Latest / The Adults In The AI Room / Earley AI Podcast Episode 66: Reengineering Knowledge for the AI Era

Transcript

Earley AI Podcast Episode 66: Reengineering Knowledge for the AI Era · The Adults In The AI Room

• 0:01
  SPEAKER_00: This is the Early AI Podcast hosted by Seth Early.
• 0:04
  SPEAKER_00: Join us as we delve deep into the passions, expertise, and experiences of thought leaders and practitioners to talk about what's possible with artificial intelligence.
• 0:14
  SPEAKER_00: The Early AI Podcast is sponsored by Early Information Science, your digital transformation journey with design and deployment of innovative technology solutions, as well as Vector.
• 0:26
  SPEAKER_00: Now enjoy the show.
• 0:29
  SPEAKER_01: Well, welcome to the Early AI Podcast.
• 0:32
  SPEAKER_01: I'm your host, Seth Early, and today we're going to be digging into some of the most important but often misunderstood aspects of AI, turning data into real action knowledge and being able to apply it to solving business problems in an efficient and effective way.
• 0:47
  SPEAKER_01: With Generative AI on the Rise, the conversation is no longer about having the data, but really about how your organizations understand your data and be able to drive better decisions.
• 0:57
  SPEAKER_01: We'll be unpacking knowledge engineering, the role of search and retrieval technologies, and how business can bridge the gap between information overload and intelligent action.
• 1:06
  SPEAKER_01: So today, joining us is someone that's been at the center of these conversations for nearly 30 years.
• 1:12
  SPEAKER_01: Industry analyst, advisor, and thought leader Tony Baer.
• 1:15
  SPEAKER_01: Thank you, Tony, for joining us.
• 1:16
  SPEAKER_01: Welcome to the show.
• 1:18
  SPEAKER_01: Well, thanks, Seth, for having me.
• 1:19
  SPEAKER_01: So Tony's been working with some of the biggest names in tech, including uh AWS, Oracle, and he helps organizations make a sense of the rapidly evolving trends in data, cloud, and analytics.
• 1:30
  SPEAKER_01: He also hosts his own podcast, shares insights regularly with LinkedIn, and is contributing voice to publications like SiliconANGLE.
• 1:36
  SPEAKER_01: So welcome.
• 1:37
  SPEAKER_01: And let's go ahead and jump in.
• 1:39
  SPEAKER_01: What I wanted to talk about is the role of knowledge engineering.
• 1:43
  SPEAKER_01: And of course, you and I met when I read your article about the need for knowledge engineers.
• 1:47
  SPEAKER_01: And that's uh music to my my ears because it's been the thing that I've been talking about for many, many years.
• 1:53
  SPEAKER_01: So talk about how that role uh as you see it has evolved and why is it very different than the prior uh classic versions of knowledge engineering where people focused on ontologies and maybe looked at this in the context of master data.
• 2:07
  SPEAKER_01: So give us your perspective on that, if you would.
• 2:10
  SPEAKER_02: Yeah, I knew I was going to pay the penalty for writing that.
• 2:12
  SPEAKER_02: But I was at a uh very interesting uh conference down in Austin, Data Date Texas.
• 2:16
  SPEAKER_02: It's been going on for about a dozen years, and it's a really good gathering.
• 2:20
  SPEAKER_02: It's a real good grassroots gathering when you kind of put your your ear to the ground as opposed to getting a message that's sanitized through technology vendors.
• 2:27
  SPEAKER_02: But anyway, and it's just something which just kind of jumped out at me.
• 2:30
  SPEAKER_02: It's not that I wasn't aware of this all these years, is that what do we know about the data that we're using?
• 2:35
  SPEAKER_02: What does that tell us about our businesses, our organizations?
• 2:39
  SPEAKER_02: And traditionally we've looked at it through a couple of different lenses, through whether, you know, you know, for instance, we might look at very structurally like what's the schema, you know, what tables are we using for this particular application.
• 2:50
  SPEAKER_02: We're taking very, or sometimes we zoom out to say a process view.
• 2:54
  SPEAKER_02: What I got from all this is that and I think maybe it's that, you know, I mean, there was an era of there was a there was about a decade of big data, but now AI has really brought the limitations of these approaches.
• 3:07
  SPEAKER_02: It's really made them, you know, very, you know, just you know, just almost uh too glaring.
• 3:14
  SPEAKER_02: Now to to answer your question, when you first asked me about knowledge engineering, I kind of thought, well, wait a second, I'm not looking, this is not like the semantic web is finally coming back.
• 3:24
  SPEAKER_02: But the thing is that uh I you know my background, I was both and a journalism major.
• 3:31
  SPEAKER_02: And so history is kind of like what happens, and journalism is what happens now.
• 3:35
  SPEAKER_02: And together, I feel it kind of gives you sort of a uh a view, uh a perspective on the why.
