 from downtown San Francisco. It's theCUBE. Covering IBM Chief Data Officer Strategy Summit 2018. Brought to you by IBM. Welcome back to San Francisco everybody. We're here covering the IBM CDO Strategy Summit. You're watching theCUBE, the leader in live tech coverage. Hashtag IBM CDO. Beth Smith is here. She's the general manager of Watson Data and AI at IBM. And of course, Interpol Bhandari who's the global chief data officer at IBM. Folks, welcome back to theCUBE. It's great to see you both again. Good to be back. So I love these shows. They're intimate, a practitioner. Really, they're absorbing everything. You're getting some good questions, some good back and forth. But Beth, share with us what you're hearing from customers. I mean, what matters for enterprises right now in the context of the cognitive enterprise, the AI enterprise? So you know, customers are looking at how did they get the benefit? They recognize that they've got a lot of valuable data. Data that we haven't always called data. You know, sometimes it's documents and journals and other kinds of really unstructured things. And they want to determine how can they get value from that? And they look out and compare it to maybe consumer things and recognize they don't have the same volume of that. So it's important for customers, how do they get started? And I would tell you that most of them start with a small project. They see value with that quickly. They then say, okay, how do we increment and grow from that? So, Inderpal, you had said, I think I got this right. This is your fourth CDO gig, right? You're not new to this rodeo. Were you the first healthcare CDO? Is that right? I was. Okay, now you got it all started. There were four of us at that time. Oh, okay. So fourth and four. Okay, I did get that right. So you obviously bring a lot of experience here. And one of the things you stressed today in your talk is you basically want to showcase IBM. So you're applying sort of data, enterprise data strategies to IBM and then you showcase that to your clients. Talk about that a little bit. Yeah, I mean, if you think about it, we are the quintessential complex enterprise, right? We're global. We're far-flung. We have literally thousands of products. We acquire companies. We move forward at, you know, at a global scale and also we're always competing at a global scale. So there literally is that complexity that enterprises face, which all our customers, you know, who are the large enterprises, have to also deal with it. So given all that, we felt that the best way to talk about an AI enterprise is to use ourselves as a showcase. All right. Okay, Beth, I got to ask you about Watson's Law. Okay, so we had Moore's Law. We all know what that is. Metcalfe's Law, the network effect. Watson's Law, and I have a noodling on this a little bit. You know, we're entering a new era, which I think is underscored by, you know, and names matter. We use a parlance in our industry, you know, whether it's cloud or big data or internet or whatever it is. And so we're trying to sort of figure out what this new era is like. What do you envision as Watson's Law? I'd love to have a little riff on that. So first of all, as we look at all those things and compare them back, they're all about opportunities to scale, right? And how things changed because of a new scaling effect. So I would argue that the one we're in now, which, you know, we like to call Watson's Law, the future will determine what it's actually called, is about scaling knowledge and applying knowledge. So it's about how to use AI, machine learning, applied to data, all forms of data, and turn that into knowledge. And that's what's going to separate people. And I would tell you that's also going to come back and give the incumbents an opportunity because the incumbents are strong in their industries, in their domains, they can leverage the data that they have, the knowledge and experience they have, and then use that for how do they disrupt and really become the disruptors of the future. So okay, so what about the math behind this? I'm kind of writing down some notes as you were talking. So my version of Watson's Law, I love your comment, is innovation in the future, current, is going to be a function of the data, your ability to apply AI or cognitive to that data, and then your ability to, to your point, scale the cloud economics. Does that make sense to you guys? Is that where innovation is going to come? It's true, but I have to go back to this point, Dave, of knowledge. So I think you take data and you take AI or machine learning, and those are sort of your ingredients. The scaling factor is going to be on knowledge, and how does that ultimately get applied? Cloud gives us, cloud again gives us an ingredient, if you will, that can be applied to it, but the thing that'll make the difference on it, just like networking was in the past and opened up opportunities around the internet, this, that, and the other, will be how folks use knowledge. It's almost like you could think of it as a learning era and how that's not just going to be about individuals, but about companies and business models, et cetera. So the, the knowledge comes from applying cognitive to the data. Yeah, that's right. And then being able to scale it. That's right. Okay, and then, into Paul, how do I address the access issue, right? I've got most, many, if not most incumbents, data are in silos. The marketing department has data, the sales department has data, the customer service department has data. How do you, as a CDO, address that challenge? Well, what you've got to do is use the technology to actually help you address that challenge. So, building data lakes is a good start for both structured and unstructured data, where you bring data that's traditionally been siloed together, but that's not always possible. Sometimes you have to let the data be where they are, but