Welcome to the Cloud Native in 15 Minutes podcast. If you haven't listened before, the goal of this podcast is simple, to explain complex but increasingly important and even strategic technologies and about the time it takes for a coffee break. We're all busy, but that is not an excuse to not understand, at least at a high level, some of the fundamental technologies that will be powering your company for the next decade. Previously, if you haven't listened, we've covered Kubernetes and Apache Kafka and MongoDB, and today we're going to be talking about artificial intelligence. I'm Derek Harris from Pivotal, and joining me to talk about artificial intelligence is Andrew Ng. If you're not familiar with Andrew, just Google him, and you'll find a litany of his comments and thoughts, because he has been a leader in the AI space for quite a while now. Andrew co-founded Coursera. He was a Stanford professor, early leader on the Google Brain team, the chief scientist at Baidu, and presently he's running a trifecta of companies, Landing.ai, which is a company which helps enterprises get started with AI, the AI Fund, which is a venture capital fund for AI companies, and DeepLearning.ai, which is an educational company. So Andrew is all about getting people educated on AI and getting them actually up and using it. In this podcast, we focus specifically on AI and the enterprise, and Andrew explains why we're hearing so much about AI right now, including how a specific technology called supervised learning is really what's driving much of the enterprise and commercial success with AI. And I should tell you, supervised learning is not something out of science fiction. It's some heavy math, but it is very much pattern recognition. Andrew also discusses the importance of properly scoping AI projects and the importance of starting small in order to deliver early wins. And then we also, we talk about a lot of stuff, but also how enterprise AI adoption is going to benefit from advances in education and tooling,
2:00higher levels of abstraction, better services, et cetera. So listen, buckle up and listen to Andrew Wing talk about AI and the enterprise now. Okay, let's get started. Andrew, can you briefly explain the techniques most people are talking about when they talk about artificial intelligence? Artificial intelligence or AI is a portfolio of techniques, but there's one of them called supervised learning that's driving 99% of the economic value that you see AI creating in the news. All supervised learning does is a way for computers to compute input outputs or A to B mappings, such as input an email and output, is this spam or not? Or input an audio clip and output the text transcript? Or input an ad and output, how likely is the user to click on this ad? And by putting these applications in the right business context, this is used in everything from spam filtering to speech recognition to showing people the most relevant ads on the large web search platforms. And this is creating tremendous business value. All right. And when you talk about supervised learning, is that related to deep learning, which maybe is the broader term that gets more attention? Yeah. You know, there's so many buzzwords like deep learning, supervised learning, unsupervised learning, floating around. Now, deep learning is maybe one of the hottest technologies in AI right now. And what we've found over the last several years is this is a wonderful technique for doing supervised learning. So, for example, if you are running a factory and you want an AI to look at the smartphones going down your manufacturing line and figure out if it is a scratch or not, you know, to do automatic defect inspection, then it turns out that deep learning is a very effective way to learn that input-output mapping, where the input would be, say, a picture of a smartphone, and the output will be, does the smartphone have a scratch on it or not? And a lot of recent rise of AI and supervised learning is driven by discoveries in deep learning,
4:03allowing us to make them build very accurate input-to-output dynamics. Okay. Interesting. And how does AI relate to other, like, you know, I'm going to say AI, you know, just broadly speaking, right, not going to get too deep into the terminology, but, like, how does AI relate to other, I think, big trends that IT execs have had to think about over the past decade? I think of big data and data science, cloud computing, IoT, like, I guess I hear, I feel like I hear AI, you know, lumped into all these discussions. I'd love to get your thought on, like, where that fits in. Let's see. The rise of a lot of earlier waves of technology, which were also important in and of themselves, such as the rise of cloud, the rise of big data, I think they helped set the stage for the recent rise of AI. We as a society, we've been just generating a lot more data, more data is aggregated in the cloud. And we actually, as a society, got a bit ahead of ourselves in terms of the data we collected relative to our ability to have computers make sense of this data. So part of the recent rise of deep learning was when we collected so many days, so much data, not just in the software internet world, but sometimes even in manufacturing or agriculture or healthcare. And the recent rise of AI algorithms, specifically deep learning algorithms, is letting us process this data in more effective ways. And this is driving tons of exciting applications, ranging from harvesting machines that are more efficient for farmers to use, to AI-assisted radiology diagnoses, to many, many more. Okay. And out of curiosity, when you talk about AI and how we got ahead of ourselves, I mean, is it now a situation where, like I said, we have data and we finally have, if I'm a CIO thinking about how I want to do AI and I want to use a cloud service, for example, I mean, is it a matter of just like we have more computing power?
