Utilizing Tech - Season 7: AI Data Infrastructure Presented by Solidigm - 2x25: AI Is Going to Be Everywhere and in Everything With David Klee

Episode Date: June 22, 2021

AI is everywhere these days, powering applications from the enterprise to industrial, medical, education, and mobility. In this episode, David Klee joins Chris Grundemann and Stephen Foskett to discuss the ubiquity of AI technology today. Although not all applications of machine learning have been compelling, we are starting to see novel uses that allow us to do things we could never do before. One exciting application is in root cause analysis across the entire application stack, which has never before been possible. Three Questions Will we ever see a Hollywood-style “artificial mind” like Mr. Data or other characters? When will we see a full self-driving car that can drive anywhere, any time? How small can ML get? Will we have ML-powered household appliances? Toys? Disposable devices? Guests and Hosts David Klee, Founder of SQLibrium & Heraflux Technologies. Connect with David at LinkedIn and on Twitter at @KleeGeek Chris Grundemann, Gigaom Analyst and Managing Director at Grundemann Technology Solutions. Connect with Chris on ChrisGrundemann.com on Twitter at @ChrisGrundemann. Stephen Foskett, Publisher of Gestalt IT and Organizer of Tech Field Day. Find Stephen’s writing at GestaltIT.com and on Twitter at @SFoskett.   Date: 6/22/2021 Tags: @SFoskett, @ChrisGrundemann, @KleeGeek, @SQLibrium, @HeraFlux

Welcome to Utilizing AI, the podcast about enterprise applications for machine learning, deep learning, and other artificial intelligence topics. Each episode brings experts in enterprise infrastructure together to discuss applications of AI in today's data center. Today, we're discussing the ubiquity of AI in today's applications. And this is something that we've noticed at our recent AI Field Day event. First, let's meet our guest, David Klee. Hi, my name is David Klee. I am the founder of Heriflux Technologies and SQL Librium Education. And I'm your co-host today, Chris Grundemann. I'm also a consultant, content creator, coach, and mentor. You can learn more at chrisgrundemann.com.

And I'm Stephen Foskett, organizer of Tech Field Day and publisher of Gestalt IT. You can find me on Twitter at sfoskett and right here on the podcast every week. So we've covered this quite a lot on the podcast. We've certainly noticed it at our AI Field Day events and our other Field Day events as well. And that's actually the reason that we started this podcast and that we started doing the AI field day, because frankly, AI really is everywhere. The truth is that it's hard to think of an application that hasn't been affected by AI, especially when you look in IT operations.

But even when you look beyond that into manufacturing and medical and pretty much everything, there's AI almost everywhere. And David, that's one of the things I think that you called attention to after the AI Field Day event. So talk to us a little bit about that. Where is AI or where is it not? I think that's the better question. Where isn't AI at this point? You sit back and you start looking at basic things, embedded devices, wireless controllers, telemetry off of your cell phones,

car data, these sort of things. And when you think about it, AI is powering everything. And it may not even be AI in the way you're interacting with it, but it might be AI in terms of telemetry being collected on the background or how things are being used and processed. Nobody's got the time to manually mess with any of this stuff anymore. And quite frankly, we're all busy. They're busy. They don't have the time to sift through petabytes of data in order to find anomalies or patterns or trends. And the most mundane equipment, to be frank, AI is in there. Whereas five, 10 years ago, no. I mean, you were pattern matching, you were making rules engines to do a

lot of this stuff. And they're finding ways to simplify their operations and their management and their telemetry and their analysis of all of this with AI. And we're starting to see it exposed to the consumer, which is really cool. That's interesting. It definitely is much more widespread than it was, like you said, five to 10 years ago. And it's seeping in everywhere. I do wonder if there's a bit of a, one of my favorite quotes is William Gibson's, the future is here, it's just not evenly distributed yet. And so when we say AI is everywhere, maybe it's worth walking through some of the places where it's been surprising to find AI and possibly, you know, are there some other examples, right? Or maybe

exceptions that prove the rule of where AI isn't quite yet. Do you have any surprising examples of where you've seen AI in play? I've got a good one. Actually, a friend of mine here in Lincoln, Nebraska is running a company and they're building artificial intelligence-based detection for hospitals. And it sounds kind of interesting, but a really big challenge in hospitals is people basically trying to get out of bed on their own when a nurse is not nearby. And he's actually building a platform that is detecting when somebody is trying to get out of bed and nobody's nearby because then they can alert a nurse, the nurse comes running and they can prevent falls