you at least have to have a catalog that tells you where all the data is, so that an intelligence system can then reason about that when working with somebody on a particular use case, actually help them find that data, and help them apply it and use it. So that's a metadata challenge, correct? It's a metadata challenge in the AI world, because the metadata challenge has always been there, but now you have the potential to apply AI to not just create metadata, but then also to apply it effectively to help business users and subject matter experts who are not data experts, find the right data and work it. You guys make a big deal about automating some of this stuff up front at the point of creation or use automating, let's, classification's a good example. How are you solving that problem from a technology perspective? Well, some of our core Watson capabilities are all about classification, and then customers use that, it can be what I will call a simple use case of email classification and routing. We have a client in France that has 350,000 emails a week, and they use Watson for that level of classification. You look at all sorts of different kinds of ticketing things, you look at AI assistance, and it comes down to how do you really understand what the intent is here, and I believe classification is one of the fundamental capabilities in the whole thing. Yeah, it's been a problem that we've been trying to solve in this industry for a while, kind of pre-AI, and you really, there's not a lot you can do if you don't have good classification. If you can't automate it, then you can't scale at another point. That's right. So from a classification standpoint, I mean it's a fundamental, always been a fundamental problem. Machines have gotten much better at it, with AI systems and so forth, but there's also one aspect that's quite interesting, which is now you have open loop systems. So you're not just classifying based on data that was historically present, and so in that sense, you're confined to always repeat your mistakes and so forth, right? You hear about hedge funds that implode because their models are no longer applicable because there's a black swan event. Now with the AI systems, you have the opportunity to tap the real-time events as they're going and actually apply that to the classification as well. So when Beth talks about the different APIs that we have available to do classification, we also have NLP APIs that allow you to bring to bear this additional stuff that's going on and go from a closed loop classification to an open loop one. So I want to ask you about the black box problem. If you think about AI, I was saying this in my intro, I can, I know when I see a dog, but if I have to describe how I actually came to that conclusion, it's actually quite difficult to do and computers can show me there's a dog, or I joked in Silicon Valley, if any of you guys watched that show, Silicon Valley, hot dog, or not, right? So your prescription at IBM is to make a white box, open that up, explain to people, which I think is vitally important because when line of business people get in the room, they go, how'd you get to that answer? And then it's going to be, it's going to actually slow you down if you have arguments, but how do you actually solve that black box problem? It's a much harder problem, obviously, right? But there are a whole host of reasons as to why you should look at it that way. One is we think it's just good business practice because when people are making business decisions, they're not going to comply unless they really understand it. From my previous stint at IBM when I was working with the coaches of the NBA, they would not believe what patterns were being put forward to them until such time as we tied it to the video that showed what was actually going on, right? So it's that same aspect in terms of being able to explain it. But there's also fundamentally more important reasons as well. You mentioned the example of looking at a dog and saying that's a dog, but not being able to describe it. AI systems have that same issue. Not only that, what we are finding is that you can take an AI system and you can just tweak a little bit of the data so that instead of recognizing it as a dog now, it's completely fooled and it will recognize it as a rifle or something like that, right? Those are adversarial examples. So we think that taking this white box approach sets us up not just tactically and from a business standpoint, but also strategically from a technical standpoint if a system is able to explain it, describe it and really present its reasoning, it's not going to be fooled that easily either. That's so some of the themes that we hear from IBM. You own your own data. I mean, all this, the Facebook blowback has actually been an unbelievable tailwind for that meme. And most of the data in the world is not publicly searchable. So build on those themes and talk about how IBM is helping its customers take advantage of those two dynamics. So they kind of go hand in hand in the sense that because customers have most of the data behind their firewall, if you will, within their business walls, it means it's unlikely that it's annotated and labeled and used for a lot of these systems. So we're focusing on how do we put together techniques to allow systems to learn more with less data? So that goes hand in hand with that. That's also back to the point of protecting your data because as we protect your data, you and your competitor, we cannot mix that data together to improve the base models that are a part of it. So therefore we have to do techniques that allow you to learn more with less data. One of the simplest things is around the customization. I mean, we're coming up on two years since we provided the capability to do custom models on top of visual recognition, Watson visual recognition. The other guys have been bragging about it