CLOSING THE LOOP
If we could measure the world, how would we manage it differently? This is a question we've been asking ourselves in the digital realm since the birth of the Internet. Our digital lives—clicks, histories, and cookies—can now be measured beautifully. The feedback loop is complete; it's called closing the loop. As you know, we can only manage what we can measure. We're now measuring on-screen activity beautifully, but most of the world is not on screens.
As we get better and better at measuring the world—wearables, Internet of Things, cars, satellites, drones, sensors—we are going to be able to close the loop in industry, agriculture, and the environment. We're going to start to find out what the consequences of our actions are and, presumably, we'll take smarter actions as a result. This journey with the Internet that we started more than twenty years ago is now extending to the physical world. Every industry is going to have to ask the same questions: What do we want to measure? What do we do with that data? How can we manage things differently once we have that data? This notion of closing the loop everywhere is perhaps the biggest endeavor of our age.
Closing the loop is a phrase used in robotics. Open-loop systems are when you take an action and you can't measure the results—there's no feedback. Closed-loop systems are when you take an action, you measure the results, and you change your action accordingly. Systems with closed loops have feedback loops; they self-adjust and quickly stabilize in optimal conditions. Systems with open loops overshoot; they miss it entirely.
Take agriculture, for example. In agriculture you plant seeds, you fertilize them, you spray them, and then you wait six months and hope for the best. Once upon a time, people really understood the land, they understood the plants, and they had this intuitive feel for what would happen. Then farms got bigger and there were fewer people on them. We didn't have the ability to walk the land, and feel it, and close the loop. In the absence of information, we had to resort to monoculture. We closed the loop on an industrial scale. We went to predictability: If I use this Monsanto seed, spray it with this chemical, in this very well understood soil, I'm likely to get a pretty good outcome. In the absence of information, we went with predictability, simplicity, and a lot of chemicals.
There's a real cost to that. Monoculture has its costs in terms of genetic diversity. Chemicals have their cost on the environment and the food. So how are we going to measure agriculture again? We have industrial farms that are massive, so we're not going to populate the farms with people walking them anymore. Obviously, we want sensors. This starts with satellites, airplanes, sensors on the ground and maybe on the tractors, and then drones. But we now have the ability to start to measure those crops again.
The hope is that by closing the loop on agriculture, we can make smarter choices. Right now we say, "Hey, it would suck if we had a fungal infection, because by the time we found out, we'd have lost the whole crop, so we're going to prophylactically spray fungicide everywhere." Why would you do that? Because the fungal infection could happen anywhere, and the farm is ten square miles. But what if I told you that you could find out, on an hourly or daily basis, whether you had a fungal infection anywhere in the crop? Then you could wait until you had a problem before acting on it, and you'd spray fungicide only where it was needed. It would lower the chemical load in the ground and the plants, and it would be more efficient. And it can be done simply through measurement.
If you knew, you would act. If you don't know, you have to do these dumb prophylactic scalable solutions, which are bad. That's just one example where we used to be able to measure, then we lost that ability, and now we have the ability to measure again by using sensors and information to compensate for boots on the ground. This is true for industry after industry.
All manufacturing right now is based on these computer-controlled machines, CNC machines—lathes, mills, and things like that. They are incredibly sophisticated and complicated, but they need these experts to program them, which is hard. They have to understand exactly what they're making—the cutting tools, the materials, et cetera. These experts are now around sixty years old, and we're not training nearly enough replacements (at least not those who can pass newly required drug tests). So the work goes to China because they're still making experts who can run the CNC machines. Why do we need experts to program CNC machines? The answer is because they're dumb machines; they're open-loop machines. You feed them the commands and they just act on them. If the commands include things that are bad like, "Tear yourself to pieces," they will.
Obviously, these machines should be measuring themselves. They should be measuring vibration, where their heads and bits and pieces are, and their material. They should be smart enough not to do stupid things. We made them just smart enough to be operated by computers, but not smart enough to operate the computers themselves. In other words, they're open-loop machines and we need to make them closed loop.
