One of the biggest issues is getting business people to understand enough about AI [so] that they can break the problem down, and then getting the AI experts to understand the business priorities so that the two can meet.
—Jeff Loucks, executive director of the US Center for Technology, Media, and Telecommunications
Tanya Ott: The people have to understand each other and the technology has to be powerful enough to bring their ideas to life. Today, on the Press Room, we’ll close the loop between artificial intelligence and 5G mobile service.
I’m Tanya Ott, and if you’re my age or younger, you probably don’t remember the first commercial wireless phone call. It was made from Soldier Field in Chicago in 1983 on a car phone that my kids would probably describe as looking like a “home phone”—it had a large, push-button display and a cord. A cord! Back then, the idea of a nationwide wireless network and phones thinner than a deck of cards that could take and send photos anywhere in the world in a fraction of a second was … well, I’m assuming it wasn’t even a thought!
But here we are in 2019, and we’re getting ready to see the first rollouts of the fifth generation of wireless technology that’s going to be even faster and more powerful than the 4G broadband that lets you video chat with your family around the world from your smartphone. It’s called 5G. It’s going to give consumers access to more information, faster than ever before, and it’ll make businesses more efficient. I’ve got three guys on the show today, who all study the convergence of technology, media, and telecommunications for Deloitte.
Paul Lee : Hi! I'm Paul Lee. I’m a partner based with Deloitte in the UK.
Jeff Loucks: My name is Jeff Loucks and I’m the executive director of the US Center for Technology, Media, and Telecommunications.
Duncan Stewart: I’m Duncan Stewart, the director of research for Deloitte Canada in the area of technology, media, and telecoms. But I also have a role in coauthoring and talking about the global predictions.
Tanya Ott: And Duncan, Jeff, and Paul are predicting big things for 5G and the technologies it enables.
Paul, let’s get started with a really basic question, which is, how long have we been talking about 5G happening?
Paul Lee: Most people haven’t been talking about 5G yet. It’s been discussed in the telecoms industry for several years, as that’s the time it takes to agree on the standards. What we’ve seen this year is the first round of commercial launches where handsets will be available. And this is the kickoff of a process that will take years, probably a decade, to conclude—even in developed countries. So, 5G really is something of [importance] now and will become more and more important over the years to come.
Tanya Ott: What’s the difference between 4G and 5G?
Paul Lee: At the current moment, there are two things which really merit explaining. One of them is [that] 5G makes mobile broadband—what we’re used to using [for] smartphones—faster, and also the ability to have a lot more devices connected. And there’s another facet of 5G, which is fixed wireless access, or using the mobile network to provide connectivity to the premises instead of using copper or coaxial cable or fiber to the home.
Tanya Ott: So, why is 2019 the year that things are really going to pop?
Paul Lee: It’s the first year in which we’re seeing quite a few launches. We’re expecting 25 launches in this year and expect the number will go up over the course of the year. So we’ll probably have at least 30. It’s the first year in which we have smartphones being available that support 5G, so probably about 20 vendors will launch this year.
Tanya Ott: You mention the 5G smartphones. Give us a sense of what it’s going to be like. Are they going to cost more? Are there going to be enough 5G networks around the world to really make sense of a lot of people adopting a 5G smartphone?
Duncan Stewart: One of the things is that 5G handsets will, at least at first, cost more to manufacture. They need new chips, and those chips cost more in the early days. Whether that price increase is passed directly on to consumers will vary around the world. But broadly speaking, one should expect a 5G handset to be at least US$100 to US$200 more than the equivalent of the 4G version. And, one of the problems from a consumer perspective is that it will tend to give you much, much higher speeds to a handset, but most of us don’t need much, much higher speeds to our handsets. Plus, in the early days that coverage will be—just like [the] rollout of 4G—spotty. It will tend to be only in the bigger cities and only in [parts] of the bigger cities. Smaller rural areas or suburbs won’t have a lot of coverage. So, our expectation for 2019 5G smartphone sales is quite low—only about a million units around the world. We think that there will be some early adopters who are interested in more expensive phones that don’t have broad coverage, but we think that the mass market really doesn’t kick in until 2020, and, if we’re honest, really more like 2021.
Tanya Ott: I’m wondering from a consumer standpoint, if they’re really not going to notice and need that faster speed, what’s the appeal?
Duncan Stewart: You get a 5G phone because everybody’s got a 5G phone.
Tanya Ott: (laughs).
Duncan Stewart: No ... this is what happens. People go out and buy the phones and you know—"Hey! My friend’s got a 5G phone and I’m on 4G. That doesn’t sound as good!” So we’re definitely going to see some of that keeping up with the Joneses.