6:03Is it just easier to do some of these things we've been wanting to do? The rise of data and compute has driven a lot of the rise of AI. The story is now bigger than that. And AI is also powered by the rise of availability of talent. For example, over two and a half million people have signed up for my machine learning course on Coursera, the one that actually started Coursera. So there's now a huge talent pool, excited people, motivated. They often study it hard in their own spare time. They want to go work on all these problems. And open source research papers, all of these things are trending up into the right, growing very rapidly. And so there are now wonderful tools for CIOs and CTOs to adopt AI projects. One of the hardest things in building an AI project still is scoping the project. It feels like there are a hundred or even a thousand things you could work on. How do you pick the most promising, you know, two or three to start on? So one of the things I published recently was an AI transformation playbook on Land AI's website, which gives some suggestions to CIOs, CTOs, also CEOs on how to choose your first few AI projects and how to help a company become good at AI. So I was going to, I wanted to ask you about that. Are there any low hanging fruit that most companies should think about? Because I think broadly when people think AI, you know, some people's brains go to this kind of fantastical thinking. And then, but really, it seems like a lot of the applications are pretty, are kind of mundane in a way, right? I mean, are there like, but are there low hanging fruits you think that most organizations could easily start with to prove out the value of something and then, you know, go deeper and deeper as they kind of prove out what they're working on? I've seen a lot more companies fail by starting too big than by starting too small. So I recommend to most companies to start with a few pilot projects, deliver a quick win in six months,
8:06and then use that to ratchet up to bigger and bigger projects. Even Google, when I was starting out the Google Brain team, there was a lot of skepticism about deep learning at that time, both within Google as well as in the wider world. Well, much as today, a lot of companies may still be uncertain how to use AI. So my first internal customer, you know, the Google Brain team was Google's speech recognition team. And speech recognition was not, maybe still is not Google's most important AI application. It's not web search or advertising. But by delivering a quick win there, it helped other teams within Google gain more fame in what, you know, my team and I could do. Then we got our second customer, which was to help Google Maps use AI to read house numbers to more accurately geolocate houses and buildings on Google Maps. And only after delivering this second win to a second internal customer did I then go on and start the more serious conversation with the advertising team. So I think the lesson to a lot of companies is start small and try to pick a project that can allow you to deliver a quick win, because that often helps the organization learn how to do AI and opens the door to them doing bigger projects over time. All right. Are those applications usually specific to the organization, do you think, or any other things? I could imagine some kind of broad, let's say, computer vision applications or other things, some sort of text recognition, right? That might be kind of useful across a broad swatch of organizations. Yeah. In the AI transformation playbook, I recommend to companies to try to do something that's specific to your business. There are going to be cloud enterprise solutions that you could buy, much as today, many people buy an email service or buy a CRM system. But AI would transform the core of how many industries operate.
