Latest / The Adults In The AI Room / The Secret Power of Collaboration in Data Science - Ben Clinch - Earley AI Podcast with Seth Earley - Episode 33

Transcript

The Secret Power of Collaboration in Data Science - Ben Clinch - Earley AI Podcast with Seth Earley - Episode 33 · The Adults In The AI Room

• 0:01
  SPEAKER_01: This is the Early AI Podcast, hosted by Seth Early and Chris Featherstone.
• 0:06
  SPEAKER_01: Join us as we delve deep into the passions, expertise, and experiences of thought leaders and practitioners to talk about what's possible in artificial intelligence.
• 0:15
  SPEAKER_01: Early AI Podcast is sponsored by Early Information Science, your digital transformation journey with the design and deployment innovative technology solutions.
• 0:27
  SPEAKER_03: Good morning.
• 0:28
  SPEAKER_03: Welcome to today's podcast.
• 0:30
  SPEAKER_03: My name is Seth Early, and I'm Chris Featherstone.
• 0:34
  SPEAKER_03: And today I want to introduce our guest.
• 0:37
  SPEAKER_03: He's an industry expert in data management, knowledge graphs, and business and data architecture.
• 0:43
  SPEAKER_03: He brings 23 years of experience in business technology and operations across some of the largest multinational organizations.
• 0:50
  SPEAKER_03: He has a passion for communicating complex ideas in relatable ways and teaching how to implement data management at scale.
• 0:57
  SPEAKER_03: He's a sought-after public speaker, having spoken at Gartner Data and AI Summit and Google Next, among many others.
• 1:04
  SPEAKER_03: He's an advocate of the Enterprise Data Management Council, a nonprofit promoting the highest standards in data management.
• 1:10
  SPEAKER_03: He's obsessed with knowledge graphs, starting the guild at British uh BT, I guess it's now called BT, uh, with 200 plus members.
• 1:17
  SPEAKER_03: He's head of the information architecture for BT Group, known as Mr.
• 1:21
  SPEAKER_03: Data Fabric.
• 1:22
  SPEAKER_03: Ben Clinch, welcome to the show.
• 1:25
  SPEAKER_02: Thank you, Seth.
• 1:26
  SPEAKER_02: Thank you, Chris.
• 1:27
  SPEAKER_02: It's great to be here.
• 1:28
  SPEAKER_00: Yeah, thanks, Ben.
• 1:29
  SPEAKER_00: It's always a pleasure to uh one just one to know you.
• 1:33
  SPEAKER_00: And then, you know, all our interactions that we've have, we have are just you know, one thought-provoking, but also very gracious.
• 1:39
  SPEAKER_00: So thanks for the time.
• 1:40
  SPEAKER_03: So, Ben, tell me, what do you think the biggest top the biggest uh misconceptions are about generative?
• 1:45
  SPEAKER_03: That is one of the areas.
• 1:46
  SPEAKER_03: I mean, we know you're you're heavy in debt data and knowledge graphs and uh and uh uh information architecture and data management and data architecture, but you know, the whole thing with generative AI is so first in uh uh in people's minds.
• 2:02
  SPEAKER_03: Uh, how does that relate?
• 2:03
  SPEAKER_03: What do you think the biggest misconceptions are these days about generative AI?
• 2:08
  SPEAKER_02: So I I think first of all, obviously, generative AI is an incredibly valuable tool um in the right hands.
• 2:15
  SPEAKER_02: But the biggest misconception, I guess, from my perspective is is that many people assume that um it's not necessary to curate and prepare data uh now now that we have these kind of capabilities.
• 2:30
  SPEAKER_02: And of course, uh generative AI models like OpenAI and others have have hoovered up huge amounts of structured data across the world and and filtered out a lot of data as well.
• 2:41
  SPEAKER_02: Um they've they've um dismissed data that they don't want to utilize.
• 2:46
  SPEAKER_02: So um I it that's very easy to overlook for the many millions of people who've interacted with these kind of large language models through um sort of chat interfaces.
• 2:58
  SPEAKER_02: So the magic is still behind the curtain, so to speak.
• 3:01
  SPEAKER_03: Right.
• 3:02
  SPEAKER_03: And you had mentioned uh when we were chatting before that you're kind of leaning on someone else's data model or someone else's knowledge in a lot of ways, uh, as it is today.
• 3:12
  SPEAKER_03: Can you explain?
• 3:12
  SPEAKER_03: Can you say more about that?
• 3:14
  SPEAKER_02: Yes, absolutely.
• 3:15
  SPEAKER_02: So so it wasn't that long ago.
• 3:16
  SPEAKER_02: I was I was at a conference and somebody said, Well, we know no longer need to do any data modeling.
• 3:21
  SPEAKER_02: Because uh because you know, uh large language models can now create data models entirely for us without us having to do any work.
• 3:30
  SPEAKER_02: And of course, you know, it's it's a great accelerator for that type of activity, but in reality, the um it's been trained on huge um sets of information like schema.org and and other sets of information freely available across the web and ontologies, and and and those are being curated and created thoughtfully by brilliant people around the world.
• 3:54
  SPEAKER_02: And um from that perspective, what you're doing is you're deferring your data modeling to a standard that everybody else is going to have access to.
• 4:02
  SPEAKER_02: And if if no data modeling goes on anymore, then then actually it's not it's going to increasingly differ and diverge from reality.
• 4:11
  SPEAKER_02: Um, so while absolutely can accelerate these types of activities, it would be a disaster to solely rely on on um somebody else's model for you.
• 4:20
  SPEAKER_03: You know, what's interesting is people used to ask me, well, don't you just have a taxonomy we can use?
• 4:25
  SPEAKER_03: And of course, taxonomy has lots of implications.
