 Technology is changing at an amazing pace, and it will impact our work life. To talk more about it, we have Randy Ciefert amongst us. Randy is the head of UX at ServiceNow, where he is using AI to automate everyday productivity tasks. He's here to tell us about how the future of work will look like. So please welcome Randy with a huge round of applause. Welcome, Randy. Thank you very much. I know you're at the end of your three days here, so I appreciate that we still have quite a bit of attendance, because I did notice the bags of the back of the room, so I see people are ready to start going home. So what I want to talk to you about today is humans versus robots, right? So we do have machine learning, AI, that's coming to the forefront of everything that we do. So myself as head of UX and product insights, this is something that's very close to my heart and things that we deal with every day. So first to start with, to set some context, ServiceNow. You might have seen our booth at the back. We're a growing company here in Hyderabad. We're a company that's based and founded out of San Diego 13 years ago. We have our headquarters in Santa Clara, and we're one of the fastest growing enterprise software companies. We were actually 2018 Forbes' most innovative company as well, so in the entire world, ServiceNow was voted as Forbes' most innovative company. Now, we don't typically talk about that that much, because at the end of the day, we don't really see ourselves as winning it because our customers want it for us, right? We're very customer focused. We want to see what our customers do. It's their needs, it's their problems, it's theirs opportunities that drive what we do. So fundamentally, we kind of have our brand slogan here, which is we make the world of work, work better for people. So we're an enterprise software company that really focuses on workflows, productivity, efficiency, ease of use, and making applications that fit into the environment that makes people's work easier at the end of the day. So because of this, I spend a lot of time obsessing about work, how work is done, the future of work, how work is evolving, everything else, so obsess about it. My research team obsesses about it. We stay up at night thinking about it, slacking each other. Did you see this cool new article? That's what we do. We're really passionate about doing this. So curious question, who here is currently working on or has worked on very recently mobile projects? Like everybody, pretty much, right? Who here is currently working on machine learning or AI projects? Maybe about 10% of the room, so a little bit less. So just like mobile, if I had asked you that question five years ago, it maybe would have been about half the room. If I ask you this question again in two years or three years about AI and machine learning, I guarantee everyone in this room is going to raise their hand. So that's what I wanted to talk about today is a little bit about this technology but also really the impact on us as user experience practitioners and also how it's impacting the future of work. So this is happening, right? This isn't a flash in the plan. This isn't a niche technology. This isn't like your 3D TV that you never use, right? This is happening, right? This isn't a passing technology. This is happening. So already we're looking at investments. 2022, the investments in AI and machine learning by companies is going to exceed $77 billion of investment, huge market. If you look at the use cases that it will be driving in the near future towards 2022, really you're looking at things like automated customer service is going to be the bulk of it, service desk, customer call centers, these kind of things. Automated threat intelligence and prevention systems, sales process automation recommendations, fraud analysis, et cetera. So these are the first use cases that are going to hit us with AI and machine learning. So it's happening, huge, huge, huge investments by many companies, huge investments both on the technology companies that are implementing it into their solutions, huge investments for the customer companies that are actually also implementing it into their companies. By 2020, Gartner predicts that you will have more conversations with a robot or a machine than you will with your own spouse. Who's already here? Okay, don't tell my wife that I put it in my hand. But this is happening, right? This is your Alexis, this is your Siri, this is everything else that you interact with. This is the chat bots, these are the things you're doing. This is the amount of interaction that you're having and so it will exceed the number of times you talk to your spouse in 2020. Just to kind of a little, I'm gonna throw out terms, AI, machine learning, deep learning, so I just wanted to take a second to kind of normalize on some definitions. So AI really refers to any kind of technique, cognitive technique where the machine is doing something that the human does cognitively. That also includes even just simple if them logic. So AI has been around a long time. If you've called up a customer support, if you've called up your bank and you go through an automated system and you go through a decision tree of if then statements, you're dealing with what is technically called AI. So it's been around for decades. Coming