• 3:40
  SPEAKER_02: And I tend to see you know with the technology that history does tend to go around in in cycles, but each time we we come from a different starting point, a newer starting point that has some more wisdom for the past.
• 3:51
  SPEAKER_01: It's sort of a moving out to pagers are coming back.
• 3:54
  SPEAKER_01: I don't know if you're from um 30 rock, uh Dennis the Pager King.
• 3:58
  SPEAKER_01: Anyway, sorry.
• 3:58
  SPEAKER_01: I didn't Okay, you just destroyed my morning there.
• 4:02
  SPEAKER_01: I'm sorry, keep going.
• 4:03
  SPEAKER_01: You just destroyed my morning.
• 4:04
  SPEAKER_01: I know.
• 4:05
  SPEAKER_02: But anyways, so I was wasn't looking at uh, you know, that we're not looking at, let's say, you know, web ontology language, even though surprisingly RDF has made a comeback.
• 4:15
  SPEAKER_02: That actually kind of surprised me a little bit.
• 4:17
  SPEAKER_02: But I think the problem with knowledge engineering in the past was that it was very static, it was based on a static snapshot, which is if you think about it, it wasn't necessarily wrong at the time, it was the best that we could do.
• 4:29
  SPEAKER_02: I mean, just an example of all this of just you know kind of like how things change is I remember I was at an AI conference in Seattle back in 1987.
• 4:38
  SPEAKER_02: And I remember writing a piece.
• 4:40
  SPEAKER_02: I mean, like machine learning algorithms had been with us for like 40 or 50 years.
• 4:44
  SPEAKER_02: Right.
• 4:44
  SPEAKER_02: A lot of them are not particularly new.
• 4:49
  SPEAKER_02: But what my the headline I wrote coming out of it was AI is out, expert systems are in.
• 4:55
  SPEAKER_02: And the reason for that is because the limitation of technology at the time, the best that we could do was use rules, which at the time made sense.
• 5:02
  SPEAKER_02: Obviously, today, looking back, it's too static.
• 5:04
  SPEAKER_02: And I think that's the same thing that's applied with knowledge engineering, in that we're not looking at trying to take a snapshot.
• 5:10
  SPEAKER_02: We need something that's more dynamic and something that uses some of the newer technologies, such as graph and AI, to essentially kind of you know continually update this picture.
• 5:19
  SPEAKER_01: No, that makes a lot of sense.
• 5:21
  SPEAKER_01: And I think that you know, when you talk about how we we were approaching these problems, and we're trying to solve the same problems with different paradigms and different lines of thinking, but again, it it's it's understanding what that information infrastructure looks like and how those organizing principles are applied across unstructured and structured processes.
• 5:39
  SPEAKER_01: And I agree that it's it's very dynamic, it's it's very fluid.
• 5:43
  SPEAKER_01: You know, I talked about the information uh metabolism of the organization, right?
• 5:48
  SPEAKER_01: Because it's always about information flows.
• 5:50
  SPEAKER_01: And what happens is when we don't understand, you know, we have a lot of manual processes, or we have to do conversions or transformations or translations or integrations by that are manual, it adds friction to the process, right?
• 6:02
  SPEAKER_01: It slows things down and and inevitably across lots of different systems and processes.
• 6:08
  SPEAKER_01: When you think about, and a lot of it is always about trying to understand the context of data or the context of the analysis or the analytics or the metrics or whatever it might be, but but it's it's putting things into context, it's understanding why they're relevant, how they apply, getting the right information in context, and uh being able to use that because you know to solve whatever problem you're solving.
• 6:28
  SPEAKER_01: I like to say, and I I think uh I think this came from Peter Morvell, it's not about, or maybe it was uh uh I forgot who who was, might have been uh another luminary, but it's not about information overload, it's about filter failure, right?
• 6:41
  SPEAKER_01: We're we're not filtering the information, and the filter is the context, right?
• 6:45
  SPEAKER_01: What do I need in this context now to make this decision?
• 6:47
  SPEAKER_01: What data do I need?
• 6:48
  SPEAKER_01: What content do I need?
• 6:49
  SPEAKER_01: And so on.
• 6:50
  SPEAKER_01: So when we talk about the context and enterprise data, you know, what do you think that we're really referring to?
• 6:56
  SPEAKER_01: When we kind of look at that across the organization, why has this been a persistent gap in in traditional data management approaches?
• 7:05
  SPEAKER_01: How do you think about context with enterprise data?
• 7:08
  SPEAKER_02: It's really, I hate to boil it down to like a ridiculously simple elevator pitch.
• 7:13
  SPEAKER_02: It's really about the five views and something which I learned in actually journalism school, which is trying to get to the bottom of something.
• 7:20
  SPEAKER_02: And so it's who, what, when, where, why.
• 7:23
  SPEAKER_02: And that's there's also a an honorary, I guess, six W, which is how.