10:08So I think that for many industries, there is a little bit of an existential thing to figure out how to use AI effectively for your industry. So it's important to start with some projects that knit more closely to your business, because the way that AI would transform most industries is not just some SaaS enterprise thing that everyone could buy, although there will be some of that, but it would transform the core of how we do manufacturing, how we do agriculture, how we do healthcare, and probably every major industry. Okay, so broadening out a little bit, even from the applications, you mentioned the AI playbook you have. You also have this Coursera course called AI for Everyone that tries to explain, she does a really good job explaining the fundamentals of AI to, let's say, executive types. What do executives really need to understand at a core level about AI in order to start effectively implementing it? So AI for Everyone was designed for a non-technical audience, including a business audience that may want to learn enough about AI to either work with AI engineers within their own organization or to navigate the rise of AI across multiple industries. So it talks about some of the buzzwords of AI, so what is machine learning, what is AI, what is deep learning, what are all of these different, how do these things relate to each other? And also, it spends a lot of time talking about how to scope AI projects so that if you're looking within your own company, you can start to understand what it would feel like to pick a project, maybe work with engineers. And I hope that it'll help many executives or other business leaders work better with AI teams to drive successful AI projects through their organization. You know, there's a lot of hype about AI, and I think a lot of executives understand that this is a critical time for them to jump onto it and, you know, be the disruptor rather than be disrupted by someone else. And AI for Everyone tries to clear the way and teach people what they need to do that.
12:12Yeah, I thought one of the, as I think about it, one of the more useful things that I think was explaining, you know, the differences between, let's say, data science and AI, because I think a lot of organizations went and built up data science teams over the past decade, let's say, and then AI comes in and it's not necessarily a one-to-one mapping of those roles, right? And there's all these little things that you don't, that get conflated, but don't necessarily, that aren't necessarily performing the same function, so. Yeah, and just the machine learning versus data science. For example, you know, I've worked on projects where we build a machine learning system that would run many times per second to show people the most relevant ads. And I think showing people ads is not the most inspiring thing to do of your life, but it tends to be, you know, very important. You can make you money. And then on the flip side, I've also had data science teams whose job was not to build an AI system that would run 24-7 and produce, you know, outcomes every fraction of a second. But I've also had data science teams analyze data and then come up with a conclusion like, hey, Andrew, did you know that if the company sends more salespeople to the travel industry, maybe we should try to sell a lot more online ads to the travel industry? So that would be an example of such a great insight that, you know, was produced by one of the data science teams I was working with that drove business results. So I think both of these are important. Both the machine learning algorithms will result in a piece of software that runs 24-7, as well as something called data science projects, which results not in a piece of software, but in a PowerPoint deck. It's really a presentation that changes how you prioritize a business. And these are two of the ways that AI, machine learning, data is treating a lot of value for companies today. All right. And then on a related note, how important is it to have actual AI experts in-house versus, say, developers who understand AI and have access to, you know, like Google or AWS cloud services, right? So there's kind of a discussion going on right now, right, about hiring for AI.
14:17Yeah. This is the old build versus buy question, but, you know, applied to the AI era. There are some projects which will be industry standard and will be more efficient to just buy it. So I think most companies today should not try to build their own cloud. Most companies today should not try to build their own, you know, deep learning framework because there are giant companies investing huge amounts into that. So those are some things you should just buy. But in terms of building, there are going to be things that are specific or maybe at the core of your business. For example, for Google, web search and advertising are at the heart of Google's business. And so those just had to be built internally. And it'll be different for different companies. I think for some companies, it may be possible to work with some outsource partners to jumpstart that process. For example, there are manufacturing companies that work in some of my teams to build AI solutions for them. But then over the long term, for companies to become really good at AI, eventually, you know, to do the custom things that are very unique to your business, eventually, I hope a lot of companies will build an in-house AI team. All right. Do you think that's a matter of time in the sense of like as the abstractions get higher and better? Because what I can't imagine, I guess, is a bunch of, is every organization having like, you know, and maybe this will change, right? But AI experts and researchers in-house, you know, and, you know, like heavy mathematician types. But it seems like the tooling will probably go, hopefully evolve, right? And the abstractions will evolve to some degree to make that more broadly applicable. Is that kind of you see it going? Yes. And we've seen this already with the rise of, you know, IT. Today, a lot of companies have an IT division. They have, you know, really efficient, pretty good computer networks, a lot of companies. And then also with the rise of the internet. Today, a lot of companies have a software engineering team. You know, many retail companies, not just the giants, but much smaller ones, have software teams that can run their own websites.