up more recently are more investments around machine learning and this is set up to be kind of a jawbreaker layered candy here because it's all kind of nested pieces. So AI is the big shell and then you go to machine learning. Machine learning is actually the algorithms that are produced. So in traditional programming models, you write a program, you tell the steps of the program and then there's some outcome. So you tell all the steps and then there's an outcome. With machine learning, the algorithm learns the step, you define the outcome. So you define the outcome, it learns the steps. That's the basis of machine learning. Deep learning takes layers of these algorithms and basically creates a neural net and then you throw a mountain ton of data at that neural net and that's when it begins to learn and see patterns that us humans would not ever be able to see. We cannot process these kinds of data sets. That's deep learning when they can look at these. So it's just kind of some definitions that you can use. Every day, open a headline, open a blog, there's something new. Like I was even adding a couple last night because there's some recent ones. So look at anything that's happening. This camera app uses AI to erase people from your photographs. So you just had an nasty breakup. You wanna take them away from all your social media streams. Bill, Jane, whatever your ex was and you can just basically hit a button. It will scrub everything. It will make your pictures look great. It will take some intelligence about the background and all of a sudden, Bill is gone. Your dream has come true. You don't have to deal with Bill or Jane anymore. So this is what it can be used for. Making deep fakes. Here's another one. You've probably seen these, right? Tom Cruise was not Iron Man. But you can actually watch videos of Tom Cruise being Iron Man because AI technology has allowed people to render highly realistic, basically hacks of videos. Just as I was putting this together, even last week, there's now a Will Smith as Keanu Reeves in The Matrix. It's kinda wild because they put Keanu Reeves' hair on Will Smith and the mustache, it was kinda wild. So all these things are popping up all the time. Why are we hearing about it so much right now? Well, we're at a perfect convergence of three things of why it's exploding and making this possible right now. It's big data, storage, right? We're producing more data than we ever have. Storage that's cheap, easily networked to put that big data on. And then we have the computing power to do that on. So put all these three together, right? The computing power to process the data that's on the storage of all this data that's being produced. So we're at the perfect opportunity where these have come together that actually now we have enough pieces in place to actually power these machine learning algorithms. So for example, so it's in 2014, right? Where the number of these devices exceeded humans on the planet. So we have over, and this is 14, over 7.2 billion devices. Now we're much higher than that because this is 2014. So we have so many of these devices that are collecting data, creating data, texts, videos, pictures, everything else that's being collected right now. And the computing power in these devices is such that there's more computing power in this than there was all the computing power put together to put people on the moon, right? That's incredible. Going a little bit closer, 2017, some 2017 data, is that electronic users generate 2.5 quintillion, I don't think I could write that number of zeros, quintillion bytes of data per day. What was mind blowing to me is that 90% of information and data that has been created by humans in the history of humankind has been in the last two years. Think about that. Lifespan of every literature, photographs, everything that has existed, cave paintings, we have exceeded 90% of all human existence and data collection has been in the last two years. That's crazy. This is the amount of data that we're creating and it's not slowing down, it's getting bigger. YouTube, every minute users watch more than 4 million videos, most of them cat videos, I'm guilty. 15 million texts, which this makes me wonder that if these machine learning algorithms are watching us and we're watching a lot of cat videos, you know, when they take over us, they'll make you wonder what they think about our cat watching behaviors. Okay, who really got anxious when they saw that? Right, you don't want to open your phone and see how many you're up to. So curious, everybody raise their hands. Everyone please raise your hands. Okay, put your hands down if you currently in your inbox have less than 100 emails, total emails. I don't think I saw a hand go down, okay? Maybe one or two, okay. What if you have less than 200? Couple 500? Oh my gosh, I'm gonna have to go higher. A thousand? Okay, some people, maybe that's their limit of 1,000. 5,000? 