• 4:27
  SPEAKER_03: It can be navigational taxonomy, it can be data hierarchies, it can be uh data models, it can be content models, right?
• 4:33
  SPEAKER_03: It can power metadata fields, there's all sorts of things.
• 4:36
  SPEAKER_03: But I'd say, you know, look at uh go into two stores that sell exactly the same things.
• 4:42
  SPEAKER_03: Think of your even a display taxonomy as your your state store layout, your shelves, your your signage.
• 4:48
  SPEAKER_03: Do you go into two stores and they look exactly the same?
• 4:52
  SPEAKER_03: If everything looked exactly the same, there would no be there'd be no competitive differentiation, right?
• 4:57
  SPEAKER_03: So when you standardize, that's good for efficiency, but differentiation is what gives you that competitive advantage.
• 5:02
  SPEAKER_03: So it sounds like you're saying the same thing.
• 5:05
  SPEAKER_03: It's it's that that's fine, but it doesn't give you any differentiation.
• 5:08
  SPEAKER_03: It's a good starting point, but it doesn't give you bad.
• 5:11
  SPEAKER_02: Absolutely agree.
• 5:12
  SPEAKER_02: And actually, so there's two angles on that that I think is is quite one that I like to pick up on, which is I always think of the um the data model being to data what the org chart is to people.
• 5:26
  SPEAKER_02: They're both incredibly powerful assets, data and people for any organization.
• 5:31
  SPEAKER_02: And um, and it's an analogy that that keeps giving in in many different ways.
• 5:36
  SPEAKER_02: And one is you wouldn't go to an external party and uh or and or maybe just go to the web and download somebody else's org chart and say, I can't be bothered to work out how to organize my people.
• 5:47
  SPEAKER_02: I'm gonna defer it entirely to somebody else's standard.
• 5:50
  SPEAKER_02: It might be a bad place to start, of course.
• 5:52
  SPEAKER_02: You know, what's the standard model for for however you would never defer your entire organizational design to somebody else's um um because people are too important and individual to uh to treat them that way.
• 6:07
  SPEAKER_00: That's also saying that, too, that that org chart comes with all of the HR responsibilities and or pay grades associated with it too.
• 6:14
  SPEAKER_00: So you may be getting a uh you know a um a bonus, or you may be getting a downgrade, depending on the I'd love to get your ideas.
• 6:24
  SPEAKER_03: We're gonna deploy this new standard org chart model.
• 6:27
  SPEAKER_03: Exactly.
• 6:29
  SPEAKER_00: That means you're you're now demoted.
• 6:31
  SPEAKER_00: All the attributions.
• 6:32
  SPEAKER_00: Sorry, CEO, you get a downgrade.
• 6:34
  SPEAKER_00: But in that vein, Ben, I'd love to get your take on how you actually, because you've been very successful at this, but how you get the business leadership within your organizations to understand the value of it, and what are those, you know, what are those pieces that you have to go through generally to help them understand it?
• 6:54
  SPEAKER_02: Right.
• 6:54
  SPEAKER_02: So the value of data or the value of data modeling or both.
• 6:57
  SPEAKER_00: Value data modeling.
• 6:58
  SPEAKER_00: Yeah.
• 6:58
  SPEAKER_00: Data data is like, you know, they it's data's crammed down their throats all day long around oh, it's the new oil and all that other, you know, yes, but data modeling.
• 7:07
  SPEAKER_02: Fantastic.
• 7:08
  SPEAKER_02: So for data modeling specifically, um, what what um and and you know to give many people context for in many organizations, data modeling has been really downplayed over many years.
• 7:19
  SPEAKER_02: And a lot of this is to do with with um various different vendor platforms saying you don't you don't need to create schemas, you don't need to cure curate data, pile it into my data lake uh platform and and people uh yeah, schema on read, or people can work it out themselves.
• 7:36
  SPEAKER_02: And what that really means is is that you've got one group of people who are putting data into the into the lake, who one would hope understands the data well or should, um, who are off the hook for creating and structuring that data in a meaningful manner.
• 7:52
  SPEAKER_02: And then you've got a huge number of teams who want to benefit from it, with teams of marauding data scientists swimming or or wading through the data lake, um, who are highly skilled individuals, I might add, but then but then they they become something akin to data janitors trying to work out what's in the data lake.
• 8:10
  SPEAKER_02: Um, and there are loads of them all doing it simultaneously and overlapping.
• 8:14
  SPEAKER_02: So rather than having one group of people who uh on the hook to actually make sure that it's well curated, you've got loads of highly paid people, highly skilled people wasting their time trying to work out what the data is.
• 8:25
  SPEAKER_02: So that's one example.
• 8:27
  SPEAKER_02: The other thing is the org chart's really powerful.
• 8:30
  SPEAKER_02: So people have seen uh that this uh you know the the issues with data scientists spending so much of their time curating data.
• 8:38
  SPEAKER_02: But also they'll say, well, do we really data modeling?
• 8:40
  SPEAKER_02: What's the return on investment?
• 8:42
  SPEAKER_02: You know, or when is the data model done?
• 8:44
  SPEAKER_02: You know, Ben, when when when when is the data model finished and it's not going to change again?
• 8:49
  SPEAKER_02: And that org chart analogy, which which I find everybody can relate to within the organization, they they I say, Well, okay, well, you tell me what's the ROI on an org chart?
• 8:59
  SPEAKER_02: Have you ever worked out the return on investment on organizing your people?
• 9:04
  SPEAKER_02: Um, you know, and they go, no, and why would I?
• 9:10
  SPEAKER_02: Exactly.
• 9:10
  SPEAKER_02: If it's a valuable asset, why not organize it and get them and maximize the value?
