you buy anything on Amazon, it'll probably go to their prologue's warehouse.
44:03
SPEAKER_05: The problem is they realized that um their warehouses, when they would go to the warehouses, sometimes there was it was kind of uh empty in the sense that hey, we could have been stacking more pallets in here and and and and right and making more money, or sometimes it was packed and they were like, Hey, there's stuff outside, like what happened, what's going on?
44:25
SPEAKER_05: Um so they when they looked at this particular problem, this one instance of this problem, there they realized somebody went all the way and they connected it to a data quality issue.
44:34
SPEAKER_05: They're like, Oh, we don't have the accurate data around the height of a building.
44:40
SPEAKER_05: So if we look at so now, so what is the height of a building?
44:43
SPEAKER_05: So if we look at so there's some data in a data first world, I'm like, here's a table and it has some numbers in it.
44:49
SPEAKER_05: Okay, what does that mean?
44:50
SPEAKER_05: Well, it's a positive number, it's a right, there's some null values in there, but what does it mean?
44:55
SPEAKER_05: Oh, it has the column name is called length.
44:58
SPEAKER_05: Okay, so I'm measuring something.
44:59
SPEAKER_05: Oh, the the column next to it it has a unit and it's feet.
45:01
SPEAKER_05: Okay, yeah, so I'm definitely measuring something over there.
45:04
SPEAKER_04: Okay.
45:04
SPEAKER_05: Uh imagine you give that to an AI engine, right?
45:08
SPEAKER_05: Or my model, like they built there's a number in there, but what does that number mean?
45:12
SPEAKER_05: Now that's a data first world is give me more data, give me more data, I need more data.
45:16
SPEAKER_05: Like, here's data, and then and and that's it.
45:18
SPEAKER_05: I focus on the data.
45:18
SPEAKER_05: I'm like, wait, so you're telling me you can't answer the question because you lack data.
45:21
SPEAKER_05: So if I give you more data, problem solve, like, no bullshit on that.
45:25
SPEAKER_05: We need what we're missing is to understand what that column means.
45:28
SPEAKER_05: Somebody tells me, oh, that column actually is the height of a building, height of a warehouse.
45:34
SPEAKER_05: Oh, okay.
45:36
SPEAKER_05: So what can I infer from that?
45:38
SPEAKER_05: Well, guess what?
45:39
SPEAKER_05: Heights of a warehouse are between 20 feet and 50 feet.
45:44
SPEAKER_05: Oh, by seeing numbers outside of that, somebody probably typed something in wrong.
45:48
SPEAKER_05: You can even take it to the next level and say, wait, if you look at the at the eight, what the what the building was built, there were regulations that allowed after the 90s or whatever, the heights could building to go up to 50, but before the 90s, they can only go up to 30.
46:02
SPEAKER_05: Oh, so I can get more even accurate data around that stuff.
46:05
SPEAKER_05: By the way, if there's a null value, well, we know that a building obviously has a height, so somebody knows what that number is.
46:11
SPEAKER_05: Let's go track that number down and get it incorrectly.
46:14
SPEAKER_05: Um, and why does this care?
46:15
SPEAKER_05: Why do we care about this?
46:17
SPEAKER_05: Making money, saving money is like, well, if I have the wrong number, I if I thought I could stack more stuff in there, I deal with it.
46:23
SPEAKER_05: I'm leaving money on the table by putting packets outside, or I thought I could, I thought I was full, but I had more space, I could have sold more stuff in here.
46:30
SPEAKER_05: Tying that directly to business value.
46:32
SPEAKER_05: Now, they they this is thinking about what does this stuff mean?
46:37
SPEAKER_05: What are the implications?
46:38
SPEAKER_05: That's a knowledge first world.
46:39
SPEAKER_05: The knowledge first world is people first, context first, relationships first.
46:43
SPEAKER_05: So, what I want the kind of people the folk people to start listening is shift from a data first world to a knowledge first world.
46:51
SPEAKER_05: And that knowledge first will we get that context, part of that context is connecting it to the business directly.
46:56
SPEAKER_05: I think there needs to be direct connection to how we make money and save money.
47:00
SPEAKER_05: Now, going back to incentives is that they their goal wasn't just to solve this problem and make sure that they're making more money, save more money just because they can stack more stuff in there.
47:10
SPEAKER_05: What they wanted to go do was a cultural change.
47:13
SPEAKER_05: So you know what they did?
47:15
SPEAKER_05: They said 20% of everybody's bonus depends on data quality.
47:23
SPEAKER_05: You work on your you work your boots on the ground in the warehouse with a clipboard, 20% of your bonus, money in your pocket depends that we all as a company improve our data quality because you're probably putting in the number and we want to make sure that we have that culture.
47:36
SPEAKER_05: So their ultimate goal was not just to improve these different types of business processes, their goal was to create a culture, an incentive of treating your data with care.
47:46
SPEAKER_05: That's all that is what that's what you can call it.
47:49
SPEAKER_05: So true leaders are people who are thinking about data that way.
47:52
SPEAKER_02: Well, that's impressive.
47:54
SPEAKER_02: I mean, that's really putting your money where your data mouth is.
47:57
SPEAKER_05: And that's executive leadership.
47:59
SPEAKER_01: Yes.
