Utilizing Tech - Season 7: AI Data Infrastructure Presented by Solidigm - 2x22: Microsoft is Democratizing AI with Steph Locke

Episode Date: June 1, 2021

Microsoft plays a large role in enterprise IT applications, from the desktop to the datacenter to the Azure cloud, and the company is active in the world of AI as well. But most of Microsoft’s work has gone unnoticed, with high-profile cloud AI and ML applications at companies like Google and Uber getting all the press. In this episode, Steph Locke joins Chris Grundemann and Stephen Foskett to discuss the place of AI inside the Microsoft ecosystem. Microsoft has built AI into search and Cortana and has also produced an AI Builder and ML workspace in Azure that allows developers to produce their own chatbots, recognize images, and more. Steph also discusses the AI-related announcements at Microsoft Build last week. We finish up with a deep discussion of accessibility and diversity and potential solutions from hiring to training to awareness. Three Questions Can you think of an application for ML that has not yet been rolled out but will make a major impact in the future? Is it possible to create a truly unbiased AI? Will we ever see a Hollywood-style “artificial mind” like Mr. Data or other characters? Guests and Hosts Steph Locke, Data Scientist and CEO of Nightingale HQ. Connect with Steph on LinkedIn or on Twitter @TheStephLocke. 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: 6/1/2021 Tags: @SFoskett, @ChrisGrundemann, @TheStephLocke, @NightingaleHQAI

Welcome to Utilizing AI, the podcast about enterprise applications for machine learning, deep learning, and other artificial intelligence topics. Each episode brings together experts in enterprise infrastructure to discuss applications of AI in today's data center. Today, we're discussing Microsoft and what they're doing in the world of AI. First, let's meet our guest, Steph LaHocke. Hi, Stephen. Thank you for having me. I run Nightingale HQ. We're an organization helping manufacturers adopt AI and primarily inside the Microsoft space because almost

everybody in enterprise has Office 365 or something like that. So it's the place to be adding AI capabilities for everyone, really. And I'm Chris Grunman, your co-host today. I'm also a mildly enhanced cyborg, a consultant, content creator, coach, and mentor. You can learn more about what I'm up to at chrisgrunman.com. And I'm Stephen Foskett, purely organic, organizer of Tech Field Day and publisher of Gestalt IT. You can find me on Twitter and those social media networks at sfoskett. And of course, you can find me here on Utilizing AI

every Tuesday. As Steph mentioned, Microsoft's ecosystem has certainly been a major player in enterprise IT for decades. In fact, these days, between on-premises and Azure in the cloud, Microsoft makes up a large percentage, maybe even more than half of enterprise IT application platforms. But of course, we don't usually talk about Microsoft when we're talking about artificial intelligence in the enterprise. And that's one thing that I was interested in when we started talking to Steph here, because of course, she knows a lot more about that than I do. So Steph, tell us a little bit about Microsoft's place in the world of AI. So Microsoft have a big vision about democratizing AI.

And you can actually see this in place right now with a lot of things that you might be using since COVID hit. The live captions inside Teams, the autocomplete on your Word emails, on your Outlook emails, and even the design ideas in PowerPoint automatically making your slides look great. So they're really, as well as giving us as data scientists and AI engineers, some really great capabilities in the cloud. focused on is making AI highly utilized and efficient for day-to-day use by non-technical people. It's really interesting to start talking about kind of those little enhancements that are here and there throughout the Microsoft ecosystem, just because I believe that, you know, a lot of AI, at least really well done AI, kind of goes unnoticed. And I think that folks kind of fail to appreciate how much model generation and inference and all the things is going on behind the scenes when, you know, those little helpful hints pop up. And it's definitely a long

way from the days of Clippy. And now it's just kind of really seamless and integrated and integrating these enhancements put in. So I wonder from that, I mean, do people need to consciously be aware of this or is this something that really should stay in the background and just kind of enhance our workflows? I try and encourage people to think about what is strategic to your business and what is an enabler of your business.