• 9:14
  SPEAKER_02: The second is when is the org chart done?
• 9:17
  SPEAKER_02: You you tell me when when it's not going to change again, you know.
• 9:20
  SPEAKER_02: Um, and I know both of those sound like sort of flippant questions uh or in rhetoric, but in fact, it does spark an interesting debate where where, of course, the data model needs to uh flex, it needs to grow and evolve with the strategy of an organization and in the same way that an org chart would.
• 9:40
  SPEAKER_02: And so I I try and use those types of analogies that are relatable and and often get you know get really good response from those.
• 9:49
  SPEAKER_03: Right.
• 9:50
  SPEAKER_03: So I I also have the analogous um arguments on the unstructured side, right?
• 9:56
  SPEAKER_03: When um I think I mentioned to you we had this project with uh uh applied materials, which is a Harvard Business Review uh uh uh article um that was um uh made it to their best of uh archive edition.
• 10:10
  SPEAKER_03: And uh it was as your data infrastructure ready for AI.
• 10:13
  SPEAKER_03: It was from uh this project we did where you know the CFO was saying, Why do we need taxonomy?
• 10:19
  SPEAKER_03: You know, why do we need uh you know, ontology and information architecture and all of this stuff?
• 10:25
  SPEAKER_03: And I said, uh, do you have a chart of accounts for your finance organization?
• 10:29
  SPEAKER_03: He said, Of course I do.
• 10:30
  SPEAKER_03: I said, why don't you just oh he said, why do we need that?
• 10:33
  SPEAKER_03: Why don't we just get Google?
• 10:34
  SPEAKER_03: And then I asked the question, I said, why don't you get rid of your org your uh chart of accounts and just get Google?
• 10:39
  SPEAKER_03: And they said, of course, that's ridiculous, right?
• 10:41
  SPEAKER_03: And I said, because a taxonomy is a chart of accounts for knowledge.
• 10:44
  SPEAKER_03: And uh the project was very successful because they tried doing it so many times before uh and failed without the core IA.
• 10:51
  SPEAKER_03: What else do organizations or you know executives and CEOs need to know about this related to generative AI?
• 10:58
  SPEAKER_03: There's the data side, and then there's the unstructured content side.
• 11:01
  SPEAKER_03: And how do you see knowledge graphs fitting in with that?
• 11:04
  SPEAKER_03: And you know, I'm sure you've heard of you know retrieval augmented uh generation.
• 11:10
  SPEAKER_03: Uh what are your thoughts there then?
• 11:12
  SPEAKER_02: So I I'm I'm we're we're I'm actively exploring that actually, and and I'm very passionate about um the fact that um graphs and and generative AI work very, very well together.
• 11:24
  SPEAKER_02: Um, you know, generative AI is is extremely eloquent and um but it it is prone to hallucination.
• 11:33
  SPEAKER_02: For those unfamiliar, that means that it be as in the course of being creative, it can it can tell very plausible lies.
• 11:42
  SPEAKER_02: Um and and of course, it can it can also unwittingly have been trained on information that that is misleading or inaccurate, which which an organization would not want to relay to it to others.
• 11:55
  SPEAKER_02: So the the fantastic aspect of a knowledge graph is that it can give structure and semantic context to to uh artificial intelligence, which which enables it to better understand what we mean when we say certain things.
• 12:10
  SPEAKER_02: Um, but also um it can provide a corporate corpora of knowledge that can be referred to and and also checked against.
• 12:18
  SPEAKER_02: So at the point of responding, a large language model, for example, can then be checked for hallucinations against facts and against reality.
• 12:27
  SPEAKER_02: And and and so it's you know, they work really beautifully together in that sense.
• 12:32
  SPEAKER_03: Yeah, no, and uh we're seeing more and more of that.
• 12:35
  SPEAKER_03: There's a lot of different ways of being able to kind of combine an information architecture and a knowledge architecture, you know, you can use that uh that uh large language model to process the query, uh, then retrieve the results based on that query, it's kind of normalized, and then process the results.
• 12:53
  SPEAKER_03: So it's conversational.
• 12:55
  SPEAKER_03: And uh, if you turn the temperature down to zero and you say only get the information from this knowledge source, um and and uh you know zero is the creativity factor, right?
• 13:06
  SPEAKER_03: One is very creative or you know, more prone to hallucinations, zero is less so, and say, tell it if you don't have the answer from this data source, say I don't know.
• 13:15
  SPEAKER_03: And that pretty much, you know, I don't want to say eliminates hallucinations, but I'm pretty sure it it vastly restricts them.
• 13:21
  SPEAKER_03: Uh, but the other thing that we had found out in our research was that adding metadata to the content and componentizing it effectively really did uh improve the accuracy and the precision.
• 13:32
  SPEAKER_03: Because the tricky thing about generative AI is it may give you an answer that's not really it's factually correct, it can be from that data source, but it's not phrased in a way where you can just look at that data source and go, is this the answer?
• 13:44
  SPEAKER_03: Like you can't take that quote search for it because it's already translated in some way.
• 13:49
  SPEAKER_03: So that's an interesting um challenge.
• 13:52
  SPEAKER_02: I absolutely agree.
• 13:53
  SPEAKER_02: And actually, you know, I so I'm a huge fan of the way you describe this.
• 13:57
  SPEAKER_02: Actually, you uh and and uh the when actually in interacting with people, I actually often uh do sort of quote um because I I think one of the the aspects of your book uh is is that you you really make it extraordinarily accessible to people to understand very complicated ideas and bring it to a business context all these business people can understand.
• 14:19
  SPEAKER_02: What book is that?
• 14:20
  SPEAKER_02: What book is that?
• 14:21
  SPEAKER_00: Sorry over your direct shoulder.