That's another example where sensors and connecting to the Internet would suddenly mean that you don't need an expert or a big overbuilt machine; it's smart enough to figure things out on its own. That is the endeavor. We've built these fantastic sensors and computing environments, and we have lots of information right now. We now need to extend the tendrils of the Internet out into the physical world and start measuring things, act on that information, and then make smarter choices about our planet.
The first notion of the power of feedback loops came from Norbert Wiener's construct of cybernetics. That was in the early days. Some of those cybernetic machines were actually mechanical. When you think of the regulators in steam machines, with these little balls that would come out by centrifugal force and slow the machine down, that's a form of feedback loop. It's an analog cybernetic machine. Then they started with the early digital cybernetic machines, with analog oscillators. We've now extended that concept of feedback into the entirety of the Internet, and soon beyond. You could call this the new cybernetics, but it's so universal that it almost doesn't need a special name. This is just measurement.
There're two kinds of failures in the failure of an idea. One is that it's wrong and abandoned. The other is that it's right and adopted. If it's wrong and abandoned, off it goes. If it's right and adopted, it's typically so right and adopted so broadly that people don't think of it as an idea anymore. If you show a kid a phone, they think that it's wireless and it has a screen. This is what they think of as a phone. You don't have to call it a wireless phone anymore. The idea is internalized.
The notion in the digital world of cybernetics is now completely internalized. Of course you've got server logs, of course you have A/B testing, of course Amazon is watching what you do to improve the experience. We don't call it cybernetics anymore; it's just called measurement. In the physical world, which hasn't adopted the digital concepts, it is still so alien that we almost need a new word. We talk about the Internet of Things, we talk about the Industrial Internet, we talk about Geospatial Information Systems; it's a whole different language than the ones we use on the Internet. Maybe the point is just that the rest of the world needs to think more like the Internet, but because they have different cultures and different incentives, they haven't adopted the same language. Maybe they never will, and that's okay.
I use the phrase closing the loop because that's the phrase we use in robotics. Other people might use the phrase big data. Before they called it big data, they called it data mining. Remember that? That was nuts. Anyway, we're going to come up with a new word for it.
It goes both ways: The tendrils of the Internet reach out through sensors, and then these sensors feed back to the Internet. The sensors get smarter because they're connected to the Internet, and the Internet gets smarter because it's connected to the sensors. This feedback loop extends beyond the industry that's feeding back to the meta-industry, which is the Internet and the planet.
~ ~ ~
How does one become a roboticist? That is a great question. I have no idea how to become a roboticist because, like any properly humbled roboticist, I don't call myself one, and I'm certainly not. But I will tell you how I became the CEO of a robotics company.
I was trained as a physicist in the late 80s. I turned out to be a bad physicist. I was just terrible compared to the other physicists at the time. I was going to school at UC Berkeley when the first generation of Chinese students were coming over. They were brilliant, incredibly focused, and picked from the best of the schools.
I struggled on, went to Los Alamos, and I thought, well maybe I'm not going to be a Nobel Prize winner, but I can still be a scientist. I was destined, like the rest of the scientists and the physicists of my era, to work on the Superconducting Supercollider. You'll remember this. It didn't end well. It was canceled because of cost overruns. All of us who were in physics and had these romantic heroes—the Feynmans, the Manhattan project—realized that our career trajectory would at best be working on one project at CERN for fifteen years. That project would either be a failure, in which case there would be no paper, or it would be a success, in which case you'd be Author 300 on the paper and become an assistant professor at Iowa State.
That didn't sound like a lot of fun. Most of my classmates went to Wall Street to become quants, and to them we owe the subprime mortgage. Some of the others went on to start the Internet. What's the connection? The Internet. The modern Internet started in that physics world.
Even though physics was blowing up, our ability to measure and have instruments was destroyed by funding. We did three things. First, we built the Internet by connecting physics labs. Second, we built the Web. The Web, as you know, started at CERN, with physics. Third, we were the first to do big data. The only big data at the time was physics data. We had supercomputers—Crays—which were half the power of your phone now, but they were the supercomputers of the time. We also had the statistical techniques. When you look at the Internet, the Web, and big data—the same stuff that today is Bayesian statistics and Google, to say nothing of Wall Street—we had portable skills.