So your strategic pieces for instance in a manufacturer might be the data specifically about your products and your factory floor and your setup and having ai that you train specifically to solve that problem is much more strategic in nature than the kind of back office productivity and chat bots to help your suppliers get answers to questions sooner so i'm really uh quite happy with how microsoft are trying to make that back office low horizontal capabilities piece, very seamless, because it doesn't make sense to spend your rare data engineer, data scientist and engineers and software devs, putting these things, making these things and putting them live when it doesn't align to your business goals. Yeah, I was kind of half joking in the intro there where I said, you know, I was a mildly enhanced cyborg, but it is something that I've been talking about for a long time, which is that, you know, even just simple tools, you know, forget about AI,

but just things like Outlook and, you know, Note and a lot of these other products that are out there on the market have been enhancing what I can do as an individual for a long, long time. And so I see this as kind of a continuation of that in a really interesting way where these kind of, you know, personal digital assistants in a way, right. But, but, but in, in specific niches and specific use cases kind of popping up to help where they can is really interesting. Right. And just, I can imagine that there's got to be some productivity gains kind of across businesses, but, but maybe across society at large, as these things kind of roll out and become more and more part of what we do every day. Yeah, that's a really great point. And I think that we're all mildly enhanced cyborgs, aren't we, Chris? Because of course, all of us are using

these digital assistants every day. And as Steph mentioned as well, many of us are using AI-driven components to applications without even knowing it. I mean, frankly, I didn't know that PowerPoint used AI. And I imagine that there's probably lots of other instances within the Microsoft ecosystem of sort of AI co-pilots that we don't even know about, right, Steph? Yeah, just yesterday I noticed that they're using computer vision on images and things that you put inside your OneDrive to surface the text in the images as part of your organizational search. So SharePoint is like it or hate it, depending on your kind of perspective. That knowledge sharing platform is really key to most businesses because almost all of us using technology are knowledge workers in some way.

And a huge portion of our day is spent trying to find information from other parts of the business or even what you did six months ago. I don't remember what past me did. I have to look it up. So the fact that we're getting search integrations that are using AI to extract data from things that we previously wouldn't be able to get that info on is really useful. And it's actually getting made better with some announcements that they're making at Microsoft Build, which is the 25th to the 27th. So in the past, when you're listening to this, They're making some announcements about how they're going to be surfacing your Azure Cognitive Search and your Dynamics 365,

so your ERP and your CRM data and Salesforce and things, and actually making that more readily integrated and searchable across your organization. Inside even your Windows search bar. And the great thing about that is that hooks into things like Cortana. So you can now do the speech to text to be, to be able to ask questions of things like your business data to ask Cortana, how many sales did I make yesterday?

And get that answer through the combined data sets inside your Power BI, your Salesforce, and your online capabilities and get that immediately. So the productivity gain that we can be getting through just that seamless AI under the hood is phenomenal for us. I was going to ask, again, since you're more familiar with this, apart from sort of the search functions, are there specific AI tools or workspaces available within the Microsoft world beyond search? Yeah. So inside the, even at the kind of business user, non-developer space, Microsoft inside their power platform suite have their AI builder capability, which integrates a lot of their computer vision, natural language processing, and speech capabilities so that people can build things like accounts payable invoice processing, they can build their own chat bots to answer questions and perform processes and even in be able to OCR text from things like forms and stuff and be able to turn that into business

meaning. So there's a lot at the kind of non-developer level and then inside the kind of more ai engineer data scientist capabilities in the cloud we've got the azure machine learning workspace which enables you to do machine learning operations or MLOps for the full life cycle of model development, production and monitoring, as well as sandbox virtual machines for trying things out, notebook capabilities to be able to get up and running quickly, and a load of off-the-shelf composable AI capabilities. So for instance, inside the Azure Cognitive Search, Azure Cognitive Service called Form Recognizer, it now has identity document recognition. So you can do instant compliance inside your app for verifying identities and even do some facial recognition using their face service to say,

does the face in the identity document look like the face that I'm seeing through the camera? So really powerful from developed, from kind of non-technical through to developer, through to AI engineer capabilities. Awesome. I mean, it's really wild to see, you know, sometimes we talk to folks here on the podcast about things that are kind of, you know, a little bit in the future. A lot of things are kind of developing and happening, but are things you would see in the future. And this sounds really grounded