16:21And I think AI, as the tools become more available, as things like, you know, causes offered by DeepLine.ai and Coursera and others help educate more people in AI, I think it'll become more accessible to more organizations to build their own in-house AI team. So today, you can go to even a mid-sized retailer and they may be building their own custom website to promote their products in their own way. In the future, I would love for that type of company to also have their own in-house AI team. But it will take a while. All right. And then, I mean, as we talk about some of these industries, you mentioned manufacturing before, and even retail gets a lot of this. Like, I'm curious about how you think about the relationship between AI and automation. Because it seems like, I think sometimes we conflate, like, automation via software with automation via robot, and then with artificial intelligence. Like, how do you think about the nexus or even the disconnect, I guess, between those two terms? Today's AI is automation on steroids. AI can do more than automation. We look at a lot of values creating. AI is taking a lot of tasks that people can do and helping engineers write software to do that automatically. One of the rules of thumb that I give teams is almost anything that you could do in less than a second of mental thought could probably now also automate using supervised learning. And while there are AI tools that do more than just a step of automation, in terms of executives trying to find ideas for projects to do with AI, starting with thinking through what you can automate, which isn't just physical automation. They're going to be able to also be automating little bits of mental work is a great starting point. Maybe one example, if you have security guards watching surveillance videos, that's one task that I think is amenable to at least partial automation. Instead of having security guards watch tons of videos to see if anything went wrong, I think AI could certainly help narrow down the set of videos you need to look at.
18:22Instead of having radiologists diagnose x-rays from scratch, I think pretty soon we're increasingly seeing AI not replacing, but assisting human radiologists to help them do so more efficiently and maybe more accurately. So I think there'll be lots of opportunities in many industries for AI to help with either partial automation or sometimes full automation. Okay, and it doesn't have to involve robots, right? I mean, I think that's just the people say robots a lot or show robots a lot. And it's just like, they're not the same thing. Yeah, I think a lot of AI work has nothing to do with physical robots. Part of it does, but most of it does not. Yeah, exactly. And then finally, Andrew, like beyond some of the bigger picture, you know, existential concerns that have gotten a lot of news over the past couple of years. I mean, are there other gotchas that enterprises should be specifically should be cognizant of when they're going on the AI path and kind of, you know, holes you don't want to don't want to fall into? Scoping the right AI projects is really hard. And there are a few gotchas on identifying the right thing to work on. One of them is just unrealistic expectations about what AI can and cannot do. For example, a few years ago, a lot of people thought chatbots could soon have fully general purpose conversations and talk about almost anything with anyone. That turned out not to be true. And companies that tried to build fully general purpose chatbots just did not succeed because the technology was not there. So I think it's important for executives to have a little bit of an understanding. So I think it's important for executives to have enough of an understanding of what AI can and cannot do to scope the valuable and feasible projects. And then also to have the appropriate mechanisms for staffing them, tracking them and helping set up a machine learning team for success.
20:22All right. Great. And with that, we are out of time. Thank you very much, Andrew. Appreciate it. Thanks a lot, Derek. It's always a pleasure. All right. There you have it. I hope you learned something. I certainly did. For more information about any of Andrew's various products, you can check out deep learning dot AI. You can check out landing dot AI or you can check out AI fund dot AI. If you visit Coursera, you can find it as AI. You can check out his AI for everyone course. If you want to go deep into the weeds, you can check out some of his earlier machine learning courses. And obviously, if you're interested in learning about artificial intelligence and even getting started writing, writing some AI models and deep learning models, you can find there. There's this plethora of information out there today to track to track down. So you can you can Google that and figure that out for more information on on Pivotal and our broad suite of products and services for building and building and managing cloud native applications. As well as transforming how you build software, check out Pivotal dot IO. And if you like this podcast and you want to keep up on kind of all things digital transformation without, you know, without getting heavy into products or or vendor spin, please check out our blog. Pivotal dot IO slash intersect, which is focused on helping business leaders, executives and other folks who aren't deep into the deep into the technology. Pivotal dot IO and get smart about how the technology works. Thanks for listening.