• 14:24
  SPEAKER_00: Oh, yeah, but yeah, go ahead.
• 14:26
  SPEAKER_02: I'm sorry, Ken, of course, I've got my copy right here.
• 14:30
  SPEAKER_02: Uh but actually, you know, interesting.
• 14:32
  SPEAKER_02: When I when I first joined BT, we have a really um involved architecture practice.
• 14:37
  SPEAKER_02: It's wonderful.
• 14:38
  SPEAKER_02: And uh and um one of one aspect of the culture that's so fantastic is we have book reviews.
• 14:44
  SPEAKER_02: And actually, I became aware of your book through a book review um just not long after it was it was um published.
• 14:51
  SPEAKER_02: And my my really brilliant colleague who doesn't specialize in data, I might add, he's now he's actually now in another organization, but he did a book review and he said, I really like this book, but but he said, from my perspective, I'm not convinced you know that organizations would be able to build these ontologies.
• 15:09
  SPEAKER_02: And and I said, Oh, tell me so.
• 15:11
  SPEAKER_02: I was very closely listening to his book review, and and I and I said, Well, look, that's really interesting you should say that because I'm here to help do exactly that.
• 15:21
  SPEAKER_02: And and um we had a really involved conversation and he joined the Graph Guild.
• 15:27
  SPEAKER_02: And we've we've been writing uh ontologies.
• 15:31
  SPEAKER_02: He was a very, very active, he was totally sold on it and in um uh as a result of the combination of the book and the combined conversation we'd had.
• 15:38
  SPEAKER_02: Nice, that's wonderful.
• 15:39
  SPEAKER_02: And now he's a proponent for knowledge graph everywhere, but the explainability, I think, is another aspect that ontologies that you bring out in the book, and and we we very much um utilize at BT is that ontologies are easy to understand for business people, right?
• 15:55
  SPEAKER_02: As well as AI.
• 15:56
  SPEAKER_03: Right, right, right, exactly.
• 15:59
  SPEAKER_00: Ben, what are your thoughts on on um utilizing generative models to generate graphs and generate you know these these types of you know technologies and stuff off of you know basic information?
• 16:11
  SPEAKER_02: Well, I think there's a great opportunity there, um, and but I think it has to be tempered.
• 16:17
  SPEAKER_02: So so I've I I have been um uh uh playing with that, and and Chris, you and I had a sort of conversation.
• 16:25
  SPEAKER_02: I've been doing a little bit of this in my free time where I was running podcasts not dissimilar to this, through um through a graph and extracting entities and and creating um graphs on the fly, which is you know great fun and and quite a powerful tool.
• 16:43
  SPEAKER_02: What I would say is that that um even writing code, you need where where it's it's non-RDF.
• 16:49
  SPEAKER_02: I I was using an RDF uh sort of um uh graph model to be able to do this, you still need to correct it heavily.
• 16:56
  SPEAKER_02: So you need to have the knowledge associated with um uh not only the structure and the content, but but also the way that's been codified to ensure that it's it is a sound ontology.
• 17:08
  SPEAKER_02: So I would say it's a great accelerator of these types of activities, but it's but from my perspective, the human in the loop is absolutely critical to ensure that you get the results that you want.
• 17:20
  SPEAKER_02: And that's a really interesting thing from an educational perspective, because people are saying this is having an impact on people being interested in learning how to code and learning these skills.
• 17:29
  SPEAKER_02: The thing is, we we're not going to be able to correct and advance the AI if if there's a lack of skills of people who can oversee and in and be that human in the loop.
• 17:42
  SPEAKER_00: So this is this is where um Seth and I, Ben and I, you know, I think um I certainly was more enamored by by him than probably the reverse, but we started to discuss the notion of Yes, I love you too.
• 17:54
  SPEAKER_03: Come on, you know.
• 17:55
  SPEAKER_00: I know, I know, no, no, no, no.
• 17:56
  SPEAKER_00: This this is you know mutually exclusive for sure.
• 17:59
  SPEAKER_00: Um but when we started to discuss this notion of being able to dynamically create something off of let's say a speech, right?
• 18:08
  SPEAKER_00: And we were talking about, you know, once once this does get created, you can definitely dissect and also come up with some really interesting relationships between what somebody speaks and maybe they believe that they're speaking in an open forum and it's public and everything else.
• 18:28
  SPEAKER_00: However, what they're also doing, because of their knowledge of the domain, is they're also outlining all of their IPs associated with it too.
• 18:38
  SPEAKER_00: And so we got into the question well, who owns that?
• 18:41
  SPEAKER_00: If you, if somebody graphs a TED talk, right, or graphs, let's say somebody else's who who you know, who cares what they're doing, you know, speaking, who owns that information, especially if you can disseminate that into some structure out of non-structure, right?
• 18:58
  SPEAKER_00: And so now we get into you know some of these more interesting scenarios around um at what point, how do you protect besides not speaking?
• 19:09
  SPEAKER_00: How do you protect some of this information?
• 19:11
  SPEAKER_00: Because you literally can you know dice decipher somebody what they're saying into a graph structured model, and now you have everything, you have it all, right?
• 19:21
  SPEAKER_00: Of where whatever the topic that they're talking about.
• 19:23
  SPEAKER_00: So that was one of the you know, the things that we jumped into was now who owns it?
• 19:28
  SPEAKER_00: If if I if I can graph it, who owns it?
• 19:31
  SPEAKER_02: It's a really interesting uh ethical piece because the I guess the thing is anybody can invest time in listening to a podcast and mapping out the entities.
• 19:41
  SPEAKER_02: It's a sort of data modeling activity.
• 19:43
  SPEAKER_02: It's it's it's too slow to do by hand to do at scale.
