(upbeat music) >> From our studios in the heart of Silicon Valley, Palo Alto, California, this is a Cube Conversation. >> Hi, I'm Peter Burris, and welcome to another Cube Conversation from our spectacular studios here in beautiful Palo Alto, California. As enterprises move forward with transformation plans that include AI, that include new classes of data-first applications, and very importantly, new storage resources necessary to support all that, you have to ask the question. Where is the talent gonna come from? How are we gonna organize that talent? What tasks are gonna be especially important? Ultimately, what will be the roles and responsibility of storage and AI administrators in the future? And to have that conversation, we've got my colleague, Jim Kobielus, from Wikibon, here to talk about this. Jim, welcome to the Cube. >> Hello, Peter, nice to be here. >> So let's start with the first question of, what is it about AI that's catalyzing these new classes of roles? >> What's catalyzing new classes of roles is regards AI is the fact that AI is not just one thing. AI is a range of applications and approaches that, to be done right, has to be built out into an organization with a specialization. A fairly fine degree of specialization between different things we've all heard, people we've all heard about data scientists, data engineers, you also need various analysts, subject matter experts. But you also need people like, when you think of normally, people who are masters of natural language processing and conversational UIs, robotics, a lot of other specialties come in depending on a type of AI project you're undertaking. It can be a fairly complex engineering process to get this stuff built and trained and working right in the real world. >> So Jim, as I like to talk about, I talk about that we are in the midst of a significant inflection point in the industry. We can characterize the first fifty years as known process, unknown technology. And by that I mean, we knew we were gonna do accounting, we knew we were gonna do finance, we knew we were gonna do HR. We didn't know what platform it was gonna be on. To now, going forward, it's known technology. We know it's gonna be cloud, and things that provide cloud-like services. We may not know the specifics of the technology, but we know what it's generally gonna look like. But more importantly, it's unknown process. Somewhere paradoxically, in the old world, the storage administrator could, based on the process, and the classes of data required to support that process, do a lot of optimization down at the physical schema layer in storage. As we move forward to these data first-world though, it opens up unknown, unexpected emergent characteristics of how the business uses data, and that's gotta put significant net new pressure on the expertise required to administer storage resources. Can you talk a little bit about that? >> Yeah, because the unknown process, I love to talk about conversational UIs which are in everything, and AI is the magic behind those, and the impact that that, for example, that application has on storage. To do conversational UI right, to do it at all really, you need natural language processing, and to do natural language processing, you need statistical models, machine learning. You need lots of textual unstructured data and so forth, and that requires a fair amount of storage, 'cause that's a lot of data, and it just takes up petabytes and beyond depending on the level to which you want your natural language processing model to be able not just to understand it, but to generate a text, a speech, with a highly read averse and millitude. So when we look at the new world of AI, natural language processing actually has been the killer app for AI ever since AI was coined as a term in the 1950s. What we're seeing is that in the last 10 plus years natural language processing, textual data in very large databases in the cloud, really catalyzed this whole big data revolution. And a big part of the big data revolution is this notion of embedded analytics, or in database analytics, and the in database analytics has gotten ever more parallelized over time to the point where you have these massive hood dupe clusters and object storage clusters and so forth. >> Jim, let me interrupt you. But doesn't that suggest ultimately that the storage administrator which used to have to know a lot about the complexities of the underlying physical device, especially when that device was spinning. That person now has to know more about the data services, how those data services are being consumed, in service too, a broader set of application issues. The storage administrators have to get more knowledgeable about how data and business come together so they can do a better job of providing and administering and optimizing data services. Is that kind of a decent summary of where the maturity's gonna be in a few years? Yeah, it comes down to storage administrators now have to master storage virtualization which is many different types of storage device or storage platforms, or database platforms, and these terms of blurring into each other have to play together within fairly complex hybrid data environments, to be able to drive an application such as conversational UI inside I would say a chat bot that's a front end to an ecommerce system that is built on being able to in real time, do heavy transactional and analytical processing. What I've just laid out, at the highest level, is an architecture that's mainstream in the database world whether you have RBBMs's, you have Calver data storage, you have distributed file stores like Hidupe File Store and so forth, and you have stream computing with some degree occasionally of persistence in a low latency one. All these different storage and data management approaches have to play together within a broader data or storage virtualization environment that very much has to be geared to what the business needs. You don't just plug in new databases as a data engineer 'cause you like them or you like the opensource project that they come out of. You do it because each of those is optimized for a particular type of data that's associated with a particular type of data source, which is also associated with a particular type of downstream usage of that data in a given business application. So the storage administrator, they have to make sure this entire storage virtualization architecture is able to align with the sources and the uses of the data, and they need a high level virtualization abstraction layer that enables, the people who build the data analytic applications, the data scientists and so forth, be able to find the data they need to put into their machine learning models to drive the magic of AI. So there has to be a greater degree of alignment with the business inside of storage, people who define the storage architecture now than ever before. Understanding the underlying formatting of the data on the spinning, well it used to be spinning disk. In the underlying of say, flash storage environment, becomes less important in terms of its relationship to the business. >> So, but I would suggest that, you tell me I got this right, it does seem to me as though as we move forward that there's going to be a tighter correspondence between what the business needs, and what the actual storage can deliver. Now we've talked about the skills gap in the data-driven world. In security, in AI, in data science, etcetera. It seems also that as we try to do a better job of matching, that we're gonna see a skills gap in how data services are conceived and operate within a business. And some of that can be filled in by new products, some of that can be filled in by new tooling, but talk to me a little bit about how partners, companies that might of historically been associated with just moving pieces of hardware, and very focused on that kind of a transaction, can step up or are going to have to step up to be more cognizant and aware, and able to participate in the process of closing that skills gap between what we need storage to do, and what the business outcomes are. >> Yeah, lemme go back to AI and lemme go back to sort of a leading edge AI use case that is everywhere, and it's gonna increasingly dominate the industrial world, which is robotics. I mean, it's not just factory floor robotics, robotics is being embedded in so many different business environments and consumer environments now, and the magic of robotics going forward is all about AI, the AI, the brains that drive. Autonomous vehicle are a type of robot, so are drones and so forth. And so, in terms of where partners come in, not that many enterprises have robotic specialists just sort of on staff on call. This is a specialized discipline that you would bring in. There's any number of robotics friends you'd bring in to a robotics related project that involves AI. You might have your staff of data scientists building out the AI. They need to call in the robotics partners, contractors, whoever they are, to provide that piece of the overall, as it were, application scenario. You might also need to call out a separate set of partners who are masters of building conversational UIs so that, to enable human beings to interact more naturally with the robots that are being built and infused with intelligence through AI. So what I'm getting is, I'm starting a sketch of an ecosystem where more companies have internal AI, or I should say data science expertise. They may increasingly call on robotics partners to provide that piece for drone related projects or whatnot. They may call on conversational UI or natural language processing partners to handle that piece. More of the UI, the interactions and so forth, they're like other specialties that are brought into these projects based on the extent to which geospatial intelligence is required. You might bring in, you know, mapping firms and so forth. Partners will provide various pieces of the overall application ecosystem needed to stand up a working AI application. Now the storage becomes important, because every one of the components in these increasingly physical, based in say robotics or drone projects or autonomous vehicles, will have its own local storage. It'll persist data locally for lots of good reasons. A, it acquires the data there 'cause it's got sensors. B, it needs to cache history in terms of over the last hour or day or last month worth of data for lots of reasons in terms of doing trend analysis. So the storage architecture that needs to be built out needs to span all these disparate assets, some of which are provided by partners, the suppliers, some of which are provided in house. So how do you build out a storage architecture that has enough flexibility to bring in more of these edge storage requirements in a unified way so that the clouds and the gateways and the edge devices and all of their disparate flash and spinning disk and what not, they all play together as a unified storage resource, while partners have to be part of that overall setting of that architecture in terms of providing some degree of, look them into the process by which the end to end storage requirements will be sort of mapped out. As you're starting to build out these more complex projects. >> Alright Jim, lemme stop you there. So it sounds as though we are in a position where there is an enormous opportunity for partners to establish a presence as helping their customers better associate what the business needs with application characteristics and the data requirements for those applications, and make it simpler and faster to associate storage resources to those so that they can accelerate these outcomes. Jim Kobielus of Wikibon, thanks very much once again for being on The Cube. >> It's a pleasure. >> And once again, I'm Peter Burris, and this has been another Cube conversation, until next time.

