As part of our AI For Growth executive education series, we interview top executives at leading global companies who have successfully applied AI to grow their enterprises. Today, we sit down with Kevin Scott, Chief Technology Officer at Microsoft.
As the CTO of Microsoft, Kevin drives the technology giant’s AI strategy and services. In this interview, he focuses on the intersection of AI and IoT and reveals how enterprises have successfully leverage the combination of these two emerging technologies to drive real business value. He also shares insights from his visits to industries ripe for disruption by AI and automation and key learnings for how managers can best prepare their workforces for the future.
Mariya Yao: Hi everyone, this is Mariya with TOPBOTS. Welcome to our AI for Growth executive education series, where we interview the top leaders and companies that are successfully applying AI to enterprise problems. Today, I’m very excited to be joined by Kevin Scott, who is the CTO of Microsoft.
Kevin, a couple weeks ago we were having a discussion over lunch, and you mentioned the extraordinary impact of the combination of AI and IoT that’s transforming enterprise workflows. Can you give our audience a sense of where you’re seeing the most opportunity and the most ROI in this space?
KS: Just stepping all the way back and looking at the trends, [this is] one of the more exciting times in computing since the early 90s. There are a few things that are happening at once that are combining in a very interesting way.
One of those trends is that IoT itself is exploding. There are different studies from a variety of different sources. The Gartner study on IoT devices indicates that we’re probably going to go to somewhere north of 20 billion devices by 2020.
These are computing devices connected to the Internet, and [as] a frame of reference, we’re probably at 11 or 12 billion IoT devices right now. There are about a billion PCs and two and a half billion smartphones, so the IoT sector is an order of magnitude larger than the largest computing platform that has emerged today.
That is in and of itself an incredibly exciting thing and a really interesting opportunity for all of us. When you combine that with the fact that silicon is becoming much more powerful at an accelerating clip [and] you are thinking about the types of silicon required for doing AI model training and AI inference, that particular type of computing power is growing by maybe a factor of 10x in terms of price performance over the past five years. We see that trend line continuing for probably another five orders of magnitude emerging over the next eight years or so.
That has obvious implications for the high-end of computing, where in the cloud you’re gonna have huge amount of additional capacity over the coming years, like build more sophisticated models. It also means that the power of AI is coming to consumer price point devices on the edge of the cloud in this IoT environment
Take that, and you take the fact that these IoT devices are increasingly sensor-equipped, you really do have what we think is gonna be a new computing paradigm. We’re calling it the “Intelligent Edge”, because it’s not just about the fact that computing is becoming ubiquitous and merging into your environment where any room that you’re gonna step into is potentially gonna have tens of these devices, each capable of sensing what’s going on inside of its environments and reacting intelligently to it.
It really is gonna require a bunch of change in the way that we’re thinking about how we build and manage these systems.
MY: What are some examples? You mentioned so many more AI applications are going to come to consumers when it comes on these edge devices. What are some applications that maybe weren’t possible before, but that businesses should now be thinking about, given the proliferation of IoT devices and of AI.
KS: You’re already seeing the early stages of these things in the intelligent speakers that are coming out, but I think that’s really just sort of the tip of the iceberg.
If we do our job right over the coming years, you’re gonna start to see more and more applications. One of the interesting ones that has been written about are these smart stores that are retail outlets where they’re using IoT devices and cameras, shelf sensors and a bunch of AI in computer vision models to identify you as you come into the store and just look at which times you are putting in your shopping cart and taking out of the store, where you don’t even have to check out.
There are more and more of these stores popping up as proofs-of-concept. It’s not that [there] necessarily is going to be this wave that sweeps through retail and redefines everything. We should all look at that as an inspiration for the sorts of things that you could do with this new technology.
Just by way of an example, my wife had surgery early this year. I was getting to hang out a lot in the surgical unit recovery area at one of our hospitals here in the Bay Area, and I was noticing all of the processes and workflows.
One of the things that that happens when you’re in recovery from surgery is the doctors want you to get some level of activity, but they want to make sure that you’re not overly active. You might injure yourself after the surgery you’ve just had.
Right now, the way that they monitor your activity is they have nurses. In this particular hospital, there were four on shifts for this entire ward, and there’s no way that these nurses can keep a close eye on every one of the patients that were in recovery.
But if you look at these IoT devices with cameras and computer vision models, it should be very easy for us to write software in this new world that would identify when my wife is in the common area walking around, and they can add to her tally of activity.