Those who didn't go to Wall Street to use their big data skills—Wall Street, by the way, was the adjacent big data space—went and started the Internet. Some people started Internet infrastructure companies, some created Netscape, and browsers, and software. I did none of the above. I went to the science journals to write about the use of this interesting communication technology in science and then outside of science. I was at Nature, then I was at Science, and then I went to The Economist to start the Internet coverage there. Meanwhile, we were reading this magazine called Wired, which had just come out in '93, and we realized that this tool we scientists use could have application for everybody.
It wasn't just about scientific data; it was a cultural revolution that could change the nation-state. It was mind-blowing. So when Conde Nast, who owned Wired at the time, asked me to take over the magazine, I was like, absolutely. This magazine changed my life. This is the biggest story of our time.
You can ask why I was drawn to physics, and that was probably because it was romanticized at the time. Why was I drawn to computational physics? Again, it was in the air. Computers were cool, and given the choice between being good at math or being good at computers, the computers seemed easier to me. Once I was in computers, why was I drawn to networks? If could telnet into the Cray, why wouldn't I? That would be awesome. Why was I drawn to the Web? The cool kids were doing it, and I was just in the right place at the right time. I have an average IQ, but incredible luck. I've always been in the right place at the right time. It just so happens that being in physics in 1987, working at Los Alamos and telnetting into the Cray, was the right place at the right time to make you realize there was something going on.
Then you have this Day-Glo magazine called Wired, which is basically showing you your same world from a much more exciting perspective. They were based in California, and they were super sexy and cool, but I actually knew this stuff. I was present at creation. This magazine, which was just covering it and not doing it, says that it's much cooler than science, and that maybe someday I might get a date. Awesome. Sign me up.
That was epiphany number one: The Internet is big; bet on it. Epiphany number two came in 2001, after the stock market crash and the dot-com bust. Everyone thought the Internet was a hoax, like it was the subprime crisis of its era. I still believed in it. I thought the crash was a stock market phenomenon, but I knew the Internet was real. So I counter-cyclically bet that this was like the first minute of the day, I moved to California, and I bought at the bottom.
There were some more epiphanies along the way. I lived in China for a couple of years, from 1997 to 2000. Everything I'd read about China was wrong. They were going to crush it, and that was a bit of an epiphany.
My third epiphany gets me back to robotics. At the time, I was the editor of Wired, with a background in computational physics, and I had five children. I wanted to know how I could get my children interested in science and technology. My wife had been an editor at Nature and our kids were playing video games. So I thought, robots, Lego Mindstorms Robots. That will be cool. The kids brought them home and they said, "No, that's totally not cool. I've seen Transformers. Where are the freaking lasers? These things are slow, they take all morning, and they do nothing." Real robots are not as exciting as CG robots. I thought, what would be cooler than a rolling robot? What about a flying robot? So I Googled flying robot. If you Google flying robot, the first result is a drone. I thought, yeah, I guess a drone is a flying robot. Then I Googled drone. A drone is an aircraft, piloted by an autopilot. I thought, yeah, a plane with a brain. Then I Googled autopilot. Okay, I get it—sensing, computation, and lots of math.
So I said, "Kids, let's build a flying robot with Lego parts." We did it, and it was constructed like a model airplane. Mindstorms came out with these sensors, gyros, accelerometers, magnetometers, and we connected it to a GPS. It flew, but barely. The kids, of course, lost interest and went back to their video games. But I was stunned. That should not have been possible. Drones were purely military stuff—Global Hawks, Predators—and we built one on the dining room table, with nine-year-olds. I thought there was a glitch in the matrix, a disturbance in the force. Something happened that made it possible for a dad and his kids to build a drone.
I set up a website—DIY Drones—on which I shared my discoveries and progress, and I started doing research. Of course, everyone else starts doing it because, like any great idea, everybody had it simultaneously. The year 2007 was the year that everybody woke up to something being different about hardware. The Nintendo Wii had come out. The Fitbit guys got a Wii and took it apart to find out what was in the stick that made the cursor move—an accelerometer. What else could you do with an accelerometer? Boom—Fitbit. Three-dimensional printing had come out. Arduino, the open-source computing program had come out. The maker movement was burgeoning. Make magazine had come out. The iPhone came out that year.