(jazzy techno music) >> From our studios, in the heart of Silicon Valley, Palo Alto, California, this is a Cube Conversation. >> Hi, I'm Peter Burris, and welcome to another Cube Conversation. Everybody talks about AI, and how it's going to dramatically alter both the nature and the productivity of different classes of business outcomes, and it's clear that we're on a variety of different vector's and road map's to achieve that. One of the most important conversations is the roll that AI's gonna play within the IT organization, within digital operations. To improve the productivity of the resources that we have put in place to make these broader, more complex business outcomes possible and operationally efficient. One of the key places where this is gonna be important is in storage itself. How will AI improve the productivity, both from a cost stand point, but even more importantly, from the amount of work that storage resources can do standpoint. Now, to have that conversation we've got Jim Kobielus, my colleague from Wikibon, to talk about his vision of how AI technology, Jim you're the, our key AI guy. How AI technologies will be embedded in storage services, data services, and the new classes of products, that are gonna make possible these new types of data driven, AI driven outcomes. Jim, welcome back to the cube. >> Thanks, Peter. >> All right, so let's start Jim. As you think about it, what is it about AI that makes it relevant to improving storage productivity? >> Well, AI is a broad term, but let me net it out to the core of what AI's all about. Core AI is what is called machine learning, and machine learning is being able to find patterns in the data using algorithmic methods in a way that can be automated, and also in a way that humans, mortal humans can't usually do. In other words you have complex datasets. Machine learning is very good at doing such things as looking for anomalies, looking for trends, looking for blotter statistical patterns among (mumbles) elements within a broader dataset. So, when you talk about storage resources, and you talk about storage resources in (mumbles) environment, you have many tables, and you have records, and you have indices, keys and so forth. >> Logs. >> Yeah, yeah. You have, yeah. So, when you have a lot of entities in various, and quite often complex relationships, that when a storage exists, if you will, a number of things to persist data, you know, as a historical artifact, but also storage exists to facilitate queries, and access to the data. To answer questions, that's what analytics is all about. If you can shorten the path that a query takes to assemble the relevant tables, records, and so forth, and deliver a result back to whoever posts the query, then storage becomes evermore efficient in serving the end user. The more complex your storage resources get, it can be across different servers, different clusters, different clouds, it can be highly distributed across the internet of things, and what not. The more complex your storage architecture and distributed becomes, the more critically you need machine learning to be able to detect the high level patterns, to be able to identify, you know at any point in time, what is the path that a given query might take to be able to respond in real time to some kind of requirement from a business user who's sitting there at a dashboard trying to call up some complex metrics. So, machine learnings able to not only identify the current patterns within distributed datasets, but also to predicting models that are built in. That's what AI often does, is predict the models. To identify the predict, how if under various scenarios, if the data were placed in different storage volumes, or devices, or cached here. Now there distributed, and (mumbles) in a particular way. How you might be able to speed up queries. So, machine learning is increasingly used in storage architectures to identify, A the current patterns, B to identify query paths, C to predict a recommended, and automatically move data around so that the performance, whether it be queries, or reporting, or data transfers, or all that. So, the performance of that data transaction or analytic, is as good or as fast as it can possibly be. >> More predictable, right? >> No, here you are. Automate it, predictably. So that humans don't have to muck around with query plans, and so forth. That the architecture, the infrastructure takes care of that problem. That's why these capabilities are autonomous operations, they're built into things like Oracle database. That's just the way database computing has to be done. There's less of a need for human data engineers to do that. I think human data engineers everywhere are saying hallelujah, that is way too complex for us, especially in the era of distributed edge computing. We can't do that in a finite amount of time. Let the infrastructure automate that function. >> So, if we look back, storage used to be machine attached. Then we went to network classes of storage. Now, we're increasingly distributing data. I think one of the big misnomer's in the industry, is that cloud was a tactic for centralizing resources. In fact, it's turning out that cloud is a tactic for more broad distribution of compute data, and related resources. All of those patterns in this increasingly distributed cloud service oriented world, have to be accommodated, have to be understood, and as you said, to improve predictability, and competence in the system, we have to have some visibility into what it's gonna take to perform, and AI can help us do that. >> Exactly. >> One thing you didn't mention Jim. I want to pick up on something though. Is the idea as we move to Kubernetes, as we move to container based, transient, even serverless types of application forms where the data is. Where all the state is really baked into the data, and not residing in the application. This notion of data assurance is important. Assuring that the data that's required by an instance of a Kubernetes cluster, is available. Can be made available, or will be available when that cluster spins it up. Talk about how that becomes a use case for more AI in the storage subsystems, to ensure that storage can assure that the data that's required is available in the form it needs to be, when it needs to be, and with the policies that are required to secure it, and insure it's integrity. >> Oh yeah, that requirement for that level of data protection requires end-to-end data replication architecture. Infrastructure that's able to assure that all the critical data, or data that's tagged by the real criticality, is always available with backup copies that are always available and close enough to the applications and the users at any point in time. Continuously, so that nobody ever need worry that the data that they need will not be available, because a given server, or storage device is down, or given network is down. End-to-end data replication architecture is automated to a degree that it's always assured, and it will (mumbles) AI, as a (mumbles). First of all, making sure that the end-to-end infrastructure always has a high level, and a very fine (mumbles). A depiction on what is where at every point in time. Also, on the path between all applications, and the critical data sources that they require. Of those paths, always include backups that are hot backups. They're just available without having to worry about it. That the infrastructure predicatively takes care of caching, and replicating, and storing the data wherever it needs to be. To assure that degree of end-to-end data protection and assurance. Once again, that's an automated, and it needs to be an automated capability, especially in the era edge computing, where the storage resources are everywhere. In high preventive architecture really, storage is everywhere, it's just baked into everything. It's the very nature of HCI. >> Right. >> So, you know, yeah. >> So, Jim. We've always, you use a term anomalous behavior, and in the storage world, the storage administrator world, regarded that, or associated that with anticipating or predicting the possibility of a failure somewhere within the subsystem, but as we move to a more broadly distributed use of storage, feeding a more rich and complex set of applications, supporting a more varied and unknown set of user and business activities, the roll of anomalous behavior, even within these data patterns, and security starts to come together. Talk a little about how AI, security, and storage are likely to conflate over the course of the next few years. >> Okay. AI, security, and storage. Well, when you look at security now, data security where everything is being pushed to the edge. You need each device now. You just know that in the internet of things, whether it be an actual edge device, or a gateway, and so forth. To be able to protect the local data resources in an autonomous or semi-autonomous fashion without necessarily having to round trip back to the cloud center, if there is a cloud center, because the critical data is being stored at the edges, so what's happening more and more is that we see something that's called. I forgot the name. Zero perimeter, or perimeterless-- >> Oh, the zero-trust perimeterless security. >> Yeah, there you go, thank you. Where the policies follow the data all the way to wherever it's stored, in a zero trust environment, the permissions, the crypto keys, and so forth, and this is just automatic, so that no matter where the data happens to move, the entire security context follows it. So, what's happening now is that we're seeing that more autonomous operation become part of the architecture of end-to-end data management in this new world. So, to enable, what's happening is that, in terms of protecting that data from any number of theft, or you know denial services, and so forth, AI becomes critically important, machine learning in particular, to be able to detect intrusions autonomously at those edge devices using embedded machine running models that are persistent within the edge nodes themselves. To be able to look for patterns that might be indicative of security threats, because fixed rules are becoming less and less relevant in terms of security rules in an era where the access patterns become more or less 360 degree, in terms of every data resource is being bombarded from all sides by all possible threats. So machine learning is the tool for looking at, what access requests, or attempts on a given data resource, are anomalous in terms of, they've not been seen before. There unusual, they fall inside the confidence intervals that would be normally expected in terms of the access request. So, those edge nodes need then to be able to take action autonomously based on those patterns according to the (mumbles). So we're seeing more of that pattern based security. The edge nodes have zero trust. They're not trusting any access attempt. Any access attempt, even from local applications on the same device, is treated as if it were coming from a remote party, and it has to come through gateway that's erected through machine learning. That machine learning that it learns in real time to adapt to the threat patterns that are seen at that node. >> All right Jim, let's wrap it up there. Once again, we've been, Jim Kobielus and I have been talking about the role that AI's going to play inside the storage capacity, the storage and data services that enterprises are gonna use to improve their business outcomes. Jim, thank you very much for being on The Cube. >> Thank you very much, Peter. >> Once again, I'm Peter Burris. 'Till next time. (techno music)

