Guest: Dr. Mark Maybury, Former Chief Technology Officer at Stanley Black and Decker
Host: Seth Earley, CEO at Earley Information Science
Chris Featherstone, Sr. Director of AI/Data Product/Program Management at Salesforce
Published on: November 9, 2022
In this episode, Seth Earley and Chris Featherstone speak with Dr. Mark Maybury - former Chief Scientist of the Air Force, CTO of Mitre, and first-ever CTO of Stanley Black and Decker - whose career spans five decades of AI research, national security, advanced manufacturing, and startup advising. Mark traces his inspiration from watching R2-D2 and C-3PO interact naturally with humans in Star Wars at age thirteen, through a PhD at Cambridge where he invented the first computer program to write multi-paragraph multilingual texts without modern machine learning, to leading digital transformation at an eleven-billion-dollar global tool manufacturer. He also shares how he helped lay the foundation for the entire field of sentiment analytics through a government-funded program called the Opinion Bank, and why the key to AI success has always been understanding the human first.
Key Takeaways:
Insightful Quotes:
"When I saw R2-D2 and C-3PO interacting with humans in a very natural, cooperative way, I walked out and distinctly remember saying - that is so cool, that's what I want to do with my life. That started my love of artificial intelligence." - Dr. Mark Maybury
"You become the learning system. There was no machine learning field. So I went in and asked: how do you do this? I interviewed doctors, English professors, mathematicians - I learned from the humans, then I taught my machine how to do it." - Dr. Mark Maybury
"I love the big because the bigs, if they can be retrained, you can make the elephant dance. And that can be quite powerful if you get them dancing in the right direction." - Dr. Mark Maybury
Tune in to hear Dr. Mark Maybury describe what it was like to do foundational AI research before modern machine learning existed, why sentiment analytics traces its roots to a government-funded annotated data initiative, and how a single spark from a Star Wars screening at age thirteen sent one scientist on a fifty-year mission to build machines that think, collaborate, and protect.
Links:
Thanks to our sponsors:
Podcast Transcript: Innovation, Digital Transformation, and Five Decades at the Frontier of AI
Transcript introduction
This transcript captures a wide-ranging conversation between Seth Earley, Chris Featherstone, and Dr. Mark Maybury about a career that has touched nearly every dimension of artificial intelligence - from inventing early natural language generation systems at Cambridge in the late 1980s, to running the Air Force's AI lab on a fleet of Lisp machines, to leading the digital transformation of a global manufacturing giant, to founding the data annotation program that became the basis for sentiment analytics. Along the way Mark shares his philosophy on cross-disciplinary learning, the power of asking humans how they think before building systems to replicate it, and what excites him most about the next generation of AI applications in robotics, nanotechnology, and national defense.
Transcript
Seth Earley: Welcome to today's podcast. I'm Seth Earley.
Chris Featherstone: And I'm Chris Featherstone. Good to be with you.
Seth Earley: Our guest today is a real Renaissance man with broad experience across new product development, innovation, digital transformation, AI and analytics, advanced manufacturing, and commercialization. He's just started a new advisory role at Ready Robotics, helping them enable intuitive interfaces to cooperative robotics. Please welcome Mark Maybury.
Mark Maybury: Seth, Chris - great to see you. I'm not too far from you, actually, just outside of Boston in Chelmsford. Beautiful fall weather here too.
Seth Earley: Give us the arc - how you got to where you are, the world according to Mark.
Mark Maybury: I've been very, very fortunate to have a rather blessed life. The arc started at home - parents who always emphasized education as a doorway to the future. My actual spark came when I was thirteen. I walked into an immersive environment - a Dolby stereo special effects theater showing Star Wars - and I had always loved math and science. But when I saw R2-D2 and C-3PO interacting with humans in a very natural, cooperative way, I walked out and I distinctly remember saying: that is so cool, that's what I want to do with my life. That started my love of artificial intelligence.
My father was a computer executive in Cambridge - an Army officer who went to MIT and then Raytheon and then ran an international company. He came home one day from a flood in the basement and we had a PDP-11 in my brother's bedroom. We got to actually learn about training machines - we had the manual and we were teaching ourselves: how do we turn this on, how do we initialize it, what is a program? Absolute influence.