in kind of practical reality today, which is pretty amazing. Obviously, you know, there's a lot of capabilities you talked about just there. And I'm wondering, you know, maybe just more specific into Nightingale and your company and what you're doing to help folks. Is this really about just like adoption of those Microsoft tools and getting the most out of those? Or is there more to it on that kind of path to finance being a decade or two behind in tech in some areas. Manufacturers are very further behind in finance in most cases, like top-notch operational technology. You know, in software engineering, we have things like Kanban and lean and all sorts of things that

we've kind of lifted from the operational side of manufacturing but there's not been a lot of widespread software adoption inside manufacturers so they need to start increasing their operating margins by decreasing their overheads. And that's things like reducing downtime of machines. You know, we can use machine learning to identify before there's a problem, when there's likely to be one and schedule downtime for a quiet period and things like logistics transport is their second largest cost and kind of managing goods if we can help them optimize that so that they reduce how much stock they hold how much they transfer from their suppliers and to their customers and there's just huge amounts of gains that manufacturers can benefit from.

Um, and I started looking at ways that we, the, what is the right approach for somebody who might not even be in the cloud as an organization who have staff who are excellent at their jobs, but have been usually doing it for 15, 20 years. In the same way what is the right approach to get that business getting those gains from software and the answer isn't a data scientist and a big it team that just is a huge amount of investment that doesn't really gain traction it's those small quick wins that productivity gain across the business in a way that is low hassle. People don't like change unless they see a value in it. So it needs to be

easy for people to pick up. So that's where instead of trying to do my own AI research and spend two, three years trying to acquire a big enough data set from people who don't want to give me their commercial data, I thought, let's go with the commoditized capabilities. Let's make it into the business processes with no training required, use no code solutions, integrate what's available that pays per use. So you're not even spending any upfront cash on kind of buying systems and going through that training exercise. And that's just be practical, make it easy and make it low cost and low risk.

That really resonates with me, especially in the manufacturing space. I know I'm working with a client right now that's a small manufacturer here in the United States. And I mean, we're really talking about basic fundamentals of like getting Wi-Fi throughout the factory floor and things like that. And definitely that idea of kind of operational technology, right, whether it's SCADA or, you know, the things that they call automation are a little bit different than what we would call AI in kind of a full IT perspective. And so I see that kind of gap as well, but also kind of a little bit of a headstart maybe because they're used to working with machinery and things

like that. But I mean, even just going back to our kind of first conversation there of, you know, autocomplete and Outlook. And I think that most folks who work on a factory floor aren't checking email the way that a lot of, you know, other knowledge workers sit there in front of an email inbox all day. And so there's definitely some gaps there. So I don't want to mischaracterize all manufacturing. I'm sure there are many that are quite, you know, forward looking and technology savvy, but my experience has been similar to yours, which is that there is a bit of a lag there, understandably, right? I mean, they're working with people's safety and things like that,

and just want to get things moving along. But there definitely seems to be a lot of room for improvement there. But I wonder, are you seeing those first baby steps? I mean, is it straight into no-code AI, or are you having conversations that are even, you know, just getting up to snuff with using kind of some more basic IT technology before they even talk about AI? Yeah, we speak to a spectrum of organizations. So I was speaking to a steel, global steel manufacturer, and they're doing data science projects throughout the year they they're they're doing predictive maintenance they're trialing now augmented reality to help people on the factory floor get help and advice looking at how they can revolutionize field services and it was great because the top 10 kind of pain points of manufacturers they were solidly working towards so there's some

real great innovative leaders out there in manufacturing but we're also talking to manufacturers who are you know tens of millions of dollars revenue every year international international global government contracts and their ERP system isn't doing what they need and and is on-prem and their active directory is so as what as we're helping them see the merits of some of these AI capabilities we're also doing that cloud adoption piece. And I'm speaking later this week with Quest on the concept of developer velocity. So how do we, organizations whose tech teams work faster and smarter, breed more innovation, more customer focus, it's all highly correlated, but organizations that can move safely and quickly outperform those who don't by like five times more annual growth. And one of the key parts of that, of course, as we kind of usually work inside a DevOps thing, the whole people, processes, then tools piece.