Moore's law, which had been trundling along at a very good rate with personal computers and networks, doubled in speed, but only with the components inside the smartphone—MEMS sensors, cameras, wireless, GPS, battery technology. All of a sudden, these things, which were operating at the economies of scale of Apple, and Samsung, and Google, those components became available to regular people. Those components then started to transform adjacent spaces so that a regular guy, like me, could become a roboticist. A little bit of background in computational physics, which is what drones are; a little bit of incentive to try new things, which is what being a dad is; and then, thank you iPhone et al, we suddenly do stuff that would have been impossible before.
Five years earlier, those sensors would have been mechanical. Each one of those sensors would have been a little mechanical cylinder with a mechanical gyro in it that cost $10,000 each, and you'd need nine of them. You'd need a $100,000. The mechanical components weighed maybe 500 pounds. They were export controlled. You would have had to have been military to have one. By the time I got there, it was a tiny chip that cost $3. I was a roboticist because anybody could be a roboticist. Suddenly, robotics got easy.
In 2007, it went from being a weekend activity to a two-weekend activity, and then setting up a website. The website took off. Everyone was contributing PCB designs, and software, and aerodynamic stuff. It was one of those homebrew computing club moments.
Then the next generation came along months later, and they liked what we were doing, but they didn't know how to compile code and fab a PCB. They asked if we could just do it for them. So we thought we better make kits. How do you do that? With bits and pieces, pizza boxes, things from China. We turned the dining room table into an assembly line, with post-it notes on all the bits and pieces, and the six year old was in charge of quality control, and we made these kits. This was a robot blimp kit called the BlimpDuino. We packed up all the boxes, took them to Maker Faire, and sold out in fifteen minutes. I said, "Good news, kids, our robot blimp is super popular. We're going to have to make more." They said, "No way."
So I found a guy on the Internet, from the DIY community, to help. That was Jordi Munoz, who I'd seen in a video flying a helicopter with a Wii controller. I said, "Jordi, you're supersmart, do you want to help me make these kits?" He said, "Yeah, I just moved to Riverside and I'm waiting for my green card. I can't work, so I've got a little time on my hands. Yeah, I'll do it." I sent him a check for $500, he bought the parts, and he started sending me pictures of his progress.
First, he starts on the dining table, then it was a garage, then it was a slightly bigger garage, and then a little industrial space. Then he had help. Then he bought some production equipment—pick-and-place machines, reflow ovens, things like that.
He kept sending me these pictures, and every time they'd get more and more professional. He's buying pick-and-place machines, and I'm Googling to find out what a pick-and-place machine is. Then he opens up a second factory in Tijuana. Again, I had only sent him a check for $500. He opens up the second factory in Tijuana, and now these people are wearing aprons and they've got electrostatic discharge things, they've got ERP systems, and e-commerce systems. He's showing me the books, and we're going to have $5 million in revenue that year. We have a Tijuana drone factory.
I was like, who is this guy? What just happened here? When I met him he was nineteen years old, in Tijuana. He hadn't gone to college. His wife was pregnant and they came to the United States to have the baby. Three years later, he's running a Tijuana drone factory. We were making more drones per month than all of America's aerospace companies combined, we hadn't taken a penny of financing, and I was still at my day job at Wired.
We had already derisked the company so it was time to grow faster. We got some venture capital. I quit my day job at Wired and took over as CEO. We built another office up here, in Berkeley, so we could have a Silicon Valley outpost. We ended up raising $125 million and became the biggest drone company in America, and that was in about three years. We started off as a father-child project. Then it was a hobby. Then it was a community. Then it was a little DIY company run out of a teenager's bedroom. Then it was a bigger company. Then it was just making bags of parts, literally, printed circuit boards in bags.
I remember when we decided to get more professional. We started soldering for people, because people suck at soldering. Then we created instructions. Then we put things in boxes with labels on them. Then we developed customer support. We took compilers out. When we originally did this, you had to compile the code yourself. Then we put the code on it for you. Years into it, we're thinking, are we supposed to be getting FCC certification? Yeah. Okay, we should do that. How you do that?