When I talk to carriers around the world, many of them are less convinced about the need to rush to 5G, but they do describe it in their local market as table stakes. If somebody in a given country pushes to 5G, everybody else more or less has to follow simply to be able to say, “Yes, we have it too.”
Tanya Ott: And there are parts of the world that are much more reliant on wireless-only connections, so that may drive that as well …
Duncan Stewart: There’s frequency—very, very high … a few years ago, it was regarded as very nearly useless. But new improvements in technology mean that at the millimeter wave, which is around 26 or 28 gigahertz, wireless signals which don’t go very far—perhaps only 500 or 800 meters or so—nonetheless are enough to have tremendous room for bandwidth. This is not mobile. This isn’t you wandering around on your phone. This is fixed wireless. So you would have a small antenna on the outside of your house or perhaps in a window pointed directly at an antenna. This would allow for very, very high-speed, high-volume, and low-price fixed wireless access. Estimates are somewhere on the order of around US$50 a month for gigabit speed. That’s the sort of thing that a number of consumers will adopt—not right away, not 2019 … this is a longer story over 2020, 2021, 2022.
Tanya Ott: What’s the big appeal there for consumers and what’s the big appeal for companies?
Duncan Stewart: You can connect up homes using this—in some territories, for perhaps US$200 dollars per home compared to US$2,000 dollars a home for wired solutions. So, it is much cheaper for carriers to roll out high speed in areas where they don’t already have things like fiber, and it’s much cheaper for consumers to get at-home data packages at very, very high speeds and, in many cases, with no data cap at all. You’re looking at gigabit speeds and unlimited data.
Tanya Ott: Where does this go long term?
Paul Lee: One of the potential differences with 5G is that you can start tailoring different packages for different types of customers, particularly at the enterprise level. There is a facet of 5G called network slicing, and what that enables the operators to do is define for a particular vertical sector or even for a particular company exactly what kind of speeds they will get either down or up, or what kind of latency they can have. If you imagine the example of a broadcaster, when they’re doing some of the breaking news, rather than sending a satellite truck, what they would do is use 5G and have a guaranteed speed available to them for uplinking their content. So that’s one of the applications which is possible. Other things which are being talked about in terms of 5G are in a manufacturing plant, [taking] the current usage of fixed broadband or Wi-Fi or also 4G and replacing all of that with 5G because it’s got much faster response times, a lower latency, better security, and better ability for connections to be moving at a faster pace, so a better rate of hand-over.
Tanya Ott: The predictions that you guys have come out with for 2019 … there’s a bunch of them across a lot of different technology sectors, but it all starts with 5G and AI. So, I want to bring Jeff Loucks into the conversation. Jeff, you have been looking at AI quite extensively for a period of time now. I would love to first just get you to give us some examples of AI technology that are already in use that even people who are not deeply steeped in this world would recognize and go, “Oh, OK, that’s AI.”
Jeff Loucks: Artificial intelligence—the basis of it right now is really machine learning and its more complex offspring, which is called deep learning neural networks. Companies who know how to use these can analyze huge data sets. They can find patterns in data and make accurate predictions. To give some examples of where it’s used today, machine learning powers virtual assistants, facial recognition; it’s the technology behind autonomous vehicles. It can be applied very broadly to a number of operations and that’s what’s so exciting about it.
Tanya Ott: I’m sitting at home this week actually doing my taxes with an online tax-prep company. I have a little virtual assistant kind of thing that pops up in the bottom corner where I can ask questions and it will answer or point me in the right direction. Is that an AI?
Jeff Loucks: It could be an AI. It could be a virtual assistant or a chatbot that’s powered by natural language processing. You say, “Hey, Alexa … play the Smiths.” Alexa is understanding what you’re saying because of natural language processing. So that’s an artificial intelligence technology that’s getting very rapid adoption and actually is covered in another of our predictions on smart speakers.
Tanya Ott: We’re going to be talking about smart speakers in another episode of this podcast. What does AI unlock for companies?
Jeff Loucks: I think we’ve had a slow start with AI because it has taken a lot of expertise to get up to speed with these technologies. But we’re seeing rapid adoption. The analogy I like to use is—as a skier, I’ve sometimes taken a look at a ski run that looked really exciting from the chairlift, but then once I got to it, I was intimidated and didn’t really know how to get down. But, if I waited long enough, usually an expert skier would come by and kind of barrel down the run and show me the best way to get down. And, in that sense, with artificial intelligence it’s a bit easier to go second for companies. We mentioned that machine learning and deep learning are the foundation of AI, but they require a lot of technical expertise. There have been a small number of AI pioneers, such as the global tech giants and other tech-savvy companies, that have really shown the way in how to do this. They’re clearing the way for everyone with a couple of innovations. One of them is enterprise software that has artificial intelligence built in. An example of this could be ERP [enterprise resource planning] software with artificial intelligence. Think about examples like SAP S/4HANA. You mentioned chatbots—it has chatbots to help improve the user experience. It also does things like making cost forecasting more accurate and automating procurement tasks. So that’s one path.