in the last four to five months. We've been doing it in production with clients who be two years in July. So you say, okay, well, what's that about? We can end up training a base model that understands some of the basics around visual type things like your dog example and some other things, but then give you the tools to very quickly and easily create your custom one that now allows you to better understand equipment that may be natural to you or how it's all installed or agricultural things or rust on a cell phone tower or whatever it may be. I think that's another example of how this all comes about to say that's the part that's important to you as a company, that's part that has to be protected. That also has to be able to learn with you only spending a few days and a few examples to train it, not millions and billions. And that base layer is IBM IP, but the top layer is client IP. And you're guaranteeing the clients that my IP won't seep into my competitors. So our architecture is one that separates that. We have hybrid models as a part of it and that piece that, as you said, is the client piece is separate from the rest of it. We also do it in such a way that you could see, there could be a middle layer in there as well. Let's call it industry or licensed data. So maybe it comes from a third party, it's not owned by the client, but it's something that's, again, licensed, not public as a part of it. That's a part of what the architecture is. And you guys, we saw the block diagrams in there. You're putting together solutions for clients and it's a combination of your enterprise data architecture and you actually have hardware and software components that you can bring to bear here, right? Can you describe that a little bit? Go ahead, it's your implementation. Yeah, so we've got, again, we're the perfect example of a large enterprise, right? So there's significant on-prem implementations, there's private cloud implementations, there's public cloud implementations. You've got to bridge all that and do it in a way that makes it seamless and easy for an enterprise to adopt. So we've worked through all that stuff. So we've got, we've learned things the hard way about, well, if you're truly going to do an AI data lake, you better have it on flash. For that reason, we have it on flash, on-prem, but also on the cloud, our storage is on flash. And so we're able to make those types of decisions so that we've learned through this, we want to share that with our clients. How do you involve deep learning in this mix? Well, it's got to be proximate to your data lake so that the servers can get to all that data and run literally thousands and thousands of experiments in time that it's going to be useful for the decision. So, all those hard learnings, we are packaging that in the form of these showcases. We're bringing that forward. Right now it's around hybrid cloud and the bridge so that because we want to talk about everything and then going forward, it's all public cloud. As we leverage that for the elasticity and the compute. Well, IBM, if you can do it there, you can do it anywhere, because it is a highly complex organization. So it's been, what, two years in for you now? Two in... Two years and a little over two years. You're making a lot of progress and I can see the practitioners eating this stuff up. And that's got to feel good. I mean, you're having an impact, obviously. It absolutely feels very good. And, you know, I've always, in fact, I kind of believe this coming into IBM that this is a company that has not only a number of products that are put into this space, but actually the framework to create an AI enterprise. Right? These are not like disparate products. These are all going towards creating an AI enterprise. And I think if you look across our portfolio of products and then you kind of map that back to our showcases, you'll see that come to life in a very tangible way that, yes, if you truly want to create an AI enterprise that IBM's the place to be because they've got the answers across all the dimensions of the problem, as opposed to just one specific dimension, like let's say a data mining algorithm. Right. Something machine learning, that's basically it. When we cover the full gamut, and you have to. Otherwise, you can't really create an AI enterprise. And the portfolio obviously coming together, IBM huge ambitions with Watson, and everybody's familiar with the ads, and so you've done a great job of getting that top of mind, but you're really starting to work with clients to implement this stuff. You had just, I know we got to end here, but you had thrown out a stat, 85% of executive CAI as a competitive advantage, but only 20% can use it at scale. So there's still that big gap. You're obviously trying to close that gap. Yeah, so and I would correct it, only 20% of them are using it at scale. I think, Dave, it's a part of how do they get started? And I think that comes back to the fact that it shouldn't be a large, transformational, scary multi-year project. It is about taking small things, starting with two or three or five use cases and growing and scaling that way. And that's what our successful customers are doing. We give it to them in a way that they can use it directly or we leverage it within a number of solutions like cybersecurity, like risk and compliance for financial services, like healthcare, that they can use it in those solution areas as well. Great, well guys, thanks so much for coming to theCUBE, sharing this story. We love coming to these events. As I say, you see the practitioners. This is a board level discussion and these guys are right in it. Good to see you guys, thank you. You too, thank you. All right, you're welcome. All right, keep it right there, buddy. We'll be back with our next guest. You're watching theCUBE live from the IBM Chief Data Officer Strategy Summit in San Francisco. We'll be right back.