• 7:28
  SPEAKER_02: And if you answer all those questions, it really gives you really the big picture of you know what it is, why you're doing it, and in many cases, also how is very important because you may be making resource decisions, for instance, with that how.
• 7:42
  SPEAKER_02: Or you may be deciding on what model type of model or approach to training or or or selection of data.
• 7:48
  SPEAKER_02: So, like how is incredibly important, and how is also very much about data lineage.
• 7:54
  SPEAKER_02: So it's really answering what I'll call the five W's.
• 7:57
  SPEAKER_01: Yeah.
• 7:58
  SPEAKER_01: And and it's it's interesting because you know, when you when you think about context, it's it's always the it's it's kind of the so what, right?
• 8:06
  SPEAKER_01: If you have a piece of information, what what's what why does that matter, right?
• 8:09
  SPEAKER_01: What's the difference?
• 8:10
  SPEAKER_01: I like to say that around deliverables, I like to say that around assessments, I like to say that around anything, right?
• 8:16
  SPEAKER_01: Taking my cat to the vet, my dog to the vet, and they want to run a test.
• 8:19
  SPEAKER_01: And I say, what is this gonna give us, right?
• 8:22
  SPEAKER_01: What insight is it gonna give us?
• 8:23
  SPEAKER_01: Is it gonna impact the treatment plan?
• 8:24
  SPEAKER_01: No, don't do the test.
• 8:25
  SPEAKER_01: Uh, but it but it's always asking the so what question.
• 8:28
  SPEAKER_01: Why do I need this information?
• 8:29
  SPEAKER_01: What is this information going to do for me?
• 8:31
  SPEAKER_01: How is it going to inform a decision or a downstream process?
• 8:35
  SPEAKER_01: And we always have to ask and answer that question.
• 8:37
  SPEAKER_01: And that is really part of that.
• 8:39
  SPEAKER_01: It's it is it is the who, what, and where, it's it's the so what as well, uh, as part of that.
• 8:44
  SPEAKER_01: Now, when you kind of look at this from the perspective of new uh technologies or relatively new technologies and approaches around graph databases, and of course, we know machine learning has been around for a long time, but it's incarnation around generative AI.
• 8:57
  SPEAKER_01: Um, you know, how has that changed how we discover and model those relationships in enterprise data?
• 9:02
  SPEAKER_01: So we have these new emerging tools, we have things that uh becoming more popularized and more mainstream these days.
• 9:09
  SPEAKER_01: How has that changed how we kind of discover and model those relationships?
• 9:13
  SPEAKER_02: It's in several different ways.
• 9:15
  SPEAKER_02: I mean, one is that the relation with graph technology, graph database technology, it's finally given us a way to explicitly represent all the connections between data.
• 9:26
  SPEAKER_02: And actually, if you take a look at what Tim Berters-Lee was hoping with the semantic way, was making those semantic connections.
• 9:32
  SPEAKER_02: So graph itself is actually, you know, we finally found a way to persist this, to instantiate this, as opposed to you're writing lengthy, you know, lengthy, you know, documents.
• 9:42
  SPEAKER_02: Um and it's also provides a mechanism by which you know this can be updated because I think a key, a key problem, a key limitation of all the you know the the approaches that we've had in the past, it's not that they were wrong, it's that they were static because that was the technology that we had at the time.
• 9:58
  SPEAKER_02: Uh database can be updated traditionally with AI.
• 10:04
  SPEAKER_02: And we'll start with just machine learning.
• 10:06
  SPEAKER_02: Machine learning can start discovering the patterns on which data is being consumed.
• 10:11
  SPEAKER_02: And the fact is that this can scale the efforts of the best human modeler.
• 10:16
  SPEAKER_02: In terms of we can get it, you know, humans can probably get an 80% idea in terms of you know the 80-20 rule.
• 10:22
  SPEAKER_02: 20% of the flows account for 80% of how we're using data, but it's the other 20%, and it's the exception conditions.
• 10:29
  SPEAKER_02: The exception conditions that, for instance, that process modeling, BPM, could never capture, that data models could never capture.
• 10:37
  SPEAKER_02: But we see these exceptions, and machine learning can discover that and help us, you know, and help us show like this is actually really how this information is being consumed, and and for what processes and under what scenarios.
• 10:50
  SPEAKER_02: So, I mean, it's never going to be perfect, but it's gonna be a heck of a lot more um reliable than a static snapshot.
• 10:58
  SPEAKER_02: So that's you know, so that's a couple things that haven't even gotten to generative AI yet.
• 11:02
  SPEAKER_02: And generative AI, you know, provides a couple of key advantages.
• 11:06
  SPEAKER_02: One, of course, is that it provides, you know, a natural language interface, it provides a way of basically being able to converse with your data.
• 11:14
  SPEAKER_02: And I use, I like to use the metaphor of how uh Captain James T.
• 11:17
  SPEAKER_02: Kirk on Starship Enterprise communicated with you know with his data, you know, and I'm not talking about you know, actually, Lieutenant Data was basically the next generation, but with data on his computer, let's put it that way.