from our studios in the heart of Silicon Valley Palo Alto California this is a cute conversation hi I'm Peter Burris and welcome back to another cube conversation we're talking today about some of the new challenges that enterprises face as they try to get more out of their data specifically we've got 10 plus years of building advanced analytic pipelines through things like data warehousing but more recently Big Data data lakes etc and we've got some new storage technologies flash and others that have been essential in improving the productivity of a lot of those systems but as we establish the baseline enterprise's are finding new and more complex way to weave together the tool chains that are essential to deploying and sustaining these very complex very rich strategic AI and related analytic applications how will the relationship of storage AI and analytics Co evolved to have that conversation I'm joined by my colleague James Kabila's of wiki bond Jim welcome to the cube thank you Peter so let's start with the problem I kind of laid it out generally but let's start with this observation am i right is there a coevolution taking place between the applications that people want the storage technologies they require and how tool chains are going to weave this all together and the new demands it's going to place on some of these storage technologies yes they're India's coevolution and what's often called the the AI or the data science pipeline which is a simply an industrialised process where my data is turned into machine learning models which are turned into inferences which drive amazing results in real-world applications and to make this pay plan work this industrialized pipeline you need specialists people like data scientists and data engineers you need specialized processes and workflows that are sometimes called DevOps or continuous integration and continuous deployment of models and apps and storage and compute and networking resources are fundamentally important at every stage of this process any data lakes are fundamental in this pipeline because data lakes where the data is stored that it from various source is ingested and then the data is then used to build the models machine learning machine learning being the heart of AI data storage our resources are used to train the models to make sure that they fit or that they're highly predictive for whatever task it is such as recognizing you know a cat in a picture or whatever it might be so storage is fundamentally important as a resource to enable high volume highly parallel ingest transformation cleansing stewardship modeling training and all of that throughout the pipeline but it becomes very most important in the in the in the process of preparing and training the models storage gives way to high volume or highly parallel compute resources as you go further toward inferencing which increasingly is real-time and when you talk about inferencing that's where it gets closer to the need for things like flash and other you know in-memory technologies to take it the next step of the game in terms of the pipeline so the pipeline is required to actually construct the models from large volumes of often small files is necessarily or is likely to be different from the storage that's associated with very high performance automated automation inferencing if I got that right right yeah bulk storage high volume bulk storage talk about petabytes and beyond of data in the back-end process of data engineering and then you have real time in memory data persistence increasingly at the forefront of what you need to do serving and inferences of the models that are created so yes different types of storage are needed at different points in the in the pipeline to make it work as an efficient end-to-end process but let's talk global the conversation because if we go above the devices or if we go up the chain so to speak of the devices we're also now talking about new classes of data services that are going to be required to support these very complex applications and the rich toolchains necessary to drive those applications we're you know encryption new types of security new types of backup and restore new types of data protection I'll give us a sense ultimately of how these new applications that they get more deeply embedded within business and become an essential feature of the revenue producing side not just the cost side of how that's gonna change the need for rethinking data services yeah well you know what happens now with AI machine learning is that it's not just the the the the underlying data that becomes important but the derived assets the machine learning models that you build from the data all that needs to be stored and persistent persistent and governed and from end to end throughout the lifecycle of a given application development so you need storage resources that can manage really you need a repositories that can manage all of those assets as a as a collection to drive the whole DevOps process to enables you know check-in checkout rollback you know transparency the end-to-end process by which and machine learning driven inference is generated so the storage resources need to become increasingly oriented towards object storage if you have complex objects it but also you need end-to-end stream computing to drive a real-time workflow it needs to be highly parallel and you need to be able to manage multiple streams in parallel from a common set of data so it needs to be dating governance baked into this whole end-to-end data persistence and data storage architecture so increasingly the storage resources storage management resources have to evolve and take on attributes of what we can used to call data management they have to know a little bit more about the applications a little bit more about the quality of the data be able to discern patterns in that data so that you can both protect the data where is but also assure the availability of the data where it needs to be and sustained security across the entire set of pipelines and execution resources have I got that right Jim that's right so that demands a degree of end-to-end auditing and logging of the audit trails for all these processes and for the state of every resource that's involved in the construction and the serving of the machine learning model who and who ends in other words you mean that's just data lakes to store the data but you also increasingly need vast petabyte scale logs for end-to-end transparency of this process and the transparency quite often increasingly will be for legal and compliance reasons as well we have mandates Ling gdpr and so for that demand a high degree of transparency into the data and the data derived assets and in general are built and trained throughout the problem they've brought the lifecycle of a given application because some there are real consequences of the machine learning model fails or makes the wrong decision that might be legally apt actionable and so more the storage architecture has to support high degree of really like queryable archives to be able to manage to search and and to be able to roll up a complete audit trail of why a given model AI model had a given result in the field Jim there's been a great conversation thanks a lot for talking about this crucial relationship between analytics applications like AI driven applications and and storage and crucially the evolving role and the impacts of on storage these very rich data pipelines or tools change that are gonna make all of that possible once again I'm Peter Burris you've been listening to another cube conversation thanks very much you [Music]

(upbeat techno music) >> Live, from Las Vegas, it's theCUBE covering VMworld 2018, brought to you by VMware and it's ecosystem partners. >> Okay, welcome back everyone, we're here live in Las Vegas for VMworld 2018 coverage. It's the final analysis, the final interview of three days, 94 interviews, two CUBE sets, amazing production, our ninth year covering VMworld. We've seen the evolution, we've seen the trials and tribulations of VMware and it's ecosystem and as it moves into the modern era, the dynamics are changing. We heard quotes like, "From playing tennis "to playing soccer," it's a lot of complicated things, the cloud certainly a big part of it. I'm John Furrier your host, Stu Miniman couldn't be here for the wrap, he had an appointment. I'm here with Dave Vallente and Jim Kobielus who's with Wikibon and SiliconANGLE and theCUBE team. >> Guys, great job, I want to say thanks to you guys and thanks to the crew on both sets. Amazing production, we're just going to have some fun here. We've analyzed this event, ten different ways from Sunday. >> So many people working so hard for such a steady clip as we have here the last three days, amazing. >> Just to give some perspective, I want to get, just lay out kind of what's going on with theCUBE. I've get a lot of people come up and ask me hey what's going on, you guys are amazing. It's gotten so much bigger, there's two sets. But every year, Dave, we always try to at VMworld, make VMworld our show to up our value. We always love to innovate, but we got a business to run. We have no outside finance, we have a great set of partners. I'm proud of the team, what Jeff Frick did and the team has done amazing work. Sonia's here's, the whole analyst team's here, our whole team's here. But we have an orchestrated system now, we have the blogging at SilconANGLE.com and Rob Hof leading the editorial. Working on a content immersion program. Jim you were involved in with Rob and Peter in the team, bringing content on the written word side, as fast as possible, the best quality, fast as possible, the analysts getting the pre-briefing and the NDAs, theCUBE team setting it up. Pretty unique formula at full stride right now, only going to get better. New photography, better pictures, better video, better guests, more content. Now with the video clipper tool and our video cloud service and we did a tech preview of our block chain, token economics, a lot of the insiders of VMworld, the senior executives and the community, all with great results, they all loved it, they want to do more. Opening up our platform, opening up the content's been a big success, I want to thank you