We literally learned the business from scratch. We went from being a big DIY company to making drones. We made lots of drones, and then we made so many drones we couldn't make them in Mexico anymore, so we moved them to China. Then we didn't make the drones anymore, our partners made the drones. Then we made 100,000 of those. Then it turns out that even though we were making them in China, the Chinese were making them even cheaper, so we decided we couldn't make drones at all, couldn't make hardware at all. Never fight a ground war in Asia, as they say.
We went from military industrial complex to the shelves of Walmart in about four years. That's never happened before. The stuff on the shelves at Walmart is better than the military stuff. The reason the toys are more sophisticated than the military drones is because the users are less sophisticated. They just want to push a button and have magic happen. The military wants to train pilots to learn how to do things right. Consumers just push a button—magic.
Eventually, we gave away all of our technology—not all of it, but we gave away the software as open source, so we basically enabled our competitors. We sowed the seeds of our own hardware destruction. Three years later, everyone took our technology and made it faster, cheaper, and better. So we were out of the hardware business.
When you're actually in Shenzhen and you're Chinese, your supply chain is better, you're faster, et cetera. We, on the other hand, were operating via contract manufacturers and through middlemen.
Many of our Chinese competitors were also willing to live on 2 percent margins. By contrast, we're a venture-backed company and we're supposed to be living on 20 percent margins. Silicon Valley companies cannot live on Shenzhen economics. So we had to go from hardware to enterprise software, which has the required economics. Now we do the software and the data around the drones. We went from technology to toys to tools, and from making drones to just using them—in short, we closed the loop.
We started with the idea that you could build your own drone. Then, you didn't have to build your own drone. Now it's like, here's what you can do with the drone. We assume the drone exists, and all we care about is the data that the drone collects and what you do with it.
~ ~ ~
We're partnered with Autodesk. Autodesk's big customers are construction companies, architecture, AutoCAD, et cetera. Construction is a perfect example of an unclosed loop. You design buildings on the screen, but the moment you dig your first spade of dirt, they're analog, not digital; you've lost the ability to manage them. They become open-loop systems. If you've ever built a house, you know that with cost overruns and mistakes your cost basically blooms 2X. You're lucky if it's only 2X. That's all because you can't measure it while it's happening, or if you can measure it, it's too late. What they want to do is they want to basically re-digitize the physical world.
By the way, construction is arguably the biggest industry in the world—offices, homes, and roads. If you could measure it while it's happening, you could manage it much better. You could spot the mistakes and know what's happening where. You'd have this chain, you'd know who did what, look at deltas and deviations. In the Autodesk parlance, this is called “reality capture.” They want to capture reality so they can manage it with the same tools they use to design reality. They have all the fantastic tools for the analysis, but what they're missing is the capture part. We capture reality so that they can analyze it.
Let me say one last thing. Much of this is due to the phone in your hand. It's the most advanced consumer device ever made, aside from the car. Basically though, think of it as a lump of black plastic. It was made probably around December of last year. If you buy a device—a car or refrigerator or whatever—you buy it, take it home from the store, and every day it gets worse. Cars lose value when you drive them off the lot. Except for this lump of black plastic. This lump of black plastic gets better. Why does this lump of black plastic get better? It's because it updates—the apps get better, the operating system gets better, the networks get better. What's unique about modern drones is that they work the same way—they’re basically the same sort of lump of black plastic as your phone, but with propellers. It inherits the same characteristics of a smartphone or any other connected device, which is it gets better over time.
Most of the devices in our life, from our cars to our homes, are “entropic,” which is to say they get worse over time. Every day they become more outmoded. But phones and drones are “negentropic” devices. Because they are connected, they get better, because the value comes from the software, not hardware. Although my device was made in December of last year, it just got a new software update today. It all of a sudden got new features, and it got better performance, and it's going to continue to get better and better for years, until we've reached the limits of its hardware. All connected devices are potentially negentropic devices, which is a very big deal—a reversal of the arrow of consumer experience.