The second path that you’re going to see is cloud services that are making it easier for companies to develop their own customized artificial intelligence. One of the problems with artificial intelligence that’s baked into an enterprise software system is that everybody can use it. It doesn’t require any expertise to use and all the use cases [are] already kind of baked in. If you want [a] competitive advantage, you need to build your own systems and that’s where these cloud development software platforms come in.
Duncan Stewart: When I talk to clients about artificial intelligence and machine learning, over and over my clients have said, “We know it’s powerful. We know it’s probably the most important thing going on in technology today.” But many of them are scared of it, frightened of it, feel they lack the tools, the staff, the competency to do it. So, this ability for clients to access the power of machine learning in AI through the cloud is really … I love Jeff’s example with the double black diamond run, but most of us are smart enough not to do that. Not everybody needs to go down a double black diamond run, but everybody does probably need to get up the curve on AI and machine learning. This transformation that Jeff’s talking about isn’t just for some sort of rarefied stratum of companies. It’s going to be every company [that] wants to access the power of AI, and the cloud is really enabling that.
Tanya Ott: For companies who, as Duncan says, have not dipped their toe in AI yet, but they want to start, where do they start? What do they need to be thinking about?
Jeff Loucks: Well, I think the companies that are just getting started with AI have, in some sense, a bit of an advantage because you don’t have to figure out how to get down that black diamond run yourself. If we think about our skiing analogy, when we think about the enterprise software that has AI built in, it’s like someone just came in and flattened every single mogul and now it’s a green run, right? Anybody can use it. With the cloud-based development tools, it’s more like taking a black diamond run and smoothing it out a little bit so it’s a blue run. You need some skill, but it’s way easier than bouncing around in the bumps. And these tools can help you accelerate your development of AI tools that you’re making yourself with things like pre-built algorithms, APIs for specific functions like machine learning, computer vision, and so forth. And then automated machine learning [can] help you select and tune [the] right algorithms. You do need some expertise, but you don’t need as much as you used to even a couple of years ago.
Duncan Stewart: The other thing I’d throw in here on the cloud development is [that] the price of it has come down. Apples for apples, you can do more complex things now, but if you’re doing the AI in the cloud, the price of it from two years ago—according to some of my clients—is down much more than you would normally [see]. It isn’t just that the tools are easier to use; they are a lot cheaper to use as well, which, of course, makes justifying ROI a lot easier.
Jeff Loucks: I just want to pile in on that one because there’s also big competition from the big tech companies and the internet giants. They’re trying to attract people and keep people on their cloud platforms, and artificial intelligence is a way to do that. If we think about cloud in two waves, the first one was helping companies to solve problems around their workloads and so forth, operating more efficiently. The second wave is about helping companies to innovate better with things like artificial intelligence. And some companies that haven’t done well in the first wave are really betting on the second wave with AI at the center.
Tanya Ott: You raise the issue of having the talent to make this work so that you can set yourself apart. And talent, you write in your report, has been an issue. Companies have been pitted against each other in bidding wars for talent. What does that look like right now and where does that go?
Jeff Loucks: That’s a great question. One of the advantages of these cloud platforms is that you don’t need as much specific AI expertise. A regular programmer or coder or data scientist can do it. But there’s also the problem of operationalizing all this. What is required is not just technical expertise, but also people who have got business savvy, who can speak business to artificial intelligence experts, or data scientists to break the problem down for them and give them a problem that they can easily [be] solved through artificial intelligence. One of the biggest issues is getting business people to understand enough about AI [so] that they can break the problem down and then getting the AI experts to understand the business priorities so that the two can meet.
Tanya Ott: I know that Deloitte recently surveyed a couple of thousand executives of companies who were starting to use AI. Did you get a sense from that survey of how adept the business folks are [at] speaking artificial intelligence and the AI tech people are at understanding business?
Jeff Loucks: Both from the survey and also from some other research I’ve done with data scientists, that’s one of the main problems, especially on the data side. They sometimes feel like they are thrown into the deep end to go and fix a business problem that is beyond their scope. But what we’ve found is that when we ask businesses in the survey—“What type of talent do you need?"—first off, they will say that they want more data scientists and AI researchers. But right underneath that, they need the type of business experts who can speak data science. They need program managers. They need people who can execute transformation projects. So, there’s a really heavy business element to the talent shortfall that people are experiencing.