guys for that. >> And I agree, I should point out that one of the things we have that say an agency doesn't offer, I used to be with a large multi national solutions provider doing kind of similar work but in a thought leadership market kind of, let me just state something here, what we've got is unique because we have analysts, market researchers, who know this stuff at the core of our business model, including, especially the content immersion program. Peter Boroughs did a bit, I did a fair amount on this one. You need subject matter experts to curate and really define the themes that the entire editorial team, and I'm including theCUBE people on the editorial team, are basically, so we're all aligned around we know what the industry is all about, the context, the vendor, and somebody's just curating making sure that the subject matter is on target was what the community wants to see. >> So I got to day, first of all, VMware set us up with two stages here, two sets, amazing. They've been unbelievable partners. They really put their money with their mouth is. They allow us to bring in the ecosystem, do our own thing, so that's phenomenal and our goal is to give back to the community. We had two sets, 94 guests this week, 70 interview segments, hundreds and hundreds of assets coming out, all free. >> It was amazing. >> SiliconANGLE.com, Wikibon.com, theCUBE.net, all free content was really incredible. >> It's good free content. >> It's great free content. >> We dropped a true private cloud report with market shares, that's all open and free. Floyer did a piece on VMware's hybrid cloud strategy, near to momentum, ice bergs ahead. Jim Kobelius, first of all, every day here you laid out here's what happened today with your analysis plus you had previews plus you have a trip report coming. >> Plus I had a Wikibon research note that had been in the pipeline for about a month and I held off on publishing until Monday at the show, the AI ready IT infrastructure because it's so aligned with what's going on. >> And then Paul Gillan and Rob Hof did a series in their team on the future of the data center. Paul Gillan, the walls are tumbling down, I mean that thing got amazing play, check that out. It's just a lot of detail in there. >> And more importantly, that's our content. We're linking, we're open, we're linking to other people's content, from Tech Field Day what Foskett's doing to vBrownBag to linking to stories, sharing, quoting other analysts, Patrick Moorehead for more insights. Anyone who has content that we can get it in fast, in real time, out to the marketplace, is our mission and we love doing it so I think the formula of open is working. >> Yeah Charles King, this morning I saw Charles, I thanked him for, he had great quotes. >> Yeah, great guy. >> He's like, "I love with Paul Gillan calls me." John, talk about the tech preview because the tech preview was an open community project that's all about bringing the community together, helping them and helping get content out into the marketplace. >> Well our goal for this event was to use the VMworld to preview some of our innovations and you're going to start to hear more from the siliconANGLE media, CUBE and siliconANGLE team around concepts like the CUBE cloud. We have technology we're going to start to surface and bring out to the marketplace and we want to make it free and open and allow people to use and share in what we do and make theCUBE a community brand and a community concept and continue this mission and treat theCUBE like an upstream project. Let's all co-create together because the downstream benefits in communities are significantly better when there's co-creation and self governance. Highest quality content, from highly reputable people, whether it's news, analysis, opinion, commentary, pontification, we love it all, let the content stand on it's own and let's the benefits come down so if you're a sponsor, if you're a thought leader, you're a news maker, you're an analyst, we love to do that and we love talking with the executives so that's great. The tech preview is about showcasing how we want to create a new network. As communities are growing and changing, VMware's community is robust, Dave, it's it's own subnet, but as the world grows in those multiple clouds, Azure has a community, Google has a community, and people have been trained to sit in these silos, okay? >> Mm-hmm. >> We go to so many events and we engage with so many communities, we want to connect them all through the CUBE coin concept of block chain where if someone's in a community, they can download the wallet and join theCUBE network. Today there's no mechanism to join theCUBE network. You can go to theCUBE.net and subscribe, you can go to YouTube and subscribe, you can get e-mail marketing but that's not acceptable to us we want a subscribe button that's going to add value to people who contribute value, they can capture it. That was the tech preview, it's a block chain based community. We're calling it the Open Community Project. >> Wow. >> Open Community Project is the first upstream content software model that's free to use, where if the community uses it, they can capture value that they create. It's a new concept and it's radical and revolutionary. >> In some ways were analogous to what VMware has evolved into where they bridge clouds and they say that, "We bridge clouds." We bridge communities all around thought leadership and to provide a forum for conversations that bridge the various siloed communities. >> Well Jim you and I talked about this, we've seen the movie and media. In the old school media days and search engine marketing and e-mail marketing and starting a blog, which we were part of, the blogging was the first generation of sharing economy where you linked to other bloggers and shared your traffic, because you were working together against the mainstream media. >> It's my major keyboard, by the way, I love blogs. >> And if you were funded you had to build an audience. Audience development, audience development. Not anymore, the audience is already there. They are now in networks so the new ethos, like blogging, is joining networks and not making it an ownership, lock in walled garden. So the new ethos is not link sharing, community sharing, co-creation and merging networks. This is something that we're seeing across all event communities and content is the nutrients and the glue for those communities. >> You got multi cloud, you got multi content networks. Making it together, it's exciting. I mean there were some people that I saw this week, I mean Alan Cohen as a guest host, amazingly articulate, super smart guy, plugged in to Silicon Valley. Christophe Bertrand, analyst at ESG, a great analysis today on theCUBE, bringing those guys, nominate them into the community for the Open Community Project. >> You know what I like, Dave, was also Jeff Frick, Sonia and Gabe were all at the front there, greeting the guests. We had great speakers, it all worked. The stages worked but it's for the community, by the community, this is the model, right? This is what we want to do and it was a lot of fun, I had a lot of great interviews from Andy Bechtolsheim, Michael Dell, Pat Gellsinger to practitioners and to the vendors and suppliers all co-creating here in real time, it was really a lot of fun. >> Oh yes, amen. >> Well Dave, thanks for everything. Thanks for the crew, great job everybody. >> Awesome. >> Jim, well done. >> Thanks to Stu Miniman, Peter Burris and all the guests, Justin Warren, John Troyer, guest host Alan Cohen, great community participation. This is theCUBE signing off from Las Vegas, this is VMworld 2018 final analysis, thanks for watching. (upbeat techno music)

>> Hi I'm Peter Burris. Welcome to another Wikibon Action Item. (upbeat music) We're once again broadcasting from theCube's beautiful Palo Alto, California studio. I'm joined here in the studio by George Gilbert and David Floyer. And then remotely, we have Jim Kobielus, David Vellante, Neil Raden and Ralph Finos. Hi guys. >> Hey. >> Hi >> How you all doing? >> This is a great, great group of people to talk about the topic we're going to talk about, guys. We're going to talk about the notion of de-risking digital business. Now, the reason why this becomes interesting is, the Wikibon perspective for quite some time has been that the difference between business and digital business is the role that data assets play in a digital business. Now, if you think about what that means. Every business institutionalizes its work around what it regards as its most important assets. A bottling company, for example, organizes around the bottling plant. A financial services company organizes around the regulatory impacts or limitations on how they share information and what is regarded as fair use of data and other resources, and assets. The same thing exists in a digital business. There's a difference between, say, Sears and Walmart. Walmart mades use of data differently than Sears. And that specific assets that are employed and had a significant impact on how the retail business was structured. Along comes Amazon, which is even deeper in the use of data as a basis for how it conducts its business and Amazon is institutionalizing work in quite different ways and has been incredibly successful. We could go on and on and on with a number of different examples of this, and we'll get into that. But what it means ultimately is that the tie between data and what is regarded as valuable in the business is becoming increasingly clear, even if it's not perfect. And so traditional approaches to de-risking data, through backup and restore, now needs to be re-thought so that it's not just de-risking the data, it's de-risking the data assets. And, since those data assets are so central to the business operations of many of these digital businesses, what it means to de-risk the whole business. So, David Vellante, give us a starting point. How should folks think about this different approach to envisioning business? And digital business, and the notion of risk? >> Okay thanks Peter, I mean I agree with a lot of what you just said and I want to pick up on that. I see the future of digital business as really built around data sort of agreeing with you, building on what you just said. Really where organizations are putting data at the core and increasingly I believe that organizations that have traditionally relied on human expertise as the primary differentiator, will be disrupted by companies where data is the fundamental value driver and I think there are some examples of that and I'm sure we'll talk about it. And in this new world humans have expertise that leverage the organization's data model and create value from that data with augmented machine intelligence. I'm not crazy about the term artificial intelligence. And you hear a lot about data-driven companies and I think such companies are going to have a technology foundation that is