48:00
SPEAKER_01: There was there was a contract manufacturer that um that I'd worked with in the past, and they wrote a book called uh um Project for Profit, right?
48:09
SPEAKER_01: And or Project Management for for Profit.
48:11
SPEAKER_01: And what they did was it's kind of exactly what you're saying, every single person, every Monday, they'd pull them in and they would show them where they were on their because they were doing hardware engineering projects and stuff, and they would show them where they were on the cost basis of if they were ahead of schedule or behind schedule.
48:26
SPEAKER_01: But everybody's bit bonus in the entire organization down to the janitor came from the perspective of where they were on the positive or negative of each of those projects.
48:36
SPEAKER_01: And so what that drove was the right level of incentive and behavior for everybody in the organization to do the right thing for the business because of the direct incentive.
48:45
SPEAKER_01: So you would see people like, well, the janitor's like, well, what can I do to clean these environments up better for you so that when you do have a cost overrun on these projects, we can actually, you know, then you know be it more adaptive and more, more uh adjacent to what you need to be doing and stuff.
49:01
SPEAKER_05: It's the perfect everything here goes back to incentives, behaviors, and skinner.
49:07
SPEAKER_01: Exactly.
49:08
SPEAKER_01: And what should we be looking for from uh from you in 2023?
49:11
SPEAKER_01: What what are you doing?
49:13
SPEAKER_01: What kind of things you want to you know talk about, promote?
49:15
SPEAKER_01: What about you?
49:16
SPEAKER_05: Yeah, so I think one of the things I'm I will continue to push more and more always is the focus on knowledge.
49:23
SPEAKER_05: We live in this world of just focusing on data, and even like I say, I don't even like the word data catalog, I don't like it because it's much more than cataloging data.
49:30
SPEAKER_05: It should be about cataloging data and knowledge.
49:32
SPEAKER_05: So I think those are the we start seeing more examples of knowledge, of connecting more to the business and being able to really call out hey, is that work?
49:42
SPEAKER_05: Is that is that is is this discussion, is this the tech thing that we're doing, is that contributing to the bottom line of organization?
49:48
SPEAKER_05: And the way to understand that is uh having more of the knowledge.
49:53
SPEAKER_05: So I think uh even with us at data.world, we're we're we're showing how we can go catalog knowledge, how we can go catalog business process.
50:00
SPEAKER_05: We talk about data lineage.
50:01
SPEAKER_05: Oh, I I change a column here.
50:03
SPEAKER_05: What is that going to impact going forward, right?
50:05
SPEAKER_05: Or hey, this dashboard has a number.
50:07
SPEAKER_05: Well, where did that number come from?
50:08
SPEAKER_05: Well, I want to know is like, hey, there is this number that is in this database goes inside some decision, a decision model that impacts this business of how we make money.
50:17
SPEAKER_05: So I want to have more business lineage.
50:19
SPEAKER_05: So I want to be able to go catalog the knowledge of the business.
50:22
SPEAKER_05: I tell people another thing we should be doing is like go ask all your executives, your leadership team, what keeps you up at night?
50:28
SPEAKER_05: Like go catalog that and go realize hey, there's all these people are actually thinking about the same thing in different words, right?
50:35
SPEAKER_05: Are we doing something for that?
50:36
SPEAKER_05: Is that actually aligned with the goals that we have in our company?
50:38
SPEAKER_05: That's understanding the business.
50:40
SPEAKER_05: That's understanding the knowledge.
50:41
SPEAKER_05: So I think the the what I want people to look at data catalogs and all this metadata management, metadata, it's it's also including the knowledge.
50:49
SPEAKER_05: How does business work?
50:50
SPEAKER_05: And that's how we're going to be able to make sure that we're tying all our data work directly to making money, saving money.
50:56
SPEAKER_05: Because hey, we live in a capitalist world, so that's what we've got to go do.
50:59
SPEAKER_02: That's fantastic.
51:01
SPEAKER_02: Well, that's uh all of our time today.
51:04
SPEAKER_02: Uh, it's really been a pleasure talking with you, Juan.
51:06
SPEAKER_02: I've always enjoyed.
51:08
SPEAKER_02: Uh, and uh it's uh it's just great to catch up with you and always like hearing your ideas, and I especially love hearing the integration of knowledge uh and data.
51:16
SPEAKER_02: So thank you again for today.
51:18
SPEAKER_02: Good luck with the new baby and uh congratulations on it as well.
51:22
SPEAKER_01: All right, cheers, everybody.
51:24
SPEAKER_01: Great, thanks, Juan.
51:25
SPEAKER_01: Appreciate it.
51:26
SPEAKER_01: Take care.
51:27
SPEAKER_01: I know.
51:31
SPEAKER_00: Thanks for listening to the Early AI podcast with your hosts, Seth Early and Chris Featherstone.
51:37
SPEAKER_00: We hope you enjoyed the program and took away some ideas that you can start to implement in your organization right now.
51:43
SPEAKER_00: If you're stuck and need help, check out www.early.com for information about Early Information Science Consulting Services.
51:51
SPEAKER_00: And of course, check the show notes for links to any resources we mentioned today.
51:56
SPEAKER_00: That's all for now.
51:57
SPEAKER_00: See you next time.