If she’s below her activity level, you can alert the nurses at their workstation or on their mobile device and say, “Patient Scott isn’t getting up to their daily level of prescribed activity today”. Of if they’re overactive, it can send an urgent alert [to] to go find this patient right now and get them back to their room.
I think there are going to be hundreds of thousands of scenarios that this flavor of software can power right now. Right now, we’ve got some packaging issues with the technology. We need to do some more work to make it more accessible to more folks, but part of the problem or challenge, I should say, is getting people to imagine what’s going to be possible in this new world.
MY: Right, because when you take IoT and put it with AI, you’re talking about bringing two huge trends, two highly technical and very difficult to understand technologies together, so there’s definitely going to be a lot of challenges implementing that on an enterprise scale.
In your experience, what are some of the things that executives can do to better prepare and increase their chances of success when implementing these kinds of AI + IoT applications?
KS: I think the biggest thing that you can do is availing yourself of some of the common infrastructure that’s emerging right now in the cloud. Basically, we’re talking about IOT and the first thing that I mentioned is the cloud, but having that the cloud is the sort of coordination backplane for everything that’s happening on IoT. Making sure that your data is in the cloud, that you’ve gotten yourself into a good state where you’re comfortable with your data governance, you understand what pieces of data you do and don’t have, will really help inform the types of AI that you’re gonna be able to build.
Then, getting your organization thinking about all of the AI tools that are available right now. Some of these things are still incredibly elite, [but] some of the tools though are getting to be incredibly easy.
Like the computer vision things—it’s being a little self-serving here as CTO of Microsoft—you can use our Azure cognitive services APIs to do computer vision stuff, for instance.
We have trained a bunch of baseline models for computer vision for you, but you can come to us with your bespoke data of things that are unique to you, and you can add your data to our models and get a customized model out of the other end that lets you do things like identify the faces of your employees, friends, and so on.
Or if you are in manufacturing, for instance, being able to identify your inventory and your parts that you are using in your manufacturing processes… Making yourself aware of what these capabilities are, I think it’s a really important thing right now.
The other thing is thinking through what your security policies are. It is really important. One of the really interesting things again that we all will have to think through with this explosion of connected devices is that it’s gonna present a security challenge that is far more interesting, even than the smartphone laptop BYOD sets of issues that enterprises have. Do you allow someone to take a smart IoT device and add it to your corporate wireless network?
Some companies are already thinking through this with these smart speakers. I’ve chatted with folks who are have no Amazon Echoes or intelligent smart speakers on their corporate networks. That may be a knee-jerk reaction that cuts you off from interesting future possibilities
MY: As AI becomes embedded in everything, there is a natural fear, especially exacerbated by the media, [that] the combination of AI with IoT is going to disrupt workforces and put people out of jobs.
I know that you’ve spent a lot of time thinking about this, and you believe that that does not have to be the case at all. Can you share more of your thoughts and stories on this particular topic?
KS: We as a society and we as a technology industry get to choose the path that we walk down. The technology industry is building these tools and capabilities, and the rest of the industry, government, and society are deciding how to get deployed.
One of the interesting and super fun things about my job is [that] I get to see a fairly broad spectrum of AI development.
For instance, two of the most inspiring things that I’ve seen technologically over the past year are the developments in precision medicine and precision agriculture. Precision agriculture, for instance, we are entering an era where this intelligent edge, like having these AI-capable devices everywhere including [and] being able to mount them in drones, is allowing you to gather more interesting data about agricultural operations.
A few years ago—and this is probably still state of the art—if you want to build a hydrology model for your crops, [such as] to understand where the wet and dry spots are in a field, to try to
optimize how you’re delivering water to make sure that you’re wasting as little water as possible, and [making sure] you are getting the exact amount of water that [your croops] need, you’d have to go through this incredibly expensive and tedious exercise of putting a bunch of water sensors all over the place and flow meters inside of your mechanical irrigation systems.
You’d have to have fairly large-scale agricultural operations to do this, and it was an elite thing. Now, you can take a thousand-dollar drone that’s got the equivalent of a Raspberry Pi running a computer vision model, like flying over a field, and they can build a fairly high accuracy hydrology model for that field. You can then optimize your irrigation [with that data]. It’s virtually free AI running on super cheap commodity hardware.
That is a flow of AI, where the technology is creating abundance. It’s not concentrating power into the hands of the few, it is making things that were inaccessible to tons and tons of people to orders of magnitude more people. I see that trend happening across the board in R&D, in agriculture, and these innovations swill be flowing out into the economy over the next five to ten years.