and accidents. And it's actually proven to be incredibly useful. And it's such a basic thing. And I would have never thought of that. Yeah, that is super interesting. A great use case for sure. And I guess at this point, right? I mean, is there some need to look at kind of what AI means or what AI is when we say that it is everywhere? Right? Because I think there's some, at least for me still, right? Some confusion there between, you know, a lot of the talk now, you know, on this podcast and other places, once we talk about AI, we're talking about machine learning, we're talking about deep learning, we're talking about, you know, supervised training, supervised learning, and a lot of things that, you know, kind of get pigeonholed into one thing.

Whereas I think if we're talking about today, AI being everywhere, we're probably taking a little bit looser view of what AI means. And where's that line between, you know, I've got a predictive algorithm and I've got AI. Absolutely. I mean, you look at basically ask 20 IT people, what's the definition of cloud? You're going to get 50 answers. Same thing with AI. It's evolving rapidly. There are tons of acronyms out there. And for people like me that are arguably coming into this from a different perspective where most of my world is either systems or relational data, the terminology and the distinction between the different pieces of AI are fascinating. I'm trying to pick it up the best that I can, but yeah, it's definitely nuanced. And I think that that's maybe truer for you than for some other people in the IT stack,

because as somebody who is so involved in data and analytics, you know, this is one area where it's just sort of as a natural complement for AI, right? I mean, I assume that you're seeing it essentially everywhere in your specific field? Starting to, yes. A lot of what I've seen in the past has been distinct data streams with rules engines to process the data to alert on different criteria. And what I'm starting to just now, really within the last year or so is intelligence behind the analytics. And it's fascinating because it's starting to creep into different systems. You know, even SQL Server now has machine learning built into it.

You know, and SQL Server is my primary bread and butter. And it's fascinating to watch how everything is sort of appearing when it may or may not have already been there. Yeah. So that reminds me of a conversation Steve and I were actually having following AI field day as well about this kind of almost like a clamshell, right. Of AI coming into folks' lives in the enterprise. Right. And what I mean is, you know, there's, there's really two sides to it, right. One is more and more kind of on that, on the DevOps side of the house or the application developer side of the house. I think the enterprises that are using software and tools to reach their customers or their partners or their employees are more and more turning to some form of artificial intelligence to help them build better systems that they're using as the enterprise itself and maybe even as their product. And at the same time,

you've got this other wave of AI, other companies doing the same thing and then selling those products into your enterprise. And so kind of more from the IT ops perspective, to your point, right? These tools we were already using, the newer versions of them are getting infused with this. And I wonder if you're seeing that same kind of interplay of both folks trying

to use AI in their own products, but also then consuming more AI in the products and tools they're using. That's a fascinating distinction. Put it this way, I'm starting to see AI pieces get added into some of the more traditional tools that people like me, DBAs, are starting to use. But here's the challenge, because a lot of people aren't really used to this. They don't know the terminology. And if a new product comes along to say, I can do this better using AI, a lot of people simply look at it and go, but I've already got a monitoring tool. They're treating it very similar when they're very different. And when new features with AI powering the features getting added to existing technology, I see people adopting it a lot more, but otherwise they are hesitant because they don't understand

the distinction. I actually helped out with a software product. This was four or five years ago. Now they were a little bit too early to the market. It was using machine learning to spot anomalies in VMware performance telemetry. And it was a flop, honestly, because it was a new market, it was a new tool, and everybody that they approached with it just said, but I've already got monitoring. And it was brilliant, absolutely gifted. It can understand the patterns and everything going on in a virtual environment and give you the a hundred thousand foot overview, but at a micro level perspective to understand cause and effect. And it's, it's just interesting. It's really interesting. We're right on this

tipping point of starting to see more adoption of these new tools. I think that's the interesting point, isn't it? That as we're, I think all three of us saying, you know, there are already tools that do things. And bringing in an AI-powered tool that does things is not necessarily compelling unless it's doing things better or more efficiently, or, you know, unless there's some real impressive feature to it. And as you're saying for monitoring, I'm in the IT infrastructure space myself, and there are a lot of machine learning powered monitoring tools out there.