Organizational things like talent management and product management capability have a really big significant impact on how fast you can go. Tools are one of those things and cloud adoption is really high up there as well. If you want to go fast, you need to be doing platform as a service, not just infrastructure as a service, because you need more time to build new features, try new things. And you can't do that if you're worrying about your backups and stuff. Yeah, it does seem like these are the directions that a lot of AI applications are heading. I mean, Chris and I have spent some time here on the podcast previously outside the Microsoft ecosystem talking to various players who are trying to develop, as you said, no-code AI implementations and AI platforms and all sorts of ways of accelerating the adoption of this technology.

And I think that this goes to sort of one of the true facts of utilizing AI, right? One of the things we've really learned is that when people think of AI, they tend to think huge grand applications that are world changing, autonomous taxi fleets, that kind of thing, Mr. Cabby. No, no. For the most part, it's really simple and impactful applications, like you mentioned, like image recognition and speech translation or transliteration, you know, things that, you know, we might not even notice as AI, but yet are having a massive impact across the space. I wonder if you can talk to that a little bit. What are the sort of amazing applications of AI, you know, and you did, you already mentioned image and search. What are the

ways in which AI is really transforming applications in ways that people might not even notice. I think the whole thing around speech is phenomenal. So I have a bit of an audio processing issue. I'm very bad with background noise. I take video calls instead of phone calls because being able to lip read helps. So being able to be able to get live captions is fantastic for me. It is a huge enabler and I only have a mild issue. You know, there's tons of people who are deaf or becoming deaf and have impairments that it really benefits from. And it's also a huge enabler because back when planes were a thing and they might be, you know, soon they'll be a thing again, I'd be going around the world and presenting to people whose first language wasn't English. And with even just the PowerPoint live subtitling capabilities, I could subtitle in a different language to what I was presenting in.

So all of a sudden, I was able to give support for people who are non-native English speakers, even though I myself don't speak the language. And the fact that with speech to text and text to speech capabilities and that live speech translation from one language to another, we can now be a more inclusive and global citizenry is just really huge for me. I think that is one of the best things that we can have. We have a Babel fish and Microsoft give it to us for free. Like it's amazing. Yeah, exactly. And you've got companies like Microsoft and Google

and Apple and all these consumer companies delivering the future in a way. You know, I mean, I use this, I'm using translate, you know, both spoken and text and image translation features fairly frequently. And it's funny because, you know, I don't think of it as, you know, oh, this is an AI application. I think of it as, oh, this is something that my phone or my platform can do. But yet, you know, we're living in the future, man. We're living in the

future here. We are indeed. I think, though, it reminds me of that William Gibson quote, which is that the future is here. It's just not evenly distributed yet. And I wonder if there's any impacts there. I mean, obviously, Steph, I mean, this is something you're doing, right, which is trying to take this out and along with Microsoft, right, democratize AI and kind of make people aware of what's available. Are there any other aspects of that that are important to talk about as far as, you know, ensuring that folks actually have access to this and are using it if they do? So there's a couple of parts to that.

One is responsible AI development and making sure that we are inclusive in how we build it. And that even comes down to things like our chatbots. So chatbots are phenomenally useful. I'm a millennial. As well as having a hearing problem, I just generally hate phoning people. But I also don't meet accessibility standards. So they couldn't be used with a screen reader or it'd be difficult to find where the speech to text button is and the inputs.

And there's also a big age divide in how people use computers and what they consider to be acceptable communications. So when we're building AI solutions, we really have to think about who's going to use it and how can we maximize that access? How do we add that fallback, like with the chatbot, that actually you can give it these days, you can have it act as an automated voice assistant on a telephone number. You know, we could make it accessible to all groups

using the same code base, but be responsible and be inclusive. So I think that's one thing that we need to be much more conscious of is how do we build things to make it accessible? And then the other big part is really around training so digital literacy is really critical I saw a stat the other day that jobs that require advanced digital skills pay 40% more than those that don't.