• 11:28
  SPEAKER_02: I don't want to mix metaphors there.
• 11:30
  SPEAKER_02: But the other thing is that it also brings unstructured data into all this.
• 11:36
  SPEAKER_02: And we briefly looked at this during the era of big data, but big data is really more about variably structured data.
• 11:42
  SPEAKER_02: But here with generative, we can actually find and basically we can introspect all that unstructured, whether it be text, image, audio.
• 11:50
  SPEAKER_02: There's so many things that are now accessible to us that we can now mine and basically understand what's being used, why it is, and all that sort of wonderful stuff.
• 11:59
  SPEAKER_01: That's a great, those are great points.
• 12:01
  SPEAKER_01: And I and I think that what's interesting about generative AI is in a way, it's it has some really great applications, but it's really the application that it can answer questions based on its understanding of the world, but it's better at answering questions about your data, right?
• 12:16
  SPEAKER_01: It's about better at answering questions when you give it your information, your content, your data, your practices, whatever it might be, your thoughts, your, you know, your line of thinking, and then processing that as opposed to giving an answer that is coming from its model of the world.
• 12:31
  SPEAKER_01: It wants to understand your model of the world, right?
• 12:34
  SPEAKER_01: Your differentiators, your knowledge, your data, your processes.
• 12:37
  SPEAKER_01: And that way we can we can uh use that most effectively.
• 12:41
  SPEAKER_01: It's almost a little bit of a red herring to say, have the AI, have the generative AI answer this.
• 12:45
  SPEAKER_01: It's no, have the generative AI understand your question and then retrieve that information and then make it more conversational, right?
• 12:52
  SPEAKER_01: And of course, there's lots of stuff we can do, can do in between, you know, in terms of processing the content and the information.
• 12:58
  SPEAKER_01: One of the things we've done is we've used uh our methodologies with large language models to kind of accelerate the process of building knowledge architectures, building metadata models, building vocabularies and taxonomies and even use cases.
• 13:11
  SPEAKER_01: And what that does is it says, here's here's the approach we're doing it, here are the definitions, now here's the data.
• 13:17
  SPEAKER_01: Now give me the answers about this data.
• 13:19
  SPEAKER_01: And that's what's really super exciting to us is that the ability to, because we do need that structure on that knowledge and that unstructured data in order for generative AI to retrieve it and to do something useful with that.
• 13:31
  SPEAKER_01: And and getting that structure to that content and that data is is is what's so time consuming and inefficient.
• 13:37
  SPEAKER_01: So when we're adding efficiencies there, that is very, very helpful.
• 13:40
  SPEAKER_02: Yeah, the couple of things I'm thinking from what you're saying there.
• 13:43
  SPEAKER_02: One is that this is actually in a way very evolutionary.
• 13:46
  SPEAKER_02: In terms of we've we've we've aimed towards a lot of these goals for a long time.
• 13:50
  SPEAKER_02: We've had business glossaries, for instance, and these are obviously you know, we we've had basically process models, we've had data models.
• 13:57
  SPEAKER_02: But the idea with AI is we can basically now turn this into something which is like a continual discovery process that can keep that sync and updated.
• 14:05
  SPEAKER_02: The other takeaway I'm getting from what you're just saying, and I think it's really important, and something which I've been saying for a long time, is that it really has to be on your data.
• 14:12
  SPEAKER_02: And you really have to get those guards.
• 14:14
  SPEAKER_02: I mean, we we don't want to get another example like you know, you we heard with Cursor, which basically where its chatbot came up with some totally fictitious policies, which number one was not just okay, hallucination, but it was also um this guess is some broader issues which get outside our discussion where cursor was not telling its customers that there's a chatbot, not a human.
• 14:34
  SPEAKER_02: That's a whole other issue.
• 14:36
  SPEAKER_02: But but either way, the importance of guardrails, and the thing is there are a number of different approaches that are emerging for that, whether it be graph rag, which I think finally is given graph databases kind of like a real reason to be.
• 14:48
  SPEAKER_02: Because until now, graph is the type of thing for very specialized use cases to uncover financial fraud, you know, really sort of edge in terms of edge and extreme use cases.
• 14:58
  SPEAKER_02: But I think graph rag is when you put the two together, really puts us in like this really makes a lot of sense.
• 15:04
  SPEAKER_02: And it can be, you know, that makes a lot of sense to, you know, every diggers, but it's also, of course, fine-tuning as well.
• 15:09
  SPEAKER_02: And where you apply Graph Rag and versus what, you know, do you basically just want to just keep doing a fresh query on the data versus do you want to keep updating the bottle?
• 15:16
  SPEAKER_02: And there are there are basically use cases for both, scenarios for both.
• 15:19
  SPEAKER_02: I think the important takeaways here are one, we do need guardrails, and there's not to be any one single silver bullet to doing that.