• 19:48
  SPEAKER_02: But when you can start to deploy graphs, then you can hoover this stuff up pretty quickly.
• 19:53
  SPEAKER_02: I don't think anybody would object.
• 19:56
  SPEAKER_02: I wouldn't, somebody mapping what I'm saying publicly, um, you know, in around graph, particularly, I'm sure they wouldn't be interested in my other opinions.
• 20:05
  SPEAKER_02: But um, and and actually saying, well, look, this is this is the kind of thing that Ben talks about.
• 20:10
  SPEAKER_02: Um, and and you know, I mean, that would be great flattery if they were manually doing that.
• 20:15
  SPEAKER_02: If they're automatically hoovering up my knowledge, well, again, I mean, somebody's gonna benefit from that, which is a great thing.
• 20:21
  SPEAKER_02: I wouldn't speak openly about something if I didn't want people to benefit from it.
• 20:25
  unknown: Right.
• 20:26
  SPEAKER_03: So, a quick question.
• 20:28
  SPEAKER_03: What what what concerns you?
• 20:30
  SPEAKER_03: Is there anything that kind of keeps you up at night in your in your role these days?
• 20:34
  SPEAKER_03: Or, you know, are there things you're worried about uh in trends and the industry and what you're doing, uh, implementation or buy-in or deployment, or what what what are the things that you're concerned about?
• 20:47
  SPEAKER_02: Wow.
• 20:48
  SPEAKER_02: Oh, so I mean, there's so many things that I'm always because as with any role, it's always about managing risk and managing reward.
• 20:56
  SPEAKER_02: Um, you know, I spent 22 years in financial services, so that's expectations and expectations, right?
• 21:02
  SPEAKER_02: And expectations, yes, absolutely.
• 21:04
  SPEAKER_02: So, I mean, some of the things I think are really interesting in the AI space, right?
• 21:08
  SPEAKER_02: Particularly if we're going to hone in on that area, I think the ethics, not dissimilar to the question that you, you know, you you just posed, Chris, which I think you know, I'm I'm intrigued by.
• 21:20
  SPEAKER_02: And of course, I don't think there are very clear answers or guidance, you know, across the industry or or uh uh uh from regulators on a lot of this stuff.
• 21:29
  SPEAKER_02: Things like who owns the the intellectual property of or or of generated images that have been obviously they're derivative by nature from from other sources and so on and so forth.
• 21:43
  SPEAKER_02: So I think the and and cross border uh discrepancies in regulation and things like that, I think that's a really interesting area to explore.
• 21:52
  SPEAKER_02: Um, I you know, I mean, I think there's some really great position papers from the from the UK government, among others, that that you know we're we're very interested.
• 22:00
  SPEAKER_02: And an inputting into as an organization.
• 22:03
  SPEAKER_02: So I think one of the that's that's really interesting to see where it's going to go and making sure that we're always holding ourselves to the highest ethical and regulatory and legal standard possible.
• 22:14
  SPEAKER_03: Yeah, I think that's such a concern of the C-suite is the risks, you know, the audit trails.
• 22:19
  SPEAKER_03: And again, some of the things we talked about can really mitigate that, right?
• 22:23
  SPEAKER_03: If you can say, here's our data, here's our data source, you know, you're only using the LLM to process a query, and then you're retrieving from your data, you have control, you have an audit trail, you're you eliminate the hallucinations for the most part with certain parameters, and then you're just processing the outcome outcome.
• 22:42
  SPEAKER_03: And there's a bunch of different ways of doing that.
• 22:44
  SPEAKER_03: So I think that's a really important way to kind of think about organization, how organizations can deploy this.
• 22:50
  SPEAKER_03: Can you talk a little bit about um a problem that your team has recently solved?
• 22:55
  SPEAKER_03: Like what's been an interesting um business problem, technical problem that that you've applied uh some of these principles to recently?
• 23:04
  SPEAKER_02: Fantastic.
• 23:05
  SPEAKER_02: So one of the things that we're super excited about is actually ontology based.
• 23:10
  SPEAKER_02: It's we're we've we've done a cloud data management capabilities assessment, um, which is a um uh a framework uh that's that's um from the non non-profit EDM Council that I helped contribute to to uh to writing and leading some of the subcapabilities.
• 23:30
  SPEAKER_02: Um and what we it brought together, sort of just to give background, it brought together about 300 professionals from from um about 100 of the world's largest financial services organizations primarily, as well as the hyperscalers um and all the all the major hyperscalers inputted into it.
• 23:47
  SPEAKER_02: And what we did was we created a C what we call the CDMC information model, which is an RDF model of all of the metadata necessary to complete CDMC certification, cloud data management capabilities.
• 24:00
  SPEAKER_02: And and it's it's so it's it provides a large ontology for metadata based on W3C standards like um uh DCAT, which is a cataloging standard, um, DQV, which is data quality, Provo, which is a lineage um ontology, and and so or standard from W3C.
• 24:23
  SPEAKER_02: And we're in the process of utilizing that to structure our internal metadata in a in a consistent standard.
• 24:31
  SPEAKER_02: So that's that's something that we're really excited about.
• 24:33
  SPEAKER_02: And it it allows us to be able to automate some of the data management activities.
• 24:38
  SPEAKER_02: It perfectly aligns with with with sort of um the AI-powered enterprise, I would say.
• 24:44
  SPEAKER_03: So so you know, before we you talked about there are some things that uh are inherently valuable that you have to do that don't necessarily have a direct ROI or measurable ROI.
• 24:54
  SPEAKER_03: Was is this one of does this fall into that classification, or is it has it been able to be tied directly to kind of a business outcome?
• 25:02
  SPEAKER_03: Or is it too much of an enabler?