10,000. Okay, there's a few stragglers left. Okay, machine learning AI will help you. Anyone have more than this? Okay, no one has more than 32,000 right now. Okay, good. Machine learning AI will help you with this, right? If you think of the rules and the simple things you do to disposition email, right, how many emails do you get now? There's tons of emails, you get your Slack, but then you get an email that you got a Slack, right? You get status notification updates on workflows. So a lot of things happen. To me, this is the promise as I switched to kind of the enterprise discussion, the work discussion. To me, this is the promise I'm looking for and we're really close, right? Microsoft is already starting to do this where they kind of have the dual inbox. It's kind of dumb right now, but it's moving in that direction. But to me, this is one of the things that will help me out most. So I'm not necessarily looking for driving cars. I'm not necessarily looking for the robot that can clean my house. I'm just looking for someone to help with my inbox. This is just a couple of weeks ago. Elon Musk was at the World Artificial Intelligence Conference and he basically said that AI is gonna take over jobs. Well, maybe if you're an AI developer, then you'll have a job until the AI is smart enough to write its own code. Kind of went on a little bit and then he actually said, well, you know, humans still wanna interact with humans. There's still human interactions. So anyone that's working in the field of human interactions or human interactions in combination engineering, that's probably a pretty good bet too. So I thought, hey, sounds like us, right? So that's kind of us, right? We're living in the world between engineering and humans. So we have a shot. Okay, so it's good. So we have a little bit of a shot. But to kind of echo what Elon's saying, the World Economic Forum is kind of looking at the trending of data and looking at work that's done. And so that they're predicting only by 2025 is that you're gonna hit that point where bots and technology have taken over a significant amount of workflows. So it's about 30% right now, 29%, where already there's a lot of automation. So this includes your factory automation. This includes kind of automated workflows. So everything's bundled in there. And so by 2025, you're gonna have this 50-50-ish kind of separation of the work between bots, whether it's automation, robots, factory automation, or chat bots or things like that. So another little game here is we think about enterprise. What is the top IT help desk support issue? Is it A, hard drive failures and issues? B, network connectivity. C, forgot password or password resets. Or D, coffee spilled on my laptop. Who thinks it's A? B, network connectivity. C, password. D, coffee and tea. Man, I guess it's just my laptop. Okay, it varies by company, but pretty much you're right. The room is pretty much right. It's C, password resets and things like that. So if you think about this simple task, this is actually one of the opportunities. These are actually some of the top issues. Not necessarily coffee or tea, but like my laptop is damaged, something's wrong, I need a new laptop. So these four actually kind of change positions, but these four are actually the top issues. Let's take D as an example. So this is the kind of thing that you probably are already starting to see, or you will see. So at ServiceNow, we use our own technology. This is our chatbot mobile app. So I can, after I spill tea on my laptop, I can go ahead with a chatbot, get onto my ServiceNow app and order a new laptop. So I can have a complete interaction with a chatbot and order a new laptop. Huge time savings, right? I don't have to like log on. I don't have to authenticate myself with my laptop, which I can't do anyways because it's not working. Then find a portal or a contact number where I make the request, right? I can just open up my app and start chatting and order a new laptop. So this is happening. This is the cases that you're gonna see first in the enterprise. If you're not seeing it, so when you go to the work and the work that you're working in the companies, these are the first AI use cases from a work standpoint that you'll see. It'll be service desks, it'll be customer service if you're talking to, you know, trying to get help. You're probably already dealing with it, probably already seeing it. These will be by far the number one use cases because these are tasks that are easily optimized, right? Staffing IT service desks to do these things, it's staffing is hard. And so if you can take the burden of these transactional tasks off them, they can focus on the harder things to solve for. Here's another example. So it's not just about a whole technology, right? It's not about just having a whole bot. Sometimes it's actually integrating machine learning and AI into things that we already have, apps and technologies. So this is actually where we've integrated image recognition into our app. So I sat on my glasses, I need to figure out if they'll pay for another set of glasses, if my insurance will pay for another set of glasses, I can whip out my app. These are screenshots from the actual live app that I have on my phone, take out my app. I'm gonna scan my glasses instead of hopping on portal and is it glasses, is it healthcare? Took a picture, it knew what it was. It recommended some categories. There it is, FSA, flexible spending account, insurance, click on that, boom. Working, I can write from there, start submitting a ticket or a claim against what I need to do. Imagine this too, you go to a conference room, you need something done, bulbs out. You know, the temperature's off in the room, you