increasingly described as autonomous, aware, anticipatory, and importantly in the context of today's discussion, self-healing. So able to withstand failures and recover very quickly. So de-risking a digital business is going to require new ways of thinking about data protection and security and privacy. Specifically as it relates to data protection, I think it's going to be a fundamental component of the so-called data-driven company's technology fabric. This can be designed into applications, into data stores, into file systems, into middleware, and into infrastructure, as code. And many technology companies are going to try to attack this problem from a lot of different angles. Trying to infuse machine intelligence into the hardware, software and automated processes. And the premise is that meaty companies will architect their technology foundations, not as a set of remote cloud services that they're calling, but rather as a ubiquitous set of functional capabilities that largely mimic a range of human activities. Including storing, backing up, and virtually instantaneous recovery from failure. >> So let me build on that. So what you're kind of saying if I can summarize, and we'll get into whether or not it's human expertise or some other approach or notion of business. But you're saying that increasingly patterns in the data are going to have absolute consequential impacts on how a business ultimately behaves. We got that right? >> Yeah absolutely. And how you construct that data model, and provide access to the data model, is going to be a fundamental determinant of success. >> Neil Raden, does that mean that people are no longer important? >> Well no, no I wouldn't say that at all. I'm talking with the head of a medical school a couple of weeks ago, and he said something that really resonated. He said that there're as many doctors who graduated at the bottom of their class as the top of their class. And I think that's true of organizations too. You know what, 20 years ago I had the privilege of interviewing Peter Drucker for an hour and he foresaw this, 20 years ago, he said that people who run companies have traditionally had IT departments that provided operational data but they needed to start to figure out how to get value from that data and not only get value from that data but get value from data outside the company, not just internal data. So he kind of saw this big data thing happening 20 years ago. Unfortunately, he had a prejudice for senior executives. You know, he never really thought about any other people in an organization except the highest people. And I think what we're talking about here is really the whole organization. I think that, I have some concerns about the ability of organizations to really implement this without a lot of fumbles. I mean it's fine to talk about the five digital giants but there's a lot of companies out there that, you know the bar isn't really that high for them to stay in business. And they just seem to get along. And I think if we're going to de-risk we really need to help companies understand the whole process of transformation, not just the technology. >> Well, take us through it. What is this process of transformation? That includes the role of technology but is bigger than the role of technology. >> Well, it's like anything else, right. There has to be communication, there has to be some element of control, there has to be a lot of flexibility and most importantly I think there has to be acceptability by the people who are going to be affected by it, that is the right thing to do. And I would say you start with assumptions, I call it assumption analysis, in other words let's all get together and figure out what our assumptions are, and see if we can't line em up. Typically IT is not good at this. So I think it's going to require the help of a lot of practitioners who can guide them. >> So Dave Vellante, reconcile one point that you made I want to come back to this notion of how we're moving from businesses built on expertise and people to businesses built on expertise resident as patterns in the data, or data models. Why is it that the most valuable companies in the world seem to be the ones that have the most real hardcore data scientists. Isn't that expertise and people? >> Yeah it is, and I think it's worth pointing out. Look, the stock market is volatile, but right now the top-five companies: Apple, Amazon, Google, Facebook and Microsoft, in terms of market cap, account for about $3.5 trillion and there's a big distance between them, and they've clearly surpassed the big banks and the oil companies. Now again, that could change, but I believe that it's because they are data-driven. So called data-driven. Does that mean they don't need humans? No, but human expertise surrounds the data as opposed to most companies, human expertise is at the center and the data lives in silos and I think it's very hard to protect data, and leverage data, that lives in silos. >> Yes, so here's where I'll take exception to that, Dave. And I want to get everybody to build on top of this just very quickly. I think that human expertise has surrounded, in other businesses, the buildings. Or, the bottling plant. Or, the wealth management. Or, the platoon. So I think that the organization of assets has always been the determining factor of how a business behaves and we institutionalized work, in other words where we put people, based on the business' understanding of assets. Do you disagree with that? Is that, are we wrong in that regard? I think data scientists are an example of reinstitutionalizing work around a very core asset in this case, data. >> Yeah, you're saying that the most valuable asset is shifting from some of those physical assets, the bottling plant et cetera, to data. >> Yeah we are, we are. Absolutely. Alright, David Foyer. >> Neil: I'd like to come in. >> Panelist: I agree with that too. >> Okay, go ahead Neil. >> I'd like to give an example from the news. Cigna's acquisition of Express Scripts for $67 billion. Who the hell is Cigna, right? Connecticut General is just a sleepy life insurance company and INA was a second-tier property and casualty company. They merged a long time ago, they got into health insurance and suddenly, who's Express Scripts? I mean that's a company that nobody ever even heard of. They're a pharmacy benefit manager, what is that? They're an information management company, period. That's all they do. >> David Foyer, what does this mean from a technology standpoint? >> So I wanted to to emphasize one thing that evolution has always taught us. That you have to be able to come from where you are. You have to be able to evolve from where you are and take the assets that you have. And the assets that people have are their current systems of records, other things like that. They must be able to evolve into the future to better utilize what those systems are. And the other thing I would like to say-- >> Let me give you an example just to interrupt you, because this is a very important point. One of the primary reasons why the telecommunications companies, whom so many people believed, analysts believed, had this fundamental advantage, because so much information's flowing through them is when you're writing assets off for 30 years, that kind of locks you into an operational mode, doesn't it? >> Exactly. And the other thing I want to emphasize is that the most important thing is sources of data not the data itself. So for example, real-time data is very very important. So what is your source of your real-time data? If you've given that away to Google or your IOT vendor you have made a fundamental strategic mistake. So understanding the sources of data, making sure that you have access to that data, is going to enable you to be able to build the sort of processes and data digitalization. >> So let's turn that concept into kind of a Geoffrey Moore kind of strategy bromide. At the end of the day you look at your value proposition and then what activities are central to that value proposition and what data is thrown off by those activities and what data's required by those activities. >> Right, both internal-- >> We got that right? >> Yeah. Both internal and external data. What are those sources that you require? Yes, that's exactly right. And then you need to put together a plan which takes you from where you are, as the sources of data and then focuses on how you can use that data to either improve revenue or to reduce costs, or a combination of those two things, as a series of specific exercises. And in particular, using that data to automate in real-time as much as possible. That to me is the fundamental requirement to actually be able to do this and make money from it. If you look at every example, it's all real-time. It's real-time bidding at Google, it's real-time allocation of resources by Uber. That is where people need to focus on. So it's those steps, practical steps, that organizations need to take that I think we should be giving a lot of focus on. >> You mention Uber. David Vellante, we're just not talking about the, once again, talking about the Uberization of things, are we? Or is that what we mean here? So, what we'll do is we'll turn the conversation very quickly over to you George. And there are existing today a number of different domains where we're starting to see a new emphasis on how we start pricing some of this risk. Because when we think about de-risking as it relates to data give us an example of one. >> Well we were talking earlier, in financial services risk itself is priced just the way time is priced in terms of what premium you'll pay in terms of interest rates. But there's also something that's softer that's come into much more widely-held consciousness recently which is reputational risk. Which is different from operational risk. Reputational risk is about, are you a trusted steward for data? Some of that could be personal information and a use case that's very prominent now with the European GDPR regulation is, you know, if I ask you as a consumer or an individual to erase my data, can you say with extreme confidence that you have? That's just one example. >> Well I'll give you a specific number on that. We've mentioned it here on Action Item before. I had a conversation with a Chief Privacy Officer a few months ago who told me that they had priced out what the fines to Equifax would have been had the problem occurred after GDPR fines were enacted. It was $160 billion, was the estimate. There's not a lot of companies on the planet that could deal with $160 billion liability. Like that. >> Okay, so we have a price now that might have been kind of, sort of mushy before. And the notion of trust hasn't really changed over time what's changed is the technical implementations that support it. And in the old world with systems of record we basically collected from our operational applications as much data as we could put it in the data