The tech skills gap that implies the demand versus supply is huge. We do not have enough digital literacy across everyone. I even find it really interesting with like my niece and young people that we interview who are considered digital natives that actually they lack things like an understanding about emails and passwords and things because this and HTML and stuff because they've been they've grown up in this world of the app and the facebook login and that kind of whole connected

off experience so their approach with ai is very much that you, it's what my phone does. It's what my app does. But they don't have as much of the savviness around the privacy and the security concerns that a lot of older people will be thinking about. So training and improving how we approach things and how we understand things is really critical. And it's a big part of what I try and do. So I try and make knowledge as accessible and cheap to acquire as possible. So I think I worked out, I've delivered on the attendee side more than 50,000 free hours of training over the past like five years so that's 50,000 hours of professional development and one of the things that I've been trying to do more recently is not just do that for the tech community, which I still have a strong passion for, but start doing that for the SME, small medium enterprise and business community to

really try and grow that awareness that you need to be trained up at an executive level for this to be successful across your organization, but that you need a broader data and digital literacy campaign across the business. Otherwise, people can't identify potential areas of improvement and deliver that for themselves so that they have autonomy and agency inside the productivity gains instead of it just being pushed on them. One of the things that you mentioned there I think is really important, and that is the whole aspect of democratization of technology and the fact that different people in different situations, whether it's different age groups or different cultures, approach technology in

different ways. And one of the challenges I think that happens when people are developing technology is that they approach it from their, you know, the native mindset that they're in. So as you mentioned, you know, a company that is predominantly a web-based, browser-based kind of company is going to develop web-based, browser-based applications that might be inaccessible or just challenging to people who are not used to using smartphones or tablets. And certainly there's also the challenge of bias. But as we've mentioned before, a lot of the challenges, the problems of bias in machine learning models aren't down to like radical prejudice. It's simply

down to the environment that the people live in and the data sets that they use to train their systems. If you train a system only with Silicon Valley people, it's going to work great for Silicon Valley people, but it might really not work in China because it's a completely different computing and user environment there. How do you think the industry can overcome this apart from maybe opening development centers and getting product managers and project managers from different parts of the world and different socioeconomic status, different age, different cultural backgrounds. I mean, how can we actually approach this problem? Laura Cragunen- Broader diversity is definitely an important component, but I think

we can all start by just reflecting a bit more on the world and being curious about other people's experiences and learning some empathy. And I've been going through this myself because I've been a hiring manager for a long time and I would throw out CVs if they had a typo in. Like how can somebody with a word processor still get a typo these day and ages? But people with dyslexia and dyspraxia and other experiences may still get typos. So I was ditching their CV out of expediency when they're a software engineer, their code wouldn't compile or would throw an error if they got a typo and they would fix it.

Why was I removing them from the pipeline unnecessarily? And I was doing it because my experience was key to me. I needed it to be faster. I needed to get to interviews. I needed a higher because recruitment's expensive. But I wasn't being, I didn't have that knowledge of why other people might make typos beyond just poor attention to detail. And I was doing people a disservice. So learning about accessibility, learning about temporary and permanent impairments that impact human-computer interactions, and understanding how different cultures work

is really something we can all be doing. It doesn't cost you anything to Google a topic. We can be more empathetic, more inclusive as we go without needing to solve that great big diversity challenge by ourselves. Just kind of leapfrogging back one step to the training. I mean, one step, I mean, 50,000 hours is really, really impressive. And two, what it made me think about was a book I read, which I can't remember the name of right now, that just really talked about the idea that we've moved into a new age where being able to program is kind of a new literacy. And I think that applies to AI in a lot of ways where we need to move beyond kind of the opaque AI system, just spitting out results and we act on them and really understand kind of what's going

on there on a pretty broad basis, right? I mean, not everybody needs to be a data scientist, but it does feel like this is something that we really, as a global community, really need to understand a bit better to be able to shape the future that we want from it. Yeah, I think that's a great point, Chris, and I really, really do agree. And I appreciate bringing these things up here on the podcast. We've actually gone a little longer than usual. So before we break, let me just dive in here with our traditional end of the podcast questions. As a reminder to the audience, we love to ask our guests three questions that they are totally unprepared for in order to catch them with

some interesting insight. Let's go ahead. Are you ready for this, Steph? Yep. All right, so here we go. And again, she hasn't been prepared, but I think she'll do a good job with these. So number one, brand new question, first time. Can you think of an application for machine learning that has not yet been rolled out, but will have a major impact in the future? Excellent question. Uh-oh, maybe this one's too challenging. I'm currently running through what we currently have. And so where AI is, we're slowly but surely learning cognitive tasks. One of the biggest challenges is that stronger induction, that being able to make conclusions