• 15:27
  SPEAKER_01: And it's a really good point.
• 15:28
  SPEAKER_01: And I and I think you know, connecting some of the dots, like people may have heard of ontologies and you know, heard of graphs, and you know, of course, they understand a little bit about retrieval augmented generation.
• 15:39
  SPEAKER_01: Uh, but the ontology is sort of the knowledge scaffolding of the organization, right?
• 15:43
  SPEAKER_01: It covers the is-ness and the aboutness.
• 15:45
  SPEAKER_01: What are these things, important big buckets and organizing principles for the organization?
• 15:50
  SPEAKER_01: And then what are the relationships between those concepts, right?
• 15:52
  SPEAKER_01: And so that's kind of the structure.
• 15:54
  SPEAKER_01: And then when we overlay that structure on our data and our content, we end up with a knowledge graph, which allows us to have different entry points.
• 16:01
  SPEAKER_01: So you can think of IMDB, the movie database, where you can look at uh a movie and then who the direct who's the director, and then what are the movies that he direct, and what actors were in that, and what actors that's how you play the six degrees of kept Kevin uh Bacon, right?
• 16:14
  SPEAKER_01: You look for those other uh movies and so on.
• 16:16
  SPEAKER_01: But that's a graph structure based on an ontology.
• 16:19
  SPEAKER_01: And and that what we're saying is organizations need to build that graph structure so that they can properly contextualize whatever their large language model is doing for them, right?
• 16:28
  SPEAKER_01: What their generative AI is doing, it's based on the reality of the business and based on the reality of it, it's data structures.
• 16:34
  SPEAKER_01: And and when we talk about retrieval augmented generation, that is they're all retrieval mechanisms, in my opinion, right?
• 16:40
  SPEAKER_01: When you start looking at these things, you know, from search to advanced virtual assistance, they're they're they're a mechanism to get it information.
• 16:47
  SPEAKER_01: And at the end of the day, retrieval augmented generation comes back to retrieval, which is search, which needs to have good data and content hygiene.
• 16:54
  SPEAKER_01: So it kind of comes back to all of that.
• 16:56
  SPEAKER_01: So when you think about informing it based on the graph uh data in the organization, that organ that data needs to be put into that structure, right?
• 17:04
  SPEAKER_01: We need to have those organizing principles and we need to be able to classify and and and tag that information so that we have the right context, right?
• 17:13
  SPEAKER_01: So all of that kind of looks at this ecosystem, even though it's evolving, you have the organizing principles which are less quickly changing, right?
• 17:21
  SPEAKER_01: But the data is obviously changing very quickly day to day as the organization operates.
• 17:26
  SPEAKER_02: And this really brings it all back home to our original topic.
• 17:29
  SPEAKER_02: We talk about knowledge engineers, right?
• 17:30
  SPEAKER_02: And what you were describing just now is really what I see is really the modern, you know, the role of the modern knowledge engineer.
• 17:36
  SPEAKER_02: The modern knowledge engineer is not necessarily the master of all these skills, but the knowledge engineer is someone who basically can help basically others in the organization, basically both domain specialists and data engineers put you know two and two together, or to use another overworked metaphor, make one and one plus one equals three.
• 17:54
  SPEAKER_02: In other words, it's knowing how to put together those knowledge graphs.
• 17:58
  SPEAKER_02: It's knowing how to basically where to look and maybe also having an idea in terms of what are some of the best tools in the trade, and also understanding, having a good ability to listen to domain experts.
• 18:10
  SPEAKER_02: Yes.
• 18:11
  SPEAKER_02: To understand what they are meaning here, which can kind of provide some, you know, some, I guess, sort of uh, you know, what's the word?
• 18:18
  SPEAKER_02: I guess of semantic assistance.
• 18:19
  SPEAKER_02: Uh you know, in other words, like here's here are really the here's how to really connect these these two nodes.
• 18:25
  SPEAKER_02: Yeah.
• 18:25
  SPEAKER_02: And these are like this is the nature of the connection.
• 18:27
  SPEAKER_01: Yeah, yeah.
• 18:28
  SPEAKER_01: And being able to help uh those different stakeholders, those subject matter expert experts, those domain experts, the data experts, the architecture experts, helping them understand this kind of common framework and being able to facilitate the development of that because we do need subject matter experts, we do need data architects, we do need representation from the business and from IT in order to do this.
• 18:49
  SPEAKER_01: IT is going to facilitate, but you know, business needs to own their data in a lot of ways, right?
• 18:54
  SPEAKER_01: They can't fix all of it, but they can certainly, you know, be responsible for the data integrity uh based on their processes, right?
• 19:01
  SPEAKER_01: And I think that knowledge engineer is that facilitator that helps to pull all those pieces together.
• 19:06
  SPEAKER_01: So I thought that's a great, great observation.
• 19:08
  SPEAKER_02: Yeah.
• 19:09
  SPEAKER_02: I really like the term facilitator.