warehouse and it's data marked satellites. And we try to govern it within that perimeter. But now we know that data basically originates and goes just about anywhere. There's no well-defined perimeter. It's much more porous, far more distributed. You might think of it as a distributed data fabric and the only way you can be a trusted steward of that is if you now, across the silos, without trying to centralize all the data that's in silos or across them, you can enforce, who's allowed to access it, what they're allowed to do, audit who's done what to what type of data, when and where? And then there's a variety of approaches. Just to pick two, one is where it's discovery-oriented to figure out what's going on with the data estate. Using machine learning this is, Alation is an example. And then there's another example, which is where you try and get everyone to plug into what's essentially a new system catalog. That's not in a in a deviant mesh but that acts like the fabric for your data fabric, deviant mesh. >> That's an example of another, one of the properties of looking at coming at this. But when we think, Dave Vellante coming back to you for a second. When we think about the conversation there's been a lot of presumption or a lot of bromide. Analysts like to talk about, don't get Uberized. We're not just talking about getting Uberized. We're talking about something a little bit different aren't we? >> Well yeah, absolutely. I think Uber's going to get Uberized, personally. But I think there's a lot of evidence, I mentioned the big five, but if you look at Spotify, Waze, AirbnB, yes Uber, yes Twitter, Netflix, Bitcoin is an example, 23andme. These are all examples of companies that, I'll go back to what I said before, are putting data at the core and building humans expertise around that core to leverage that expertise. And I think it's easy to sit back, for some companies to sit back and say, "Well I'm going to wait and see what happens." But to me anyway, there's a big gap between kind of the haves and the have-nots. And I think that, that gap is around applying machine intelligence to data and applying cloud economics. Zero marginal economics and API economy. An always-on sort of mentality, et cetera et cetera. And that's what the economy, in my view anyway, is going to look like in the future. >> So let me put out a challenge, Jim I'm going to come to you in a second, very quickly on some of the things that start looking like data assets. But today, when we talk about data protection we're talking about simply a whole bunch of applications and a whole bunch of devices. Just spinning that data off, so we have it at a third site. And then we can, and it takes to someone in real-time, and then if there's a catastrophe or we have, you know, large or small, being able to restore it often in hours or days. So we're talking about an improvement on RPO and RTO but when we talk about data assets, and I'm going to come to you in a second with that David Floyer, but when we talk about data assets, we're talking about, not only the data, the bits. We're talking about the relationships and the organization, and the metadata, as being a key element of that. So David, I'm sorry Jim Kobielus, just really quickly, thirty seconds. Models, what do they look like? What are the new nature of some of these assets look like? >> Well the new nature of these assets are the machine learning models that are driving so many business processes right now. And so really the core assets there are the data obviously from which they are developed, and also from which they are trained. But also very much the knowledge of the data scientists and engineers who build and tune this stuff. And so really, what you need to do is, you need to protect that knowledge and grow that knowledge base of data science professionals in your organization, in a way that builds on it. And hopefully you keep the smartest people in house. And they can encode more of their knowledge in automated programs to manage the entire pipeline of development. >> We're not talking about files. We're not even talking about databases, are we David Floyer? We're talking about something different. Algorithms and models are today's technology's really really set up to do a good job of protecting the full organization of those data assets. >> I would say that they're not even being thought about yet. And going back on what Jim was saying, Those data scientists are the only people who understand that in the same way as in the year 2000, the COBOL programmers were the only people who understood what was going on inside those applications. And we as an industry have to allow organizations to be able to protect the assets inside their applications and use AI if you like to actually understand what is in those applications and how are they working? And I think that's an incredibly important de-risking is ensuring that you're not dependent on a few experts who could leave at any moment, in the same way as COBOL programmers could have left. >> But it's not just the data, and it's not just the metadata, it really is the data structure. >> It is the model. Just the whole way that this has been put together and the reason why. And the ability to continue to upgrade that and change that over time. So those assets are incredibly important but at the moment there is no way that you can, there isn't technology available for you to actually protect those assets. >> So if I combine what you just said with what Neil Raden was talking about, David Vallante's put forward a good vision of what's required. Neil Raden's made the observation that this is going to be much more than technology. There's a lot of change, not change management at a low level inside the IT, but business change and the technology companies also have to step up and be able to support this. We're seeing this, we're seeing a number of different vendor types start to enter into this space. Certainly storage guys, Dylon Sears talking about doing a better job of data protection we're seeing middleware companies, TIBCO and DISCO, talk about doing this differently. We're seeing file systems, Scality, WekaIO talk about doing this differently. Backup and restore companies, Veeam, Veritas. I mean, everybody's looking at this and they're all coming at it. Just really quickly David, where's the inside track at this point? >> For me it is so much whitespace as to be unbelievable. >> So nobody has an inside track yet. >> Nobody has an inside track. Just to start with a few things. It's clear that you should keep data where it is. The cost of moving data around an organization from inside to out, is crazy. >> So companies that keep data in place, or technologies to keep data in place, are going to have an advantage. >> Much, much, much greater advantage. Sure, there must be backups somewhere. But you need to keep the working copies of data where they are because it's the real-time access, usually that's important. So if it originates in the cloud, keep it in the cloud. If it originates in a data-provider, on another cloud, that's where you should keep it. If it originates on your premise, keep it where it originated. >> Unless you need to combine it. But that's a new origination point. >> Then you're taking subsets of that data and then combining that up for itself. So that would be my first point. So organizations are going to need to put together what George was talking about, this metadata of all the data, how it interconnects, how it's being used. The flow of data through the organization, it's amazing to me that when you go to an IT shop they cannot define for you how the data flows through that data center or that organization. That's the requirement that you have to have and AI is going to be part of that solution, of looking at all of the applications and the data and telling you where it's going and how it's working together. >> So the second thing would be companies that are able to build or conceive of networks as data. Will also have an advantage. And I think I'd add a third one. Companies that demonstrate perennial observations, a real understanding of the unbelievable change that's required you can't just say, oh Facebook wants this therefore everybody's going to want it. There's going to be a lot of push marketing that goes on at the technology side. Alright so let's get to some Action Items. David Vellante, I'll start with you. Action Item. >> Well the future's going to be one where systems see, they talk, they sense, they recognize, they control, they optimize. It may be tempting to say, you know what I'm going to wait, I'm going to sit back and wait to figure out how I'm going to close that machine intelligence gap. I think that's a mistake. I think you have to start now, and you have to start with your data model. >> George Gilbert, Action Item. >> I think you have to keep in mind the guardrails related to governance, and trust, when you're building applications on the new data fabric. And you can take the approach of a platform-oriented one where you're plugging into an API, like Apache Atlas, that Hortonworks is driving, or a discovery-oriented one as David was talking about which would be something like Alation, using machine learning. But if, let's say the use case starts out as an IOT, edge analytics and cloud inferencing, that data science pipeline itself has to now be part of this fabric. Including the output of the design time. Meaning the models themselves, so they can be managed. >> Excellent. Jim Kobielus, you've been pretty quiet but I know you've got a lot to offer. Action Item, Jim. >> I'll be very brief. What you need to do is protect your data science knowledge base. That's the way to de-risk this entire process. And that involves more than just a data catalog. You need a data science expertise registry within your distributed value chain. And you need to manage that as a very human asset that needs to grow. That is your number one asset going forward. >> Ralph Finos, you've also been pretty quiet. Action Item, Ralph. >> Yeah, I think you've got to be careful about what you're trying to get done. Whether it's, it depends on your industry, whether it's finance or whether it's the entertainment business, there are different requirements about data in those different environments. And you need to be cautious about that and you need leadership on the executive business side of things. The last thing in the world you want to do is depend on data scientists to figure this stuff out. >> And I'll give you the second to last answer or Action Item. Neil Raden, Action Item. >> I think there's been a lot of progress lately in creating tools for data scientists to be more efficient and they need to be, because the big digital giants are draining them from other companies. So that's very encouraging. But in general I think becoming a data-driven, a digital transformation company for most companies, is a big job and I think they need to it