The ones that I think have the most success are the ones that are doing things you could never do. Essentially, they are collecting more data, they're processing more data, they're looking for trends and looking for anomalies that would not be visible using conventional rules. And I think that that's the challenge is that it's very easy, well, maybe not very easy,

but it is straightforward, let's say, to implement machine learning as a replacement for, you know, just rule sets. But it's actually fairly hard to implement them in a way that does something novel and different and useful. And that leads us to this whole world of essentially AI washing, where people are, you know, oh, well, we've got machine learning in this, we've got AI in this. And yet, what's the point? What is it doing? I assume that you're seeing that, you know, outside the infrastructure space, right, David? Yeah, I see it in a limited sense. But what I see is that people trying to tackle the same problems with improved tooling, whereas the challenges is not just with the same old problems.

A lot of these have been solved fairly efficiently. But what I want, and this is, I'm going to be self-serving here. I want an improved perspective on what needs to be tuned and what needs to be monitored and adjusted. For example, I deal largely in Microsoft SQL Server land. What happens if the infrastructure underneath is having a problem, and now I get SQL Server availability solution failovers randomly during the day or at night or during critical processes and it causes problems? directly unrelated, but indirectly they're related because one depends on the other. But when a telemetry stream is coming from underlying where you're focused at, I need smarts that are smart enough to understand, hey, this was a cause and effect. This problem over here had a direct relation to a failure over here. I need this stuff to work. And that's where

I'm not seeing things yet, but I know it'll get there. That reminds me of a talk I heard Tim O'Reilly give at a network operator group meeting. Man, it's probably been five or six years ago now, but the thrust of the point he made that really resonated with me at that time was this idea that not necessarily AI will replace our jobs. And obviously there are jobs that may get replaced by AI, but he was more focused on the jobs where AI will enhance it. And I think it's kind of a lot about what you're talking about here, David. And the way he described it was essentially that everyone, right, whether no matter how much of an individual contributor you are, you will become a manager. However, you will become a manager of bots, that there will be

these tools working on your behalf, doing these things kind of, you know, to your point, right? Maybe under the covers, maybe another layer down below the abstraction, maybe one layer up in the abstraction, or maybe just interacting between different systems, right? And then going out and doing these things maybe aren't your specialty, but that are related enough that you need that information or need those actions taken. So I really, I don't know those two things kind of resonate together. They do. And I'd love to have the intelligence to say this interacts with this and to be able to put them together automatically so I don't have to. This is really interesting. It reminds me of what we were talking about on last week's episode when we were talking about MLOps and the connections that are made between the various constituencies that are developing and

producing applications. And what if we get a little meta and we start thinking, well, what happens if we apply ML as a way to spot these connections? I mean, as a way to see really what's going on in a way that no human being ever could truly see because we don't have enough insight into the entire stack. I think that's where it starts getting exciting and novel, right? That's where I can start using it today to solve problems. Because I mean, quite frankly, throwing buzzwords at it, if nobody's looking at the stuff, if nobody's putting the pieces together, hey, cool, it's another tool that I'm paying a lot of money for that nobody's gonna go look at. But if I can do this and solve problems, that's where it really interests me,

especially because my world is dissimilar technologies that are related enough. And when these things cause problems, they're big. Million dollar outages, things like that. I'd love it if it could just say, hey, yeah, you had a storage pathing failover here that caused Windows to hiccup, that caused SQL server to failover, that caused the application to disconnect for 22 seconds. I'd love that. And I think it'll get there. It's just got to take some time.

Yeah, that channel, like cause and effect, kind of root cause analysis, I definitely see that at least a lot of folks seem to be in the infrastructure space seem to be kind of pushing in that direction. Right. I mean, that seems to be where a lot of at least the most useful kind of above the covers AI and ML I've seen so far. So it tends to be in that kind of observability realm of, you know, whether it's in a seam. Right. And we're looking at security logs and trying to understand what's actually relevant or if it's on a network side and you're talking about, hey, this cascading failure just happened,

but this is the domino that fell first. Those are definitely, I think, magical in the way they can speed up, at the very least, speed up our troubleshooting and maybe in a perfect world, right? Start to kind of understand when those lead dominoes are about to fall and maybe even give us some kind of predictive, Hey, there's a, there's a, you know, this, this piece is