Tanya Ott: There’s a connection between 5G and AI, and some of it lies in edge devices and some other things that we’re starting to see. Let’s talk about that a little bit. How do you pull these two threads together?
Jeff Loucks: I’ll take a quick shot at this and then I’d love to hear what Duncan has to say. So, with edge devices what we’re going to see is more artificial intelligence analysis and inference is going to be happening in devices at the edge of the network—such as smartphones, production equipment, and sensors. Instead of having all this data travel to data centers or cloud services, a lot more is going to be happening at the edge of the network. And this is really driven by three things. Duncan mentioned earlier there’re going to be more AI processors in phones and devices like that. Also, there’s just more storage in edge devices. Finally, there are more low-power options—better batteries and sensors and so forth—that’s going to unlock a lot of power at the edge.
Duncan Stewart: I think this is the “dark horse” kind of topic out there. Historically, if you want to build a robot or a turbine or a vehicle and do the machine learning for whatever function it was, the only place you could do that was in a very, very large data center connected through the cloud. Inside those data centers were special chips that cost hundreds or thousands of dollars and consumed hundreds or thousands of watts. Obviously, those sorts of chips could not be placed on a relatively cheap robot in the field, so the business case on all this is: “I need machine learning and I can’t do it on the edge device, therefore I’ll do it through the cloud.” And this is where 5G comes in. 5G is incredibly good and will be incredibly good at handling the massive amounts of data that might come from a robot that’s looking around 360 degrees and can process it and get it back in a millisecond or two, which is what that robot needs. That’s that “very low” latency that Paul referred to with 5G.
Now that’s still going to be a factor going forward, but as Jeff mentioned, as more and more is being able to be done at the edge on chips that don’t cost thousands of dollars and thousands of watts but instead cost pennies and use milliwatts, what we see is perhaps a reduced reliance on 5G for Internet of Things applications. Edge machine learning at some level competes with 5G for that edge inference, that processing at the very edge of a network by the robots and the devices out there. We don’t yet know exactly how that’s going to work, but in talks with clients, they’re evaluating—"Where do we do the processing? If it’s at the edge, maybe we don’t need 5G. If it’s in the core, we absolutely do need 5G.” And there’s going to be a mix. There’s going to be a lot of hybrid solutions out there.
Jeff Loucks: There’s one other thing I wanted to mention about AI at the edge. I think it ties 5G with the cloud phenomenon of AI. That’s that tech giants are already releasing machine learning tools that can automate the process of deploying models on devices at the edge. So, basically, you can develop a machine learning model and then these tools can allow you to deploy it anywhere—whether it’s in the cloud or in an IoT device or a smartphone. As that becomes easier, I think it’s going to make more inference at the edge possible.
What I mean by that is that a lot more of that data analysis is going to be happening in the devices when it’s easier for companies to create one set of machine learning models and then be able to deploy them anywhere, whether it’s in the cloud or in [an] IoT device, without having specific programming for the device.
Duncan Stewart: Just to amplify that, when I was talking about the real barrier to edge machine learning in the past was the chips—that is true. You can’t stick a US$5,000 chip using 1,000 watts on an edge device. But the other part is [that] hardware is only half the story. I think what Jeff essentially is saying is it isn’t just the hardware. It’s the software tools as well. Edge machine hardware has changed dramatically in two years, but the software has [changed] as well, dramatically reducing the barriers for entry for people who want to do more edge machine learning. It’s hardware plus software together working in a synergistic fashion.
Tanya Ott: Synergy … that’s where it’s at in 2019. You can read more of Duncan Stewart, Jeff Loucks, and Paul Lee’s predictions for 5G and artificial intelligence in their TMT Predictions 2019 report at deloitteinsights.com. While you’re there, be sure listen to the other podcast episodes in this series. The [TMT] team has predictions on China’s rollout of 5G and their work in the semiconductor industry, quantum computing and 3D printing, the importance of sport gambling and eSports to television, and—one of my favorite topics—smart speakers and the future of radio.
Duncan Stewart: We love our smartphones. We love our computers. We are tied to these devices. When you’ve got your hands two feet inside a turkey, you don’t want to be touching your smartphone screen.
Tanya Ott: Subscribe to the podcast—it’s free and you can have these conversations delivered directly to your device. You can also connect with us on Twitter at @deloitteinsight and I’m on Twitter @tanyaott1. Catch you again in two weeks.
This podcast is provided by Deloitte and is intended to provide general information only. This podcast is not intended to constitute advice or services of any kind. For additional information about Deloitte, go to deloitte.com/about.