in piece parts because if they try to do it all at once they're going to be in trouble. >> Alright, so that's great conversation guys. Oh, David Floyer, Action Item. David's looking at me saying, ah what about me? David Floyer, Action Item. >> (laughing) So my Action Item comes from an Irish proverb. Which if you ask for directions they will always answer you, "I wouldn't start from here." So the Action Item that I have is, if somebody is coming in saying you have to re-do all of your applications and re-write them from scratch, and start in a completely different direction, that is going to be a 20-year job and you're not going to ever get it done. So you have to start from what you have. The digital assets that you have, and you have to focus on improving those with additional applications, additional data using that as the foundation for how you build that business with a clear long-term view. And if you look at some of the examples that were given early, particularly in the insurance industries, that's what they did. >> Thank you very much guys. So, let's do an overall Action Item. We've been talking today about the challenges of de-risking digital business which ties directly to the overall understanding of the role of data assets play in businesses and the technology's ability to move from just protecting data, restoring data, to actually restoring the relationships in the data, the structures of the data and very importantly the models that are resident in the data. This is going to be a significant journey. There's clear evidence that this is driving a new valuation within the business. Folks talk about data as the new oil. We don't necessarily see things that way because data, quite frankly, is a very very different kind of asset. The cost could be shared because it doesn't suffer the same limits on scarcity. So as a consequence, what has to happen is, you have to start with where you are. What is your current value proposition? And what data do you have in support of that value proposition? And then whiteboard it, clean slate it and say, what data would we like to have in support of the activities that we perform? Figure out what those gaps are. Find ways to get access to that data through piecemeal, piece-part investments. That provide a roadmap of priorities looking forward. Out of that will come a better understanding of the fundamental data assets that are being created. New models of how you engage customers. New models of how operations works in the shop floor. New models of how financial services are being employed and utilized. And use that as a basis for then starting to put forward plans for bringing technologies in, that are capable of not just supporting the data and protecting the data but protecting the overall organization of data in the form of these models, in the form of these relationships, so that the business can, as it creates these, as it throws off these new assets, treat them as the special resource that the business requires. Once that is in place, we'll start seeing businesses more successfully reorganize, reinstitutionalize the work around data, and it won't just be the big technology companies who have, who people call digital native, that are well down this path. I want to thank George Gilbert, David Floyer here in the studio with me. David Vellante, Ralph Finos, Neil Raden and Jim Kobelius on the phone. Thanks very much guys. Great conversation. And that's been another Wikibon Action Item. (upbeat music)

>> Announcer: From the SiliconANGLE Media office in Boston, Massachusetts, it's theCUBE. Now here's your host, Dave Vellante. >> In the early days of big data and Hadoop, the focus was really on operational efficiency where ROI was largely centered on reduction of investment. Fast forward 10 years and you're seeing a plethora of activity around machine learning, and deep learning, and artificial intelligence, and deeper business integration as a function of machine intelligence. Welcome to this Cube conversation, The Skinny on Machine Intelligence. I'm Dave Vellante and I'm excited to have Jim Kobielus here up from the District area. Jim, great to see you. Thanks for coming into the office today. >> Thanks a lot, Dave, yes great to be here in beautiful Marlboro, Massachusetts. >> Yes, so you know Jim, when you think about all the buzz words in this big data business, I have to ask you, is this just sort of same wine, new bottle when we talk about all this AI and machine intelligence stuff? >> It's actually new wine. But of course there's various bottles and they have different vintages, and much of that wine is still quite tasty, and let me just break it out for you, the skinny on machine intelligence. AI as a buzzword and as a set of practices really goes back of course to the early post-World War II era, as we know Alan Turing and the Imitation Game and so forth. There are other developers, theorists, academics in the '40s and the '50s and '60s that pioneered in this field. So we don't want to give Alan Turing too much credit, but he was clearly a mathematician who laid down the theoretical framework for much of what we now call Artificial Intelligence. But when you look at Artificial Intelligence as a ever-evolving set of practices, where it began was in an area that focused on deterministic rules, rule-driven expert systems, and that was really the state of the art of AI for a long, long time. And so you had expert systems in a variety of areas that became useful or used in business, and science, and government and so forth. Cut ahead to the turn of the millennium, we are now in the 21st century, and what's different, the new wine, is big data, larger and larger data sets that can reveal great insights, patterns, correlations that might be highly useful if you have the right statistical modeling tools and approaches to be able to surface up these patterns in an automated or semi-automated fashion. So one of the core areas is what we now call machine learning, which really is using statistical models to infer correlations, anomalies, trends, and so forth in the data itself, and machine learning, the core approach for machine learning is something called Artificial Neural Networks, which is essentially modeling a statistical model along the lines of how, at a very high level, the nervous system is made up, with neurons connected by synapses, and so forth. It's an analog in statistical modeling called a perceptron. The whole theoretical framework of perceptrons actually got started in the 1950s with the first flush of AI, but didn't become a practical reality until after the turn of this millennium, really after the turn of this particular decade, 2010, when we started to see not only very large big data sets emerge and new approaches for managing it all, like Hadoop, come to the fore. But we've seen artificial neural nets get more sophisticated in terms of their capabilities, and a new approach for doing machine learning, artificial neural networks, with deeper layers of perceptrons, neurons, called deep learning has come to the fore. With deep learning, you have new algorithms like convolutional neural networks, recurrent neural networks, generative adversarial neural networks. These are different ways of surfacing up higher level abstractions in the data, for example for face recognition and object recognition, voice recognition and so forth. These all depend on this new state of the art for machine learning called deep learning. So what we have now in the year 2017 is we have quite a mania for all things AI, much of it is focused on deep learning, much of it is focused on tools that your average data scientist or your average developer increasingly can use and get very productive with and build these models and train and test them, and deploy them into working applications like going forward, things like autonomous vehicles would be impossible without this. >> Right, and we'll get some of that. But so you're saying that machine learning is essentially math that infers patterns from data. And math, it's new math, math that's been around for awhile or. >> Yeah, and inferring patterns from data has been done for a long time with software, and we have some established approaches that in many ways predate the current vogue for neural networks. We have support vector machines, and decision trees, and Bayesian logic. These are different ways of approaches statistical for inferring patterns, correlations in the data. They haven't gone away, they're a big part of the overall AI space, but it's a growing area that I've only skimmed the surface of. >> And they've been around for many many years, like SVM for example. Okay, now describe further, add some color to deep learning. You sort of painted a picture of this sort of deep layers of these machine learning algorithms and this network with some depth to it, but help us better understand the difference between machine learning and deep learning, and then ultimately AI. >> Yeah, well with machine learning generally, you know, inferring patterns from data that I said, artificial neural networks of which the deep learning networks are one subset. Artificial neural networks can be two or more layers of perceptrons or neurons, they have relationship to each other in terms of their activation according to various mathematical functions. So when you look at an artificial neural network, it basically does very complex math equations through a combination of what they call scalar functions, like multiplication and so forth, and then you have these non-linear functions, like cosine and so forth, tangent, all that kind of math playing together in these deep structures that are triggered by data, data input that's processed according to activation functions that set weights and reset the weights among all the various neural processing elements, that ultimately output something, the insight or the intelligence that you're looking for, like a yes or no, is this a face or not a face, that these incoming bits are presenting. Or it might present output in terms of categories. What category of face is this, a man, a woman, a child, or whatever. What I'm getting at is that so deep learning is more layers of these neural processing elements that are specialized to various functions to be able to abstract higher level phenomena from the data, it's not just, "Is this a face," but if it's a scene recognition deep learning network, it might recognize that this is a face that corresponds to a person named Dave who also happens to be the father in the particular family scene, and by the way this is a family scene that this deep learning network is able to ascertain. What I'm getting at is those are the higher level abstractions that deep learning algorithms of various sorts are built to identify in an automated way. >> Okay, and these in your view all fit under the umbrella of artificial intelligence, or is that sort of an uber field that we should be thinking