loose. It screws loose over here. And if you don't tighten it, you know, this whole thing could fall over kind of thing. Especially. Yeah. A predictive detection and eventually a proactive remediation where I don't have to worry about it. Take care of the basic stuff. I don't want to have to get up in the middle of the night to go deal with common

repetitive issues. Let it do it for me. I'd pay good money for that. Yeah, I think most of us would pay good money for that because, you know, the idea that we would be able to get truly to root cause analysis because, you know, Chris, I think that's such a great thought. That's really kind of what I'm trying to get at here is root cause analysis. But unfortunately, so many of the applications that have claimed to do that don't really actually do it. It's more, I don't know, root symptom analysis or correlated symptom analysis.

Essentially, when this happened, this was breaking. Or when this happened, this looked like this. And maybe that's not the cause. I don't know, though, if we really will be able to get there with ML. Maybe we will. I don't know. What do you guys think? I think ML is a good start.

I think ML is going to give me more of the anomalous data on an automated basis that I can do something with. Because if I'm looking at a petabyte of telemetry data, I can't do much with that. But if I can say, give me the top 0.1% of anomalies in here, and then give me the same thing from a different perspective. So database, firewall, security, app, give me all these and then line them up in order of time. What lines up? Start to understand the patterns there. Use the tool for that. And then give me the cliff notes. I'll put those together. That's easy. It's getting the data points that I care about. And if the AI can spot the data points that matter to what I'm doing, awesome. Yeah.

Well, and that's something I've seen definitely in some of these applications like network management and security where it is surfacing, you know, machine learning is surfacing data points that we might not have known to surface or that we might have filtered out even. Or even, you know know metrics that we wouldn't have collected because we could just couldn't handle the volume of data and that i think is something that that really is um impressive because of course if there's one thing that machine learning is really really good at it's basically find the thing that doesn't match the pattern or find the thing that does match the pattern. And that basically is what we're doing with network management, right, Chris? Yeah, I think so.

And I mean, at the very least, you can save me a bunch of kind of SED and AUC and kind of shell scripting to parse through logs. You know, best case, to David's point, right? I mean, I think, theoretically, anyway, a learning system, right, a system that's using machine learning to understand kind of what I'm tagging as I go, should be able to follow along with some of these kind of, you know, troubleshooting is a great example here, I think, of, you know, what did I find relevant? And, and then, you know, if there's a way for me to provide feedback to say, okay, this is actually where the problem was, I would hope that that's something that could build up over time. You know, if you had enough users and enough different systems doing that and tagging it as it went, um, you should be able to come up with a pretty robust system for actually finding those root causes. Um, I think there was a lot of ifs in that statement, um, but I think that's definitely the direction we're going. And I'm hopeful that we'll be able to see some really kind of mind-boggling things in that space in the not too distant future. It's the prescriptive side of it. Yeah.

I can't wait to see this. And I can't wait to see the other applications outside of just core infrastructure. And I was daydreaming about this earlier today. And what if cameras in your car can detect when you're about to get hit and tighten your seatbelts, cut the wheel one way or the other so you don't run into the motorcyclist in front of you and lock your brakes, you know, stuff like that. I'd love it if it could actually preemptively

get in front of these things. And every industry has got it. My car supposedly has that, but it's horrible. It just alerts me to all sorts of stuff that's not going to happen. So yeah, I like how it worked. Yeah. Well, let it learn and train, but not react. And then eventually give it enough miles, it'll get there. But isn't that what Tesla says that they're doing with their cars? I mean, isn't this literally what they say they're doing with their autonomous vehicles? I know we're not autonomous vehicle experts, but isn't that what they're doing?