• 19:11
  SPEAKER_02: Um because I mean that basically implies that you don't need to know these obsolete, basically semantic web languages.
• 19:18
  SPEAKER_02: You bring your skills into the modern day, into basically helping people basically bring people together, bring people and processes together.
• 19:25
  SPEAKER_01: Absolutely.
• 19:26
  SPEAKER_01: And when you look at some of the large efforts that we've seen in the past around business process uh management, master data management that are all trying to structure, you know, the data.
• 19:35
  SPEAKER_01: Like you said, it was more of a snapshot, it was more static.
• 19:38
  SPEAKER_01: But what are the lessons that we should take from those initiatives, especially, you know, when they fell short?
• 19:42
  SPEAKER_01: What are what other lessons should we take?
• 19:43
  SPEAKER_01: I know you talked about some of them, but if you kind of look back and say, well, you know, we put a lot of effort into those things.
• 19:49
  SPEAKER_01: There were some good outcomes, but what are the real uh takeaways?
• 19:52
  SPEAKER_02: Well, let's not get overly ambitious.
• 19:54
  SPEAKER_02: Let's not try and recreate that enterprise data model in this master enterprise knowledge graph.
• 19:59
  SPEAKER_02: I think what we look back and where BPM, business process management, and master data management were most successful, we're in very basically well-contained, well-bounded, high-value processes.
• 20:13
  SPEAKER_02: I think that's actually where we start.
• 20:16
  SPEAKER_02: Um, and and in some cases, the you know, the value comes into how frequently those processes are used.
• 20:22
  SPEAKER_02: So a good example of this is basically where you're doing with the customer experience.
• 20:27
  SPEAKER_02: Now, there might be other processes that have a quote unquote higher value, but but here's a very basic problem.
• 20:34
  SPEAKER_02: It's the 8020 rule.
• 20:36
  SPEAKER_02: So it's knowing how to really define the bounds of your problem and not try to boil the ocean.
• 20:40
  SPEAKER_01: Yeah, yeah.
• 20:41
  SPEAKER_01: And I think that makes a lot of sense.
• 20:43
  SPEAKER_01: You know, when we've done these kinds of projects in the past, it's taking a very specific domain and then looking at use cases that are measurable and that pertain to a particular process, such as the customer experience, right?
• 20:56
  SPEAKER_01: Maybe it's about customer support, maybe it's about a customer acquisition, maybe it's about wallet share, maybe it's about cross-sell or upsell.
• 21:02
  SPEAKER_01: But what you're doing is you're you're taking a broad view of the organization to make sure that domain model is adaptable enough and it doesn't miss anything big.
• 21:11
  SPEAKER_01: But then you're digging, you're diving deeply into that process in order to build out those pieces.
• 21:17
  SPEAKER_01: Otherwise, it is boiling the ocean and you can't do all of that.
• 21:20
  SPEAKER_01: And you really do need to kind of stack the deck to solve the problem that you're most interested in, right?
• 21:25
  SPEAKER_01: And that's okay, because we want to have what I call the excuse case, right?
• 21:28
  SPEAKER_01: The use case that gives you the excuse to do all these other things.
• 21:32
  SPEAKER_01: What's measurable, what's specific, you know, what can you where can you move the needle?
• 21:36
  SPEAKER_02: It's like a lesson we learned from the decade of big data, which is bigger is not always necessarily better.
• 21:42
  SPEAKER_01: Absolutely, right?
• 21:43
  SPEAKER_01: It and when you start looking at those efforts to kind of get everything in one place and get all the data.
• 21:48
  SPEAKER_01: And you know, we've had folks in the past say, oh, just point your AI to all of it.
• 21:52
  SPEAKER_01: And you know, AI needs context, right?
• 21:54
  SPEAKER_01: Just like P context.
• 21:55
  SPEAKER_01: So I wanted to kind of talk a little bit about so we talked about the skills of knowledge engineers.
• 21:59
  SPEAKER_01: I really Like the way we're kind of uh arriving at that in terms of the facilitator, right?
• 22:04
  SPEAKER_01: Understand those knowledge principles, they need to understand information architecture, they need to understand a lot of these other pieces, but really they are that facilitator to bring those parties together and to make sure that we're doing things that are sustainable and that are reflective of the present reality of the organization, but then have the ability to adapt.
• 22:23
  SPEAKER_01: And one of the things that's super important is building in those processes to get feedback, feedback loops, to do course corrections, to measure outcomes, to compare those outcomes with the expectations, you know, and to shampoo rinse.
• 22:35
  SPEAKER_01: Uh, is there anything else you'd like to add uh to that in terms of governance, in terms of change management, in terms of processes that uh that really uh uh have that knowledge engineering kind of mindset?
• 22:46
  SPEAKER_02: Well, I guess I mean, number one, uh just because something is old does not make it wrong.
• 22:53
  SPEAKER_02: And so I think there are a lot of disciplines that we've understood over the, you know, that we've built up over the years about what governance should be all about, what data stewardship should be all about.