• 25:05
  SPEAKER_02: Well, it's so that's a really, really fantastic question because actually building a digital twin in this in this sense, um, it can be used for multiple use cases, of course, each of which have um an ROI.
• 25:18
  SPEAKER_02: But I I I would recommend for anyone who wants to do that kind of thing, choose a couple of really big hitting initiatives that have the you know are going to benefit associated with this.
• 25:31
  SPEAKER_02: Look at the efficiency saves that you can make.
• 25:34
  SPEAKER_02: So, from our perspective, um things like we we're on a simplification um exercise across the organization.
• 25:42
  SPEAKER_02: This is um uh talked about across the industry, and um we can we can make that easier by tracking lineage from a graph perspective and tracking all the metadata associated with inter-system dependencies in this way.
• 25:59
  SPEAKER_02: So it's a very healthy and and and well-trodden use of knowledge graph um to do that.
• 26:05
  SPEAKER_03: And and you know, I I like to call the business uh capability that you're trying to develop the excuse case, right?
• 26:13
  SPEAKER_03: The use case that's the excuse for doing something that has uh applicability across lots of other domains, but we have to focus on one particular area like customer 360 or you know, or customer experience, or you know, uh top of the funnel, whatever it might be, or or compliance or you know, anything around security, but something that can have some kind of a some kind of measurable outcome and say this is what enabled that.
• 26:39
  SPEAKER_03: And you know, because data supports a process, process supports an outcome, outcome supports you know, the enterprise uh strategy, and uh and when we have that kind of connection and lineage, uh, it's much easier to say this was enabled by that.
• 26:54
  SPEAKER_03: That is what that capability depends on.
• 26:56
  SPEAKER_02: You see, I love that, and and the graph guild, you know, is very much how we're pulling together lots of different use cases from the organization that are building graphs, and we have a large number of graphs already.
• 27:07
  SPEAKER_02: And what we've created a community where we can leverage those to build out domains of an enterprise-wide knowledge graph, of which the meta model sort of structures that upper, what we would call an upper ontology.
• 27:21
  SPEAKER_02: So um, you know, it's a powerful means of of making sure that as an as people are delivering value for their particular customer-facing unit or corporate unit, as we call them internally, that we can leverage that for for uh cumulative benefit across an organization.
• 27:39
  SPEAKER_03: And I think building libraries of use cases that you maintain over time allows um the uh testing of your capabilities and and you know, use cases need to be unambiguous.
• 27:51
  SPEAKER_03: So you can so they're testable and verifiable.
• 27:54
  SPEAKER_03: And if you're building use cases for those knowledge graphs, you just build them up over time and you keep them as a corporate asset along with your your knowledge graph, uh your knowledge graphs and all of your data models.
• 28:05
  SPEAKER_00: That's perfect.
• 28:06
  SPEAKER_00: I know you to double-click on something, because you guys just you know kind of eloquently outlined it, but for you know, for everybody else, um it's it more along the lines of how do you put the governance and guardrails around the evolution of a graph and make sure that what is there is relevant and updated and not stale.
• 28:29
  SPEAKER_00: Right.
• 28:29
  SPEAKER_00: So maybe talk about that a bit too.
• 28:31
  SPEAKER_03: And not too and and not too uh uh academic or or theoretical, right?
• 28:35
  SPEAKER_00: Sure, absolutely reality based.
• 28:40
  SPEAKER_02: So that's one of the challenges for people who haven't done um much in the way of semantic modeling or on ontologies who are listening to the podcast.
• 28:48
  SPEAKER_02: I mean, one of the one of the the interesting things is it's so flexible that you can model anything and you can model the interconnectedness of everything and anything.
• 28:57
  SPEAKER_02: The the real challenge is to ensure that it's what you're modeling is relevant to the use case and relevant to the organization.
• 29:05
  SPEAKER_02: And there's there's you can infer additional connections as required, and depending on which which sort of graph uh tool you're using.
• 29:14
  SPEAKER_02: The the analogy that I always use is one that my dad uses.
• 29:18
  SPEAKER_02: He used to say there's uh and um well, he still does.
• 29:21
  SPEAKER_02: It takes two people to paint a great painting, one to paint it and the other one to tell them when to stop.
• 29:30
  SPEAKER_02: That's for me.
• 29:32
  SPEAKER_02: One of the challenges is is of course, you know, ontologists love to model, that's why they do it.
• 29:38
  SPEAKER_02: And um, so that that's one aspect.
• 29:42
  SPEAKER_03: Model and model and model, yes.
• 29:44
  SPEAKER_03: And there's two types of taxonomists in the world there's splitters and there's lumpers.
• 29:48
  SPEAKER_03: So if you put an ontologist with a splitter, you're you're you're you're bound to have a very long project.
• 29:56
  SPEAKER_03: Uh anyway.
• 29:57
  SPEAKER_02: You're so right, because it obviously if you overnormalize or over-generalize, then it's not as useful.
• 30:02
  SPEAKER_02: If you if you make it too granular, then then it's it's going to be a beast.
• 30:06
  SPEAKER_02: So, I mean, one of the other angles I think is really interesting is to actually analyze the usage um and and determine whether certain paths are redundant between nodes, and and then um, and actually, I think that could be something that AI would be really powerful at because you could then actually have a self-correcting ontology over time.
• 30:26
  SPEAKER_02: Um, and and and that could be quite powerful in terms of you know, understanding the weighting of nodes is right an AI use case.
• 30:34
  SPEAKER_03: It's uh and Chris, I think you hit the nail on the head around governance, right?
• 30:39
  SPEAKER_03: Because we need uh metrics, I mean we need feedback mechanisms, right?
• 30:43
  SPEAKER_03: Are people using this, right?
• 30:45
  SPEAKER_03: It are are we satisfying the use cases?
