 from Las Vegas, it's theCUBE, covering AWS re-invent 2018. Brought to you by Amazon Web Services, Intel, and their ecosystem partners. We are live here at the Sands Expo, one of eight venues that are actually here for AWS re-invent this week as we continue our three-day coverage here with you Tuesday, Wednesday, and Thursday as well live from Las Vegas here on theCUBE. Along with Justin Warren, I'm John Walls, and we're now joined by Phil T, who is the CEO and co-founder of Moogsoft. Phil, good to see you this afternoon. Great to be with you. Thank you for inviting me along. You bet, nice to have you. In fact, I'm going to be the only guy without a charming accent on this segment, as a matter of fact, you, Kay, and Anasi here. Your world's data, right? And it kind of reminds me of the old movie Jaws when Robert Shaw says we're going to need a bigger boat. Right, AIOps, is that your bigger boat? I mean, is that how you're handling this world of data? I think that's exactly spot on. And one of the things that we observe at Moogsoft with our customers is just this crazy complexity that they have to deal with. I mean, we cover everything from large financials, telephone companies, e-commerce businesses, and the drive to adopt agile and cloud and software defined in the enterprise has driven complexity to the point where the poor old human brain is just out of luck, right? The unaided sort of, I'll figure it out by myself approach, dead in the water, and you've got to use this artificial intelligence approach precisely as your bigger boat to go catch that shark. Right, so tell us, there's a lot of hype around AI and machine learning and all of these different buzzwords are getting thrown around. Dial us in a little bit to explain what you mean by AI and in the context that you're talking about things with Moogsoft. So I think that's very perceptive. There's a tweet going around at the moment which describes the difference between machine learning and AI as, you know, if it's written in Python, it's probably machine learning. If it's on a PowerPoint deck, it's probably AI, which is kind of funny, but to the point. And there is a ton of hype going around artificial intelligence. There are some purists who claim there is no such thing as AI and everything that we talk about today is really machine learning or data-driven algorithms. And there's some truth in that. Really what we mean by AI is the full panoply of both feature detection, you know, looking for patterns that are not obvious to the human eye. All the way through to deep learning neural nets, convolutional neural nets, where you are training a system to recognize features of the data as representative of something underlying that you're hunting for. So in the case of AI ops, it's looking for the cause of or looking for the presence of a potential sort of service impacting outage. In the data that we monitor, in the events. But one thing it's not going to do is it's not going to unplug itself from the internet and come and kill you anytime soon. It's really quite benign and very useful to our customers with what they deal with. Okay, so I mean, to that point, because you have so much data and it seems like I hate to say most of it isn't needed or most of it isn't of value, but a lot of it isn't, if not most. How do you then discern, you know, how do you assess value and assign value to what really is important and then put it to use today when you're getting so much more information than you were even a year ago? So, and just to put a little bit of context on the amount of data. So, you know, way back before cloud and virtualization, you know, a typical enterprise, a high event rate would have been, you know, 100, 200 events a second. You know, nowadays in an average customer of ours, you add a zero or two to that rate, maybe even three. And it's kind of one of the reasons why a lot of the legacy systems really struggle with that data. So, you know, job one is, you know, if you accept and I certainly do, that the most of that data is junk, most of it is inconsequential. You've got to have an algorithmic way of getting rid of that. You know, the old fashioned way was creating lists of, ignore it because it's a certain severity, ignore it because it comes from, you know, this list of hosts, you know, the whole listing approach. What we do is we use information science so we can measure the semantic content and the informational content of an event to work out whether it's telling us something of import. And we use that technique with great effectiveness to eliminate as much as 90, 95% of the in-band data is effectively affecting nothing. So that narrows the data lake, if you like, down to the point where we can process it in real time through much more compute-intensive AI algorithms to kind of get that high quality indication of an incident or a potential incident. Yeah, well, a lot of machine learning in AI is based on learning from history. So we've seen all of this stuff before and we know what that means, or even encouraging the machines to go and look at historical data to then pull out the details. As you said, even things that a human might miss, you'll look at that data and then learn new things. How does that work when we're doing all of this innovation, when there's all of this change and novelty coming in? How does the AI system cope with that kind of environment? So you have to have a dual approach. So I mean, I guess everybody's familiar with Nicholas Nassin-Talib's book, The Black Swans. He was trying to explain why it is that you can get a bunch of Nobel Prize winners in a room to design a hedge fund and it can still go bankrupt in the blink of an eye, the long-term capital management. And the truth of the matter is, is yes, an awful lot of the techniques that are supervised and based upon essentially the training set are vulnerable to the kind of the unknown unknowns to misquote Donald Rumsfeld. And that's why we use a combination of unsupervised feature detection and supervised learning. The unsupervised feature detection just knows something as an unusual, highly correlated pattern or feature in the data and needs no prior understanding of what's going on. Now interestingly, there are some hybrid techniques now. You may have heard of something called transfer learning, which is the idea that you sort of partially train a neural net on some kind of standard corpus. If you're like the stuff that you already know and adapting that sort of partially trained net to something that is kind of very, very, very adapted to the system that it's monitoring, it does that very quickly rather than having to wait for a certain critical amount of data before the net is converged. And so, you know, those sort of techniques which we also experiment with are MOOCs off, I think are going to be interesting directions for us in the future with our platform. But, you know, there's, I don't know, maybe 100 PhDs a week given out in AI and machine learning these days is definitely getting a lot of focus and there's a ton of innovation that's coming down the line. One thing that we're particularly committed to is kind of shortening the distance between when something's invented and when we can get it into our customers' hands. Right, there's usually quite a lag. I mean, historically, it's about 10 years before someone discovers something and then it actually makes it into the business world. So we could shorten that cycle that would be quite useful. So we know an academic called Professor Maggie Bowden who's just getting ready to retire and she was one of the original authors of the neural net papers in the 1960s. So that kind of gives you an idea of the lack, right? So, you know, it can be many, many decades. And it's a shame because the truth of the matter is the pressure on all the people coming to a show like this that want to benefit from the public cloud, new ways of thinking about the application development tool chain, they don't have time to wait around for that innovation to come to them. We've got to drive it a lot faster and certainly we view that as one of our missions at Moogsoft as being passionately involved in sort of shortening that gap between innovation and a production implementation as something really cool. So what have you seen at the show so far that you think that you want to take to your customers and say, well, actually, this is happening. You need to get onto this now. So one thing I've observed here is, you know, I guess if we would have been here two years ago, nobody was talking about AI ops. There was, you know, I mean, essentially, you know, the entirety of how people looked at the cloud was same old stuff, just live somewhere different. You know, we don't need to, you know, we can use all of the old techniques. You know, you walk around here, there's a bunch of startups, more established companies, you know, recognizing that a new approach is necessary. And, you know, my sense of it is, is that this market, which, I mean, let's be honest, we were pretty lowly in it two or three years ago, is starting to feel like it's a little bit more populated. And, you know, that's goodness. You know, we're very happy about that. So that is definitely a takeaway. You know, to go to customers and say, you know, this is no longer bleeding edge, it's simply leading edge. Not just a gap in the market, there is actually a market in that gap. Yeah, yeah, yeah. You're at the bigger boat. Well, we hope so. All right, Phil, thanks for being with us. We appreciate the time here on theCUBE. And once again, have a great show. And we do thank you for your time, sir. Thank you very much, Indy. Great talk to you both. From Moogsoft, joining us here on theCUBE we're at AWS re-invent and we're at the Sands and we're in Las Vegas.