need supplies easily done with this kind of technology with things like image recognition. There are some challenges though. So if you think about work and enterprise, there are some challenges of applying machine learning and AI technology into the enterprise. Number one is most companies will have concerns about security or sharing data. So when you make a transaction with Alexa, Alexa, order me a pizza. When you say that or Siri, that's actually processed in the cloud. It's not processed on your phone. That's data going up to the cloud. So a lot of companies are kind of hesitant to implement these technologies because they don't want things people are talking about. They don't want pictures taken in their environment going up to another service that's offering this. Now the alternative is with a lot of these technologies is you can implement it on your stack internally which is possible, but then you deal with the situation that machine learning algorithms are best when there's huge amounts of data. So if you don't combine it with everybody else's data, if you're just using your company's data, the machine learning algorithm isn't as smart. One consequence of a machine learning algorithm not being as smart is biases in the data. I'm gonna come back to that. So these are some of the limitations that we're dealing with when we think about bringing AIML into the workplace. So kind of Elon's fear isn't real or at least the research says it's not real. What's really gonna happen is, yes, there's a shift of skill sets. There's things that bots are good at doing. It will move some jobs. There will be jobs impacted because of it. So they're expecting that 75 million jobs will be lost by 2022, but because of the technology, they actually predict that 133 million new jobs will be created, new kinds of jobs, technology jobs, shifting jobs. So the types of jobs that are declining are the things that bots and machine learning are good at. Data entry, assembly factory workers, client information, customer service workers. So these are the kind of things that bots are good at. These are the type of jobs that will be displaced. A little bit deeper in this report, definitely worth reading. So these are the kind of skills that bots are good at, image recognition, natural language processing, transactional mathematics. These are the type of jobs that will be displaced. What machines are not good at doing are analytical thinking, innovation, complex problem solving, reason problem solving, system analysis, emotional intelligence, creativity. These are things that they're not good at. So these will be the rising skills needed to offset. This is where the 177 million other jobs come in, is these are the skill sets that will need to be enhanced. Is that, those characteristics feel kind of relevant to anybody in this room? We need you, we need you bad. Don't let this happen. I was in Germany two weeks, this is my picture. This is actually in Germany, took my son to Germany a couple of weeks ago, did a heritage trip with my father. And we were riding the trains and had to go into get some customer support and went in and saw this, the creepiest kiosk that I've ever seen. So you have this like weird projection thing on this plastic cone and then it still had a tablet that you interact with. So I instead kind of moved the side, took my number, waited in line with everybody else and watched. No one approached this. So I can just imagine how costly this thing was to implement, but no one used it. So here's a big investment and no one's using the technology because it's not approachable. It's just creepy. So back to the comment I made about biases, so the three constraints in the enterprise, biases. So when you don't have enough data or you're using the wrong data set, machine learning or the output of machine learning is only as smart as the data that you're feeding it. So who's hiring right now? Who's doing a lot of hiring? Any hiring managers in the room? It takes a lot of time to go through candidates. And a lot of those things that you do are actually the cognitive reasoning you're making, you're looking for your experience, you're looking for some skill sets. That task is actually really well-suited for machine learning. So if you can filter out candidates as a starting point, obviously you wanna do due diligence on great candidates, but if you can filter out candidates and kind of really focus on it, that's gonna save you a lot of time. Amazon did this. They actually implemented it. The problem was to feed the machine learning algorithm, what they do, they took the resumes from all the successful hires that they'd had in the past. And there was a bias in there because historically over the life of the company, more male engineers tended to be hired. So that bias in that data, the bot learned that bias and started to filter out female candidates because of the historical trend of the company. So even though the company was actually trying to focus on this as a strategic initiative, to have a more diverse workforce, it actually worked the opposite. So they didn't know what was going on. Luckily, some clever engineers figured it out and they had to turn it off, right? These are the things we have to think about. Human behavior, the data, how things actually work, it's very easy to