• 30:47
  SPEAKER_03: Are we breaking something when we make changes?
• 30:50
  SPEAKER_03: Because again, you know, if you if you go to in too too deep in too detailed uh a manner and you're splitting and splitting and splitting, you know, you're looking for finer and finer granularity.
• 31:00
  SPEAKER_03: Well, do you have a use case that backs it up?
• 31:02
  SPEAKER_03: And in fact, are people using that?
• 31:04
  SPEAKER_03: So that's where I think, as you know, as you're saying, Ben, you can have some self-correcting mechanisms by tying it to metrics and usage, and then being able to tie that to the business case and the use case and and and really again going from the very granular to the bigger, you know, macro scale.
• 31:21
  SPEAKER_03: I think that's a that's a great uh thing to think about, and that's what organizations need to get their minds around, especially in terms of governance, ongoing resource management, decision making, uh, course corrections, and uh and ROI, obviously.
• 31:36
  SPEAKER_02: I mean, it's surprising.
• 31:37
  SPEAKER_02: Oh, sorry, go ahead, Chris.
• 31:38
  SPEAKER_00: Oh, I just I was just uh all I was gonna say was I don't I'm not a fan of scoring because scoring sets a defined set of metrics for everything, not that, but there it's almost like there could be a semantic relevance score, you know, for your model and almost make up a notion of hey, here's in terms of explainability, here's what makes up our relevancy scores for these models.
• 32:00
  SPEAKER_00: And you know, because to your point, usage is king, right?
• 32:03
  SPEAKER_00: And if you're you know, and or or if it, you know, you start to see just a few areas of the synapse firing and that's it, then you've missed something, right?
• 32:12
  SPEAKER_00: So anyway, sorry, Ben, I didn't mean to catch you off.
• 32:14
  SPEAKER_02: No, I think that's fascinating.
• 32:16
  SPEAKER_02: And I'm I I totally agree.
• 32:18
  SPEAKER_02: Um, I guess uh I was also I'm also interested in because a lot of the graph packages have have originated in um often in uh academic academic backgrounds, that sometimes they they don't always have like versioning built in.
• 32:36
  SPEAKER_02: And and in terms of governance, obviously, you know, there are ways that one can supplement it with things like GitHub versioning and and other aspects, which I think is really important because you know, to the point about the org chart evolving, the graph and the ontology is going to evolve and and one needs to keep track of that kind of thing.
• 32:53
  SPEAKER_02: The other thing is is really interesting, is is actually making sure that the security aspects work well.
• 33:00
  SPEAKER_02: And so applying zero trust data entitlements type of uh thinking to an ontology and and actually breaking it up into lots of mini ontologies is one of the ways that you can that you can sort of um um help uh curate almost graphs that people some people are allowed to see, and some people it's off limits for.
• 33:20
  SPEAKER_03: You know, it's interesting.
• 33:21
  SPEAKER_03: You talk about mini ontologies, and I think there are a couple of ways of implementing that.
• 33:26
  SPEAKER_03: I think as long as those are connected and not independent.
• 33:30
  SPEAKER_03: And I've seen some companies break off pieces of their ontology and then go crazy, and it's like, ah no, you're doing it wrong.
• 33:37
  SPEAKER_03: It's like you're you're fragmenting things instead of doing that.
• 33:39
  SPEAKER_03: So that comes back into governance and change management.
• 33:42
  SPEAKER_02: I agree entirely.
• 33:43
  SPEAKER_02: It's a dangerous road to uh and and coordination is critical.
• 33:47
  SPEAKER_02: It's and and consistency, a common vocabulary and common standards is one of the ways that that you can um ensure that kind of interoperability.
• 33:56
  SPEAKER_02: But I I couldn't agree more if if they are, and actually that's one of the reasons why generally I favor RDF, because actually it does make things more uh interoperable um than label property graphs, although I am a big fan of label property graphs as well, I might add.
• 34:10
  SPEAKER_02: But the interoperability and common standards aspect is something that is a huge benefit.
• 34:15
  SPEAKER_03: Right.
• 34:17
  SPEAKER_03: Switch gears a little bit.
• 34:18
  SPEAKER_03: Um uh in and let's talk a little bit about you.
• 34:22
  SPEAKER_03: Uh like where where are you from and how'd you get to what you're doing?
• 34:25
  SPEAKER_03: Why don't you give us the world according according to Ben?
• 34:27
  SPEAKER_03: How did you get into what you're doing?
• 34:29
  SPEAKER_03: Fantastic.
• 34:29
  SPEAKER_03: Tell us Rebecca, tell us your origin story.
• 34:32
  SPEAKER_03: Were you a young boy thinking about structuring the world and organizing things?
• 34:38
  SPEAKER_03: And did you have all your tools lined up in a row?
• 34:42
  SPEAKER_00: All of his MMs and Skittles, as you can imagine, are all color coordinated by view, by you know, by by uh, you know, type, of course, right?
• 34:51
  SPEAKER_00: That's how they always were Legos, same thing.
• 34:53
  SPEAKER_00: It's probably his bedroom looked like you know a kaleidoscope of just organization.
• 34:59
  SPEAKER_02: I would I would like to say I was always like that.
• 35:02
  SPEAKER_02: I'm not convinced I was.
• 35:03
  SPEAKER_02: Um but I so I have a background in in engineering.
• 35:09
  SPEAKER_02: I was I I had a a master's degree in mechanical engineering um um from Bristol University and where I actually specialized in artificial intelligence, surprisingly, backpropagation neural networks.
• 35:22
  SPEAKER_02: Um and uh that was uh I I graduated back in 2000 and with a mart with his master's degree, and I was quite convinced that that was what I wanted to do with my life.