I think they're the leader in it. You know, I'm a big car nut. I think they're the leader in this so far. I think the competition is catching up, and I'm loving it because it's basically an arms race to intelligent motoring. I think it's awesome. It's just one industry. You can apply this to anything. Yeah, the automotive use case

is actually really, really fascinating for many reasons, right? One, it's something that almost everyone, at least in the United States, interacts with on some level every day, right? There's this piece of technology that is the car. And it's definitely a big part of how our cities are built and how our economy works here in the US. So it's definitely a big part of how our cities are built and how our economy works

here in the US. So it's really interesting. And I find that, you know, definitely Tesla has some really interesting stuff. And I, you know, my Tesla, just, you know, the auto drive alone, just kind of keeping the lane and keeping the right speed limit and slowing down people in front. I mean, that's, that's pretty amazing and lets me kind of have a little bit of cognitive load back um to make driving a lot less onerous than it once was um and but i do see i you know i think perhaps and maybe we're getting a little outside of our expertise here but uh i look at folks like like cruise automation who have got are going a fairly different direction right where um they're actually you know they're still using lidar even though it's it's fairly expensive and say, hey, it doesn't matter, right? We're talking about,

you know, safety in people's lives here. And their model isn't to sell cars to people, it's to have cars serve people, right? And so that's where, anyway, autonomous driving potentially upsets a big part of American life in the long run. If we're using cars more on demand instead of instead of ownership, which this is the first time where I see that actually as a possibility for American culture. I don't know if you guys are observing the same thing there or if we should swerve back into IT. Well, that's the thing. So we can swerve back into IT, but not if there's a motorcycle there. And by looking at exactly what you're saying. So the question is, are we getting

a thing that does a thing except better? Or are we getting a thing that does a thing we could never have done? And I think that what you're kind of getting at there is that, you know, yeah, it's, it's, and don't get me wrong. It is awesome to think that a car could drive itself, but it's awesomer to think what does a car that drives itself do to the entire landscape of mobility and transportation? And that's kind of what I'm trying to get at here as well with like network automation and everything is, okay, I'm super glad that your ML-based firewall can detect attackers that we couldn't detect before. That's cool. And it is. But what is it doing for us that's truly novel? What is it doing for us that could never be done before? And maybe the answer is, just to kind of pick something out

of the air here, what if there was a intelligent firewall everywhere on every network connection that was intelligently detecting, you know, that would be truly transformative from a security perspective. So instead of saying, okay, it's great that the perimeter firewall has ML and it detects new kinds of attacks. What if we had a firewall in everything? And that's what we've talked about, too, with cameras. I personally believe that no camera will not have machine learning in it in five years, because I think it's just too valuable to take video data and do stuff with it. And we'll see what that stuff is. But also think maybe uh no network connection will not have machine learning in it what do you guys think yeah that's wild right and we've definitely talked about the micronization um the way that the shrinking of ai and ai at the edge you know

several times um and and it's definitely a fascinating you know space of you know if we make these chips and this software cheaper and more accessible, and if it can be embedded everywhere and you can do inference, like you said, on the switch port or right behind it, that's pretty wild. I don't know, David, I mean, what would that look like from a database perspective? I mean, how close to the data can you get with ML potentially? You could be on the data. That's the beauty of it. You know, you could say, hey, this is a backup. We're writing a much larger packet. This thing is really intense. So give it priority. Let it go through as fast as possible. Oh, do I need to enable jumbo frames or all these

other things? Let it communicate into the app to say, hey, would receive side scaling help this process go quicker? Okay. turn this on here. Would jumbo frames help you? Turn it on here. But use the intelligence to not only have a single port with some AI in it, but to have it end to end, to be able to communicate and say, this entire network stream needs to have this enabled in there. Go, do it. Don't tell me about it, just do it. I'd love that. Yeah, and that seems to me to be a bit of an extension of computational storage, which is something that our listeners may not be aware of, but there's a whole trend to put intelligence and data processing on storage devices and even micro storage devices.

So not just like a storage array that has data processing, but drives that have intelligent processing on them. And of course, that's also the potential of a lot of these newly nascent neural engine chips that are appearing that are smaller and smaller or neural processing units that we're starting to see. So, for example, all the latest ARM chips have that. We've just heard about Intel investing in a company called Sci-5 to put RISC-5 based AI and ML processing in embedded devices, including disk drives and disk controllers. So yeah, I think that this is so totally happening, right, Chris? It sounds like it. Yeah. And that's definitely wild. It reminds me of there was, I think it was an IBM paper. Attribution's not accurate here, but where they were looking into kind of software-defined networking before we were calling it software-defined networking. And their vision, though, was actually that you would potentially run applications in the network, right? And so that the packets themselves would actually trigger events as they pass through the network,

right? Which I think is kind of what you're talking about, David, in some ways, right? Which is that there's some kind of meta information or actually maybe programming logic that's actually attached to the packets as they move through the network. And then routers become servers and switches become servers. And this whole thing becomes really intelligent. It's definitely a little mind boggling. But and something that just didn't seem like it could quite happen before.