• 23:01
  SPEAKER_02: I don't think we we don't necessarily throw that stuff out just because they were static, we just basically you know we just adapted.
• 23:08
  SPEAKER_02: We basically say, look, the principle was right.
• 23:10
  SPEAKER_02: I mean, I mean, we weren't wrong back when we were doing rule basically, you know, you know, the rules management system.
• 23:17
  SPEAKER_01: There's still a rules, there's still a need for rules.
• 23:19
  SPEAKER_01: I mean, there's still a need for rules-based classification and things like that.
• 23:22
  SPEAKER_01: There's still a need for the deterministic stuff, even when you have statistical outcomes.
• 23:26
  SPEAKER_02: Yeah, I mean, and the other thing too, also is again it's also more a broader thing I wrote uh on AI, which is that AI is not monolithic.
• 23:34
  SPEAKER_02: And of course, a lot of times when people talk about AI today, they're talking about generative AI.
• 23:38
  SPEAKER_02: Well, there are other there's classical machine learning.
• 23:41
  SPEAKER_02: The fact is fashioned AI, right?
• 23:43
  SPEAKER_02: Yeah, it's funny to be saying that.
• 23:45
  SPEAKER_02: It's really funny to be saying that in a way it is true because we've had with machine learning, we've had so those algorithms for 50 or 60 years, but we've only used them in the last 10.
• 23:53
  SPEAKER_01: That's right.
• 23:53
  SPEAKER_01: And it's so interesting, you know.
• 23:54
  SPEAKER_01: And when I wrote uh my book, The AI Powered Enterprise, right?
• 23:57
  SPEAKER_01: You know, I'm in my shoulder there, uh, it was, you know, it was before LLMs were really popular, it was back in 2020.
• 24:04
  SPEAKER_01: But every and when I first saw Large Angels Moss, I'm like, is my work no longer necessary, right?
• 24:09
  SPEAKER_01: And I very quickly uh disabused myself of that notion uh with my research to say, no, it's more important than ever.
• 24:16
  SPEAKER_01: Uh, even you know, someone had uh re uh had reposted one of my LinkedIn, one of my old articles to LinkedIn saying, you know, this is more important today than it was when it was written in 2016.
• 24:26
  SPEAKER_01: So a lot of this stuff is is valid, and everything in that book is still valid because it's this basic blocking and tackling of figuring out what those architectures should be, because those are the underpinnings of all of this stuff.
• 24:38
  SPEAKER_02: Yeah, it's under basic, it's uh it's important to basically understand what we you know, what our learnings over each of these cycles as we become more and more, let's say, knowledgeable or proficient.
• 24:48
  SPEAKER_02: And we also have this and the school of hard knots, the places where we basically have gotten our heads run over by a tractor.
• 24:54
  SPEAKER_02: We you know, we need to internalize all that and and and and take and look and use it going forward to say we've learned this, this is valuable.
• 25:02
  SPEAKER_02: I mean, it's just like, for instance, that you know, it's like my approach to my business, which is that I actually prefer to be proven wrong because when I'm proven wrong, that for me is a learning experience.
• 25:10
  SPEAKER_02: Yeah, I think we have to take that same attitude with regard to how we basically you know approach knowledge engineering.
• 25:16
  SPEAKER_01: I agree.
• 25:16
  SPEAKER_01: And by the way, you don't forget when your head is run over by a tractor.
• 25:19
  SPEAKER_01: You don't that is something that stays in your mind and you say, I'm not gonna do that again.
• 25:23
  SPEAKER_02: Yeah, look closely, you see all those tracks there.
• 25:26
  SPEAKER_01: They see the track marks.
• 25:27
  SPEAKER_01: But yeah, no, it it makes a lot of sense because again, we are building on that stuff.
• 25:31
  SPEAKER_01: And you always see this when you have new evolutions of technology that people coming to it don't remember or don't understand or don't know what's come in the past.
• 25:39
  SPEAKER_01: So, you know, when the web first came out, nobody thought about library science, right?
• 25:43
  SPEAKER_01: Oh, there are ways of organizing stuff that have been proven and refined for thousands of years, right?
• 25:48
  SPEAKER_01: Like those organizing principles are still valid, right?
• 25:51
  SPEAKER_01: And it and it's kind of interesting because kind of you always have these new technologies and new people saying, oh, this is all brand new, and they forget about the lessons at the best.
• 25:59
  SPEAKER_02: When I first got into this about my first job out of school, I was a tech writer for the Federal Solar Energy Research Program and was in a library part, you know, library, you know, since it was information, it was considered part of library science.
• 26:10
  SPEAKER_02: And uh that year I was a member of a professional organization called the International Society of Information Science.
• 26:16
  SPEAKER_02: Do not basically, you know, do not pronounce the acronym.