of. >> Yeah, artificial intelligence as the broad envelope essentially refers to any number of approaches that help machines to think like humans, essentially. When you say, "Think like humans," what does that mean actually? To do predictions like humans, to look for anomalies or outliers like a human might, you know separate figure from ground for example in a scene, to identify the correlations or trends in a given scene. Like I said, to do categorization or classification based on what they're seeing in a given frame or what they're hearing in a given speech sample. So all these cognitive processes just skim the surface, or what AI is all about, automating to a great degree. When I say cognitive, but I'm also referring to affective like emotion detection, that's another set of processes that goes on in our heads or our hearts, that AI based on deep learning and so forth is able to do depending on different types of artificial neural networks are specialized particular functions, and they can only perform these functions if A, they've been built and optimized for those functions, and B, they have been trained with actual data from the phenomenon of interest. Training the algorithms with the actual data to determine how effective the algorithms are is the key linchpin of the process, 'cause without training the algorithms you don't know if the algorithm is effective for its intended purpose, so in Wikibon what we're doing is in the whole development process, DevOps cycle, for all things AI, training the models through a process called supervised learning is absolutely an essential component of ascertaining the quality of the network that you've built. >> So that's the calibration and the iteration to increase the accuracy, and like I say, the quality of the outcome. Okay, what are some of the practical applications that you're seeing for AI, and ML, and DL. >> Well, chat bots, you know voice recognition in general, Siri and Alexa, and so forth. Without machine learning, without deep learning to do speech recognition, those can't work, right? Pretty much in every field, now for example, IT service management tools of all sorts. When you have a large network that's logging data at the server level, at the application level and so forth, those data logs are too large and too complex and changing too fast for humans to be able to identify the patterns related to issues and faults and incidents. So AI, machine learning, deep learning is being used to fathom those anomalies and so forth in an automated fashion to be able to alert a human to take action, like an IT administrator, or to be able to trigger a response work flow, either human or automated. So AI within IT service management, hot hot topic, and we're seeing a lot of vendors incorporate that capability into their tools. Like I said, in the broad world we live in in terms of face recognition and Facebook, the fact is when I load a new picture of myself or my family or even with some friends or brothers in it, Facebook knows lickity-split whether it's my brother Tom or it's my wife or whoever, because of face recognition which obviously depends, well it's not obvious to everybody, depends on deep learning algorithms running inside Facebook's big data network, big data infrastructure. They're able to immediately know this. We see this all around us now, speech recognition, face recognition, and we just take it for granted that it's done, but it's done through the magic of AI. >> I want to get to the development angle scenario that you specialize in. Part of the reason why you came to Wikibon is to really focus on that whole application development angle. But before we get there, I want to follow the data for a bit 'cause you mentioned that was really the catalyst for the resurgence in AI, and last week at the Wikibon research meeting we talked about this three-tiered model. Edge, as edge piece, and then something in the middle which is this aggregation point for all this edge data, and then cloud which is where I guess all the deep modeling occurs, so sort of a three-tier model for the data flow. >> John: Yes. >> So I wonder if you could comment on that in the context of AI, it means more data, more I guess opportunities for machine learning and digital twins, and all this other cool stuff that's going on. But I'm really interested in how that is going to affect the application development and the programming model. John Farrier has a phrase that he says that, "Data is the new development kit." Well, if you got all this data that's distributed all over the place, that changes the application development model, at least you think it does. So I wonder if you could comment on that edge explosion, the data explosion as a result, and what it means for application development. >> Right, so more and more deep learning algorithms are being pushed to edge devices, by that I mean smartphones, and smart appliances like the ones that incorporate Alexa and so forth. And so what we're talking about is the algorithms themselves are being put into CPUs and FPGAs and ASICs and GPUs. All that stuff's getting embedded in everything that we're using, everything's that got autonomous, more and more devices have the ability if not to be autonomous in terms of making decisions, independent of us, or simply to serve as augmentation vehicles for our own whatever we happen to be doing thanks to the power of deep learning at the client. Okay, so when deep learning algorithms are embedded in say an internet of things edge device, what the deep learning algorithms are doing is A, they're ingesting the data through the sensors of that device, B, they're making inferences, deep learning algorithmic-driven inferences, based on that data. It might be speech recognition, face recognition, environmental sensing and being able to sense geospatially where you are and whether you're in a hospitable climate for whatever. And then the inferences might drive what we call actuation. Now in the autonomous vehicle scenario, the autonomous vehicle is equipped with all manner of sensors in terms of LiDAR and sonar and GPS and so forth, and it's taking readings all the time. It's doing inferences that either autonomously or in conjunction with inferences that are being made through deep learning and machine learning algorithms that are executing in those intermediary hubs like you described, or back in the cloud, or in a combination of all of that. But ultimately, the results of all those analytics, all those deep learning models, feed the what we call actuation of the car itself. Should it stop, should it put on the brakes 'cause it's about to hit a wall, should it turn right, should it turn left, should it slow down because it happened to have entered a new speed zone or whatever. All of the decisions, the actions that the edge device, like a car would be an edge device in this scenario, are being driven by evermore complex algorithms that are trained by data. Now, let's stay with the autonomous vehicle because that's an extreme case of a very powerful edge device. To train an autonomous vehicle you need of course lots and lots of data that's acquired from possibly a prototype that you, a Google or a Tesla, or whoever you might be, have deployed into the field or your customers are using, B, proving grounds like there's one out by my stomping ground out in Ann Arbor, a proving ground for the auto industry for self-driving vehicles and gaining enough real training data based on the operation of these vehicles in various simulated scenarios, and so forth. This data is used to build and iterate and refine the algorithms, the deep learning models that are doing the various operations of not only the vehicles in isolation but the vehicles operating as a fleet within an entire end to end transportation system. So what I'm getting at, is if you look at that three-tier model, then the edge device is the car, it's running under its own algorithms, the middle tier the hub might be a hub that's controlling a particular zone within a traffic system, like in my neck of the woods it might be a hub that's controlling congestion management among self-driving vehicles in eastern Fairfax County, Virginia. And then the cloud itself might be managing an entire fleet of vehicles, let's say you might have an entire fleet of vehicles under the control of say an Uber, or whatever is managing its own cars from a cloud-based center. So when you look at the tiering model that analytics, deep learning analytics is being performed, increasingly it will be for various, not just self-driving vehicles, through this tiered model, because the edge device needs to make decisions based on local data. The hub needs to make decisions based on a wider view of data across a wider range of edge entities. And then the cloud itself has responsibility or visibility for making deep learning driven determinations for some larger swath. And the cloud might be managing both the deep learning driven edge devices, as well as monitoring other related systems that self-driving network needs to coordinate with, like the government or whatever, or police. >> So envisioning that three-tier model then, how does the programming paradigm change and evolve as a result of that. >> Yeah, the programming paradigm is the modeling itself, the building and the training and the iterating the models generally will stay centralized, meaning to do all these functions, I mean to do modeling and training and iteration of these models, you need teams of data scientists and other developers who are both adept as to statistical modeling, who are adept at acquiring the training data, at labeling it, labeling is an important function there, and who are adept at basically developing and deploying one model after another in an iterative fashion through DevOps, through a standard release pipeline with version controls, and so forth built in, the governance built in. And that's really it needs to be a centralized function, and it's also very compute and data intensive, so you need storage resources, you need large clouds full of high performance computing, and so forth. Be able to handle these functions over and over. Now the edge devices themselves will feed in the data in just the data that is fed into the centralized platform where the training and the modeling is done. So what we're going to see is more and more centralized modeling and training with decentralized execution of the actual inferences that are driven by those models is the way it works in this distributive environment. >> It's the Holy Grail. All right, Jim, we're out of time but thanks very much for helping us unpack and giving us the skinny on machine learning. >> John: It's a fat stack. >> Great to have you in the office and to be continued. Thanks again. >> John: Sure. >> All right, thanks for watching everybody. This is Dave Vellante with Jim Kobelius, and you're watching theCUBE at the Marlboro offices. See ya next time. (upbeat music)