I mean, obviously, theoretically, it's kind of an interesting idea to talk about these smart packets that are, you know, actually running a live application as they move through the network. But it didn't seem possible to get there. And maybe, right, Stephen, with some of the advances you're talking about, this is actually a time where we can get small AI into these devices in a way that this is actually possible, where instead of having to have a bunch of information in the packet itself, you could have just enough that would trigger these AI models as it went along to do some interesting things. I don't know, that that could get to a place where you get

really transformational technology. I don't think it's a matter of if, I think it's just a matter of when. And I look at it this way, we have to get there. Look at the volume of data that's floating around from IoT or just the cloud moving data around globally. We have to get there. We have to be more intelligent with this stuff. And I think AI really at this version of the edge is the best way to get there.

It's such a good time to be a geek. You there. It's such a good time to be a geek. You know, it really is a good time to be a geek. And there's so much stuff happening, so much exciting stuff happening. And so it's fun to be able to geek out with you folks. And I really appreciate you joining us here, David. Before we go, though, we do have three questions for you if you're ready for that. Let's go. All right. So, so here we go. Number one, will we ever see a Hollywood style artificial mind like Mr. Data or characters similar to that?

Sure. You know, it's going to happen. The question is when the question is, what is it going to do for the task it's given or humanity or anything around it. It's just a matter of when. Number two, how about this? We talked about self-driving cars, so I got to put the self-driving car question in there. When will we have a car that can drive anywhere at any time with no human intervention. I'm going to predict 15 years from now because of the legalities around what happens if it messes up, not because of the tech. I think it's going to take longer than people realize. That's probably true. And I think that that's actually been the consistent, the consensus answer here on the

podcast. Whenever we ask this, the answer has always been, well, not as soon as you think. Exactly. And another thing that we kind of mentioned in our conversation here, and I'd love your kind of your opinion on it. How small and cheap and ubiquitous will ML really get? Will we have ML-powered household appliances? How about toys?

How about disposable ML? Look at how technology has shrunk and gotten cheaper over the last 20 or 50 or 100 years. It's bound to happen. They were already talking about smell sensors in your refrigerator to detect if something on a shelf is going bad and to alert you of this. Next thing you know, you're going to have AI-powered ballpoint pens.

It's going to happen. They're going to find a way to put AI in everything. It's going to be ubiquitous. I would love AI-powered buttons on my shirt to let me know when one pops out in the middle of a presentation. Call me weird and connected like that, but it's going to happen. How soon, how long, to what extent, don't know, but it's happening. And that's the thing. It's happening a whole lot faster than we realize because AI is becoming a part of everything, whether you see it or not. Well, I think that that certainly is the case. And that really comes back to the title of the topic of this episode, the fact that AI is going to be everywhere and in everything, and we better get used to it.

So thank you so much for joining us, David. It was a pleasure to have you here today. It was a pleasure to have you at our AI Field Day event as well. And we look forward to continuing this conversation. So if people want to continue talking with you, where can they connect with you and follow your thoughts on enterprise AI and other topics? You can hit me at Twitter primarily and go figure.

My name is Klee Geek on there because I've embraced that for a long time now. Or you can hit me at my blogs at davidklee.net or hairflux.com or out on LinkedIn at David A. Clee. And you can find me on Twitter at Chris Grunewin or online, chrisgrunewin.com. Thanks a lot. And of course you can find me as S. Foskett on most social media networks. And you can find me every week here

at the Utilizing AI podcast along with on Wednesdays with the Gisht Altalt IT news rundown. So just go to gestaltit.com for that. So thank you for listening to Utilizing AI. If you enjoyed this discussion, please do give us a rating, review the show on iTunes. It's been great to have some of our listeners do that since it really does help.

Also, please share this conversation and this podcast with your friends, other people that might be interested in these topics. This podcast is brought to you by gestaltit.com, your home for IT coverage from across the enterprise. For show notes and more episodes, go to our website, utilizing-ai.com, or find us on Twitter at utilizing underscore AI. Thanks, and we'll see you next week.