think, oh, here's a data set, throw it in. But thinking about the consequences, thinking about the human behaviors, we need to help these folks out. Let me talk about a good example. Drove a Tesla recently, great experience. Great, great experience. I don't even consider it driving a car. It's a different animal, right? And so one of the things that was very interesting to me was that as I was driving it, just the natural flow on the UI, and this is the UI that's above the steering wheel, not the big one. On the UI, it was seeing differences between motorcycles and cars and big trucks and it was seeing things coming up from my blind spots, things that I didn't see. I could see it following the lanes, right, the blue there. Like it was seeing that the lane was curving. So when I actually hit that autopilot button, I had a higher confidence that it was smart enough because it was seeing these things to just have confidence to actually say, yes, I'll hit that button. I was test driving, my wife was in the back seat, of course she covered her eyes and said, no, no, no. But I did have a lot of confidence because I could see that. So this is a great example of applying affordances to the interaction to build towards trust in the machine learning or in the bot that you're going to interact with. So this is a great example. I'm not sure if Tesla did this by design. If there's a Tesla designer in the room, kudos to you, or if this is kind of an organic thing, but it was really, really nice piece of design from that interaction. So this is a great, great example. So we need some rules. We need some rules as we think about how we as designers are designing those interactions with our bots, with machines. I like to get my mental model around kind of starting from the top. So you can't go wrong by starting here, right? Don't kill me. Isaac Asimov, right? Some basic rules of robotics. Okay, that's a good set of principles to start with. A great paper came out recently from CHI 2019. Amrishie et al actually kind of walks through 18 guidelines they walk through some great ones. Just from my experience and what we're doing at service now, I'm kind of adding a few more guiding principles too to what we're doing. One is we really have a role in identifying tasks that are appropriate in the first place for automation. So there's some things that aren't appropriate, right? That kiosk actually wasn't appropriate because most of the people in there were actually tourists, right? This wasn't something that happens regularly, that German train one. So knowing the tasks. This is an interesting one that you don't hear very often, but initial teaching of the bot. So if you're going out and you're learning the use cases, we can actually provide insight of what I call the case zero. So instead of just having sort of a dumb bot that slowly learns over time, you can accelerate its learning by feeding it a case zero workflow, right? Us as behavioral scientists, as we go out and learn about workflows, we can feed that. So for example, we recently trained a bot to read legal contracts and extract out the service level agreement so that we could put that in the database. We actually helped them because we defined the workflow the first time and that basically accelerated it. Another one is use graceful, kind of like the Tesla model, use graceful evolution towards more mature AIs. Don't just try and jump there overnight. Think about the affordances that get you there and also graceful degradations back. When the AA is not smart enough and it can't help you, how do you land softly back as well so that next time you don't reject the bot or the interaction you have, but you're actually have a little bit of trust because it was more of a natural handoff. And understand the entire system, right? Context of use, context of use is king. Understand the context of use of which that's occurring in. Back to the German trains, mostly tourists, right? It just didn't feel approachable when it was in German. Yes, I could change the language, but just wasn't approachable. Sophia, the robot, she still needs us. A little bit strange, you probably heard this, but Sophia, what? Or does she? We're getting pretty close right now. Sophia actually was granted citizenship from Saudi Arabia recently, so it's gonna actually introduce a whole lot of interesting legal and ethical concerns, but I can tell you in the near future, and hopefully what I've told you today, that we're needed, right? Our jobs aren't gonna go away by robots. In fact, we're gonna be needed more to actually make these technologies successful. So I'll wrap up with kind of my plug here. If this is interesting to you, right? I hope this talk was interesting. We're hiring. These are the interesting things that our team gets to geek out on. We're really fascinated. We wanna make work work well. We wanna make sure that these emerging technologies actually fit and help you, and if you're actually interested as well, then come see us at the back booth. Thank you. Thank you, Randy. The future of work looks very exciting, and we are all ready for it. So please accept a small token of appreciation on behalf of UX India and all the participants. I welcome Ranjeet, our core team member. Hey, good to see you again. Oh, thank you. Oh, thank you. Awesome. I will, absolutely. Thank you.