from relatively few pieces of information and if we can start bridging that old school concept of expert system of rules and how things work and being able to map out processes and combine that with our ability to start learning from data very rapidly, I think we're going to have a way that businesses can scaffold processes and create new ways of interacting and providing people with information in a way that will really help improve everybody's experience and support things like compliance and stuff. So that's kind of almost bringing back the expert system, but doing it right in a way that doesn't cause an AI winter, I think will be a really useful area for us in future. I'm glad you mentioned that. I actually am a big fan of expert systems

and I'm disappointed by the fact that we're all focused on deep learning now. And it just seems like nobody's developing better and better expert system approaches and kind of bringing in machine learning in there. So that's a great point. All right, next thing, and maybe this one will be a slam dunk for you as well. Do you think it's possible to create a truly unbiased artificial intelligence? No, every, everyone and everything has a legacy, you know, no atoms and people are spontaneously created as much as we, well, we might get there with fusion and things, but we're still in a system where the past does kind of loosely predict the future. So the bias in our data, the bias in our approaches,

there's steps that we can do to reduce the negative consequences of that bias, but we're always going to have bias in some form. And it's not always bad. So when we pick loss functions in machine learning to try and optimize a system, we have to make assumptions. We have to make, ideally, a business-focused decision that benefits and determines why we pick it. That's bias. Even just that selection that optimizes for a business outcome is bias. And it's good bias.

All right. And finally, a little more fun. Will we ever see a Hollywood- style artificial mind like Mr. Data? AI that passes the Turing test and can generally learn would be fantastic. My biggest, well we have the problem with abstracting of building that but i'm also quite concerned about the energy implications of how long it would take to how much energy it would take to make that trained capability and have it ongoing learning so perhaps once we have a lot more in quantum computing and hopefully a bit more improved energy situations, we all get to a

generalized AI. But for now, I don't think it's a good goal for the planet. Interesting. I hadn't really thought of the energy implications. Thank you. So thank you so much for joining us. I really did enjoy this conversation. I think we could have gone on for another couple hours. Maybe we will at AI Field Day. Where can we follow your thoughts on AI and computing generally? Where can we connect with you? Sure. So I'm on Twitter at the Steph Locke, and I'm regularly blogging on how we can commoditize AI, improve a business's alignment of AI to business strategy, make that a success, and specifically for manufacturers at nightingalehq.ai. And I'm on LinkedIn as well. Search Stephanie Locke. Excellent. Thank you so much. We'll include those links in the show notes. And Chris, how about you?

What are you working on lately? Yeah, I'm doing a lot of research analysis of the networking and security space for GigaOM. But everything I'm doing, you can find at chrisgrunemann.com. And you can follow me on Twitter, at Chris Grunemann, or let's have a conversation on LinkedIn as well.

Excellent. And of course, I'm Stephen Foskett. You can find me at S Foskett on most social media networks. You can find me right here every Tuesday for Utilizing AI, most Tuesdays on the Gestalt IT on-premise IT Roundtable podcast, which I promise we're using correctly. And also, you can find my writing at gestaltit.com. This week, you will see Chris and Steph and I at AI Field Day. Actually, that's last week when by the time you hear this episode, just go to techfieldday.com and you'll see the recordings of all of those presentations. You can see the questions and comments from Chris, from Steph, and from the rest of the

AI Field Day panel. You can also find those videos online at youtube.com. And if you'd like to be a part of a future AI Field Day event or a Utilizing AI podcast, please just reach out at sfoskett on Twitter. I'd love to hear from you. So thank you very much for listening to Utilizing AI. If you enjoyed this discussion, please

do remember to subscribe, rate, and review, since that does help. No kidding. And also, please do share this podcast with your friends. It's brought to you by gestaltit.com, your home for IT coverage from across the enterprise. For show notes and more episodes, go to utilizing-ai.com

or find us on Twitter at utilizing underscore AI. Thanks for joining us and we'll see you next Tuesday.