• 26:20
  SPEAKER_01: So, you know, it it's it is interesting because again, we we we see these evolutions and forget about the past and try to reinvent a lot of stuff.
• 26:29
  SPEAKER_01: And that's what we don't want to do here.
• 26:30
  SPEAKER_01: We want to say, take those lessons from process engineering, take those lessons from from uh master data management, take those lessons from you know the building the data architecture and the and the foundational principles of the organization and leverage those with these new technologies, right?
• 26:43
  SPEAKER_01: But don't forget about that past.
• 26:45
  SPEAKER_01: I wanted to, before we wrap, I wanted to get your thoughts on what are the trends and innovations you're most excited, you're personally most excited about in this convergence of data, of analytics, of AI.
• 26:57
  SPEAKER_01: And tell me a little bit what what gets you excited to see.
• 27:00
  SPEAKER_01: What are you seeing that is maybe uh not part of the hype cycle yet, but maybe it is early stages of hype cycle.
• 27:07
  SPEAKER_02: Well, maybe I'm biased because I'm only five foot five, but I like smaller, it's better, and especially when it comes to AI.
• 27:13
  SPEAKER_02: Right sizing models.
• 27:15
  SPEAKER_02: I think there's a lot of excitement about agents right now, some of it warranted and some of it way overblown.
• 27:21
  SPEAKER_02: I've actually, after coming back from Google next, I'm a little more I'm a little more optimistic about if we can base, you know, if we can approach agents on a fair, you know, take baby steps with them.
• 27:33
  SPEAKER_02: I think there are, you know, that basically if we have some task specifications, that's something which I think is very interesting.
• 27:38
  SPEAKER_02: There are new ways of training models and all this basically in terms of how we basically how we use this to get a better handle on our data.
• 27:48
  SPEAKER_02: That to me gets me up in the morning.
• 27:49
  SPEAKER_02: Besides bike riding, of course, in Tuscany.
• 27:52
  SPEAKER_02: That's great.
• 27:53
  SPEAKER_01: Uh and uh and where can people learn more about you?
• 27:56
  SPEAKER_01: I know you're on LinkedIn.
• 27:57
  SPEAKER_01: Uh your LinkedIn's profile is db insight.
• 27:59
  SPEAKER_01: But where else can people uh learn about you?
• 28:01
  SPEAKER_02: Well, my website is dbinsight.io.
• 28:03
  SPEAKER_02: Very I tend to do most of my uh you know, my output.
• 28:07
  SPEAKER_02: I'll do periodic output in LinkedIn.
• 28:09
  SPEAKER_02: I also write periodically if I have something to say in SiliconANGLE.
• 28:13
  SPEAKER_02: And I've also started up a new podcast with Matt Housley uh called It's About Data.
• 28:18
  SPEAKER_02: We're only about at the fourth episode, so I'm it's still a soft rollout, but our purpose here is basically, and it's kind of taking a little bit after what you're doing there, is like we're trying to go short.
• 28:28
  SPEAKER_02: In our case, it's about 15 minutes.
• 28:30
  SPEAKER_02: Um, but the idea is really just to more provoke it, to provoke discussion, not come up with the last word, as opposed to most podcast days tend to be like 45 or 60 minutes.
• 28:39
  SPEAKER_02: So that's something which I'm really excited about.
• 28:41
  SPEAKER_02: As I said, and we have a a piece coming out today on Oracle Multi-Cloud.
• 28:46
  SPEAKER_02: Um and so I mean I I'm very excited about what we're doing there.
• 28:50
  SPEAKER_02: So that's what's getting me up in the morning.
• 28:52
  SPEAKER_01: That's great.
• 28:53
  SPEAKER_01: So so thank you again for joining the uh the early AI podcast.
• 28:57
  SPEAKER_01: I really appreciate that that uh that insight and your expertise and kind of the historical way of looking at this.
• 29:03
  SPEAKER_01: And to our listeners, thank you for tuning in.
• 29:06
  SPEAKER_01: Um be sure to subscribe uh for more conversations and how to make AI practical, actionable, intelligent in your organization.
• 29:12
  SPEAKER_01: And until next time, we will see you.
• 29:15
  SPEAKER_01: Thanks very much, and thanks, Carolyn, for your work behind the scenes.
• 29:18
  SPEAKER_01: And Tony, take care.
• 29:19
  SPEAKER_01: We'll talk to you soon.
• 29:20
  SPEAKER_01: Bye-bye.
• 29:24
  SPEAKER_00: Thank you for joining us on another deep dive into AI innovation.
• 29:29
  SPEAKER_00: Tune in next time when we introduce another industry expert and discuss how to maximize AI in your world.
• 29:35
  SPEAKER_00: The Early AI podcast is sponsored by Early Information Science and Vector.
• 29:40
  SPEAKER_00: To learn more, visit early.com.
• 29:42
  SPEAKER_00: That's E A R L E Y dot com.
• 29:46
  SPEAKER_00: Thank you for tuning in.