• 35:35
  SPEAKER_00: Um you still live by datum structures and stuff, right?
• 35:38
  SPEAKER_00: The rest of your life, that kind of thing, right?
• 35:41
  SPEAKER_02: I hadn't found data management at that point.
• 35:43
  SPEAKER_02: So I actually what I ended up doing was working in the financial services for for about 20 years plus, um, and I didn't touch artificial intelligence as much during that period.
• 35:54
  SPEAKER_02: I did a lot of architecture around data, the benefits of relational databases, the benefits of business architecture, um, and and all of this good stuff.
• 36:04
  SPEAKER_02: But it was actually it's really interesting because I'm sort of come full circle most recently because of the the need that um the interest in in artificial intelligence has on well-curated data.
• 36:18
  SPEAKER_02: The two go well very uh very, very well together.
• 36:23
  SPEAKER_02: So consequently, you know, after you know, I did some fascinating roles, by the way, in in between.
• 36:28
  SPEAKER_02: I I I did a lot on um financial crime uh investigations and intelligence in in financial services.
• 36:36
  SPEAKER_02: Um and of course, which is a very data.
• 36:38
  SPEAKER_03: But you found nothing there, but you found nothing there.
• 36:42
  SPEAKER_02: Well, it was it was a wonderful, wonderful role.
• 36:45
  SPEAKER_02: We were doing some real good in the world as well.
• 36:47
  SPEAKER_03: Yeah, target rich area, yeah.
• 36:50
  SPEAKER_02: Yes, and and of course, data intensive.
• 36:52
  SPEAKER_02: So it really played um well to my interests and and and of course, you know, all sorts of ontologies and AI being applied in that space, continuing uh across those those organizations.
• 37:05
  SPEAKER_02: So it's really exciting.
• 37:06
  SPEAKER_02: But yes, more more recently I got into you know, I was the I was the uh global head of business data architecture for the uh for HSBC's investment bank for for a few years, and and that was really a fantastic role as well.
• 37:19
  SPEAKER_02: I really enjoyed that and and got very involved in the EDM Council as mentioned through through that role in writing these common standards so and and and promoting them.
• 37:31
  SPEAKER_03: I'm gonna ask you, we only have a few minutes left, so I'm gonna ask you an interesting question.
• 37:36
  SPEAKER_03: If you could go back to the time that you were in uh school in college, uh, and you could do something differently, or you could tell yourself something, you know, what what would that be?
• 37:50
  SPEAKER_03: If you go back in time and give yourself advice, you know, right as you're getting out of college, what what would you tell yourself?
• 37:57
  SPEAKER_02: So I would say um follow your bliss, as Joseph Campbell would say from um who who discovered the the hero's journey.
• 38:06
  SPEAKER_02: Um and and my bliss, the bliss is is following challenges, it's not just sort of um uh some kind of hedonistic uh pursuit.
• 38:14
  SPEAKER_02: It's it's look for the things that really excite you, that that that are going to push you out of your comfort zone, that that's that um give your life some purpose and joy, and go for that.
• 38:27
  SPEAKER_02: And and you will always find that that um life aligns to support you in doing in doing that when you've got your passion is aligned with with your work.
• 38:37
  SPEAKER_03: I'll tell you, it you know, when when when my wife tells me I work too much, you I I like to say you mean I have fun too much.
• 38:46
  SPEAKER_03: I do enjoy it, I do love it.
• 38:49
  SPEAKER_03: Um and um let's see, uh, what do you do outside of work?
• 38:53
  SPEAKER_02: What what kind of uh things do you like to do for fun?
• 38:55
  SPEAKER_02: So I'm a huge um advocate of cycling and um actually um sports data in general.
• 39:02
  SPEAKER_02: So I I have my own little podcast with with a dear friend of mine, and we we we collect all sorts of sports and health and wellness data, and and and we we we um analyze that and talk about it and and help people benefit from trackable and wearable data.
• 39:18
  SPEAKER_02: Um, but yes, I I try and take as much exercise as I as as as uh my my schedule will allow.
• 39:25
  SPEAKER_03: That's great.
• 39:26
  SPEAKER_03: Well, listen, I really want to thank you for being here, Ben.
• 39:30
  SPEAKER_03: It's been wonderful, and I really appreciated your time and and your insights and expertise.
• 39:36
  SPEAKER_03: And we'll absolutely have to have you back.
• 39:38
  SPEAKER_02: Well, thank you so much.
• 39:40
  SPEAKER_02: It's such a pleasure, and keep doing the great work.
• 39:42
  SPEAKER_02: Um, you're inspiring an industry.
• 39:44
  SPEAKER_02: Well, that's wonderful.
• 39:46
  SPEAKER_00: It's a pleasure to know you, my friend.
• 39:47
  SPEAKER_00: Pleasure to know you.
• 39:48
  SPEAKER_03: Thank you.
• 39:49
  SPEAKER_03: And thank um, thank I want to thank our audience and uh hope you guys learned something today.
• 39:53
  SPEAKER_03: And uh have a terrific uh uh week, wherever or weekend, or wherever, whenever you're listening to uh to this.
• 40:00
  SPEAKER_03: And again, Ben, thank you so much for being with us.
• 40:05
  SPEAKER_01: Thank you for joining us on another deep dive into AI innovation.
• 40:09
  SPEAKER_01: Join us next time when we introduce another expert and discuss how to maximize AI in your world.
• 40:14
  SPEAKER_01: Early AI Podcast is sponsored by Early Information Science, CMS Wire, and Marketing AI Institute.
• 40:21
  SPEAKER_01: To learn more about us, visit Early.com.
• 40:24
  SPEAKER_01: That is A R L A Y.com.
• 40:28
  SPEAKER_01: See you next time.