The same thing is happening with medicine, where you take this combination of increasingly ubiquitous data about the human body
that’s coming from smartwatches or fitness bands, then coupling this data with contemporary AI, like deep neural networks, and the things that you’re going to be able to do are really incredible, like predicting serious health conditions for virtually free before a patient is symptomatic when it’s relatively easier to fix the underlying health condition than it is after the patient is sick.
These things can potentially transform the world in this positive way, and what world we get is going to depend largely on whether we’re thinking about AI. Is this an empowering technology that creates abundance versus this narrowing technology that concentrates control?
I’m a huge proponent and hugely optimistic about the potential of the former.
MY: There’s no doubt that AI has so much benevolent potential for a society, especially in the areas that you mentioned, precision agriculture and precision medicine.
I want to dig into this argument that people will sometimes throw out: they’ll say, “Okay, now that you have a drone with AI doing these hydrology models, what happens to the guy whose job it was to build these models by hand?”
What does it mean to some of these people whose jobs are being automated? You are really seeing this in industrial applications, where people’s jobs have literally been automated.
What is your thought on that, and what has been your experience analyzing these different industries on what’s really happening with automation?
KS: There is disruption happening, but what I’m really seeing with these things is, if you’re a small local organic farmer in eastern Washington State [and] we have partners we’re collaborating on [that fits] this exact profile, there was no guy building a hydrology model before. This technology wasn’t accessible to folks who were running a small operation.
You take that to the developing world, where we’re really seeing some huge impacts happening right now. This definitely wasn’t a guy on the small farm in rural India building hydrology models or building AI that’s accurately predicting when folks should be planting crops.
There was no one doing the work before, and what you get when you apply the technology is just more productivity and better quality products with less detrimental side effects like to the environment.
But you are right, there are places where there’s job disruption. I’ve been doing this for a really long time, the first machine learning system I built was about 15 years ago, and the thing that I think we will see is that these machine learning systems have this huge potential to create the opportunity for people to do higher value work.
It’s not that you’re permanently displacing jobs. Usually the machine automates the most tedious things in the world, and the thing that you can free someone up to do is much higher value.
I’ll give you one final example: when I was a young engineer, one of my first jobs was working for an electronics contract manufacturer This is a company that was less than 20 people in Lynchburg, Virginia. You had a very small number of people trying to make this business work, so you had people who would do QA on circuit boards, they would do assembly, they would do post assembly testing. They were context switching across a bunch of different things.
I’ve imagined how computer vision for doing QA in my old business. It would have helped out with things, like with this process called infrared reflow soldering, you could totally put a camera on either end of this reflow solder machine. [It would] look at a circuit board before it goes in, and when the circuit board comes out of the machine, [AI would] basically replace the visual inspection that a human being would be doing.
In the context of [my] old company, it wouldn’t have eliminated anyone’s job. It would be the super tedious thing that was distracting [the workers] from something else higher value that they would prefer to be doing and generated more value for the company.
The thing with AI—this is the misconception I think people have—is [that] it is not going to be this sort of god-like, human-resembling thing that comes in and replaces that the need for human beings and the economy. It’s a thing that can come in and make tedious work across a bunch…I mean hundreds of thousands [of tasks disappear]. [There’s a] super long tail of AI applications that people are gonna build the same way that people built hundreds of thousands of applications when PCs became ubiquitous a few decades ago.
The things that they build, like there’s going to be this whole industry that gets created out of the building, it’s going to make a gazillion jobs. The things that they’re automating is going alleviate people from doing a whole bunch of tedious work, so that they can find the higher value things that human beings are uniquely situated to doing
MY: Thank you so much for that example. Kevin, I love the personal story of this tedious thing that you would love to fix with AI and IoT, so I really appreciate the positive attitude. We definitely need more of that when we are thinking about AI apps that we can build for ourselves, for our companies, and for our society.
Thank you so much, Kevin, for being on the AI for Growth executive education series. Really appreciated your commentary.
KS: Thank you for having me. It was a pleasure!
Learn How To Drive Business Results With AI & Machine Learning
To get a deeper foundational understanding of what AI & machine learning technologies are and how to use them successfully in your business, we recommend you read our bestselling book, Applied Artificial Intelligence: A Handbook For Business Leaders, which is available for digital download below and also via paperback and Kindle versions on Amazon.