In college I almost flunked out my first month and a half. They called me into the Dean's office - I was taking all upper-class courses as a freshman, including organic chemistry with pre-med students who'd already been to college. But I went from the bottom to becoming valedictorian and cadet commander. I graduated, went to Cambridge on a Rotary Foundation scholarship, got my Master's in computer speech and language, then my MBA and PhD. That led me into a career in national security - from intelligence to command and control, eventually becoming the Chief Scientist of the Air Force, the highest scientific position in the Pentagon, a three-star equivalent role. Then I came back to Mitre, which manages seven federal laboratories, as Chief Technology Officer and Chief Security Officer. And then I got a call - as often happens, you don't apply to some of these jobs - and I ended up as the first Chief Technology Officer of Stanley Black and Decker. I just retired from that role. I'm about to start a new executive position I can't disclose yet, but it will be with a very large defense contractor.
In the meantime I sit on four boards. Ready Robotics is one advisory role. I'm also advising Flor Coatings, a nanotechnology company; Halo Energy, a wind energy company that puts a fourth ring around traditional turbines to enhance the Bernoulli effect and increase efficiency by about fifty percent; and Internet Sciences, focused on IoT and cybersecurity, which is planning to go public.
Seth Earley: You mention you have no spare time, and yet somehow you do.
Mark Maybury: All of my spare time is planned. I don't wake up on a Saturday and say "what am I going to do?" Tomorrow morning I'll be up at six-thirty to drive into Boston, pick up my daughter, and we'll coach a young women's ice hockey team that is currently ranked number two in the nation. The goal is to get to number one - we'll see. But I do believe a full life where you plan and engage with your children, your grandchildren, your colleagues, and your personal joys - that energizes you and brings renovation.
Chris Featherstone: Ice hockey - I look like I have an inner ear problem if I get near the gates. But phenomenal, congratulations.
Mark Maybury: It's a wonderful sport. Team sports teach kids individual discipline and team success simultaneously. Our girls are twelve and thirteen, and we're teaching them what they want to put in their bodies, how training as individuals and teams makes them more successful. I've seen all my own children skate - one at the collegiate level, several who have been coaches with me. It's emotional, physical, strategic, and tactical. A great life experience. And there is data now about the brain and music and sports - everything from memory in the hippocampus to emotion in the amygdala to planning in the prefrontal cortex. I'm a professionally trained percussionist, I play guitar, and I have a personal affiliation with jazz. Jazz requires you to understand all the basics - read music, understand the mathematics, the structure, transpositions - and then build on top of those and improvise. Just like hockey, you learn all the basics of edge control and puck control, and then scaffold rhythm and call-and-response on top. These patterns turn out to be consistent across disciplines. Always tell people: don't be a single-sport person, cross-train. Don't be a single-discipline person.
Seth Earley: Tell us about your transition to Stanley Black and Decker. You went from being Chief Scientist to a $11 billion commercial manufacturer - what was that like?
Mark Maybury: In some respects it was exactly like other executive transitions. In others, it was completely different from what I expected. The Air Force is a hundred-and-ten billion dollar organization - if it were for-profit it would be Fortune 450 or so - so I was used to working in large organizations. Some people hate big organizations. I love both. I love small because at the atomic level you can influence and shape things - that's why I work with startups. But I love the big because the bigs, if they can be retrained, you can make the elephant dance. And that can be quite powerful if you get them dancing in the right direction.
Going from nonprofit to for-profit was a big shift - thinking about margin, about manufacturing at scale in a global company. The Air Force was principally a US organization. All of a sudden you're running plants in Israel, China, Europe, and Russia - and multi-culturalism was a massive change. Mitre, by statute, was disallowed from manufacturing. Now I went to industry and industry manufactures. One of the benefits was I had all my knowledge of the R&D space from the national laboratories and I used it at Stanley Black and Decker - including to invent a COVID deactivation product.
Nine out of ten cars in your parking lot are held together by robotically inserted fasteners that Stanley Fastening Systems makes. We make a billion of them. Electronic doors, all of these kinds of things. We grew e-commerce to two and a half billion dollars. And I managed all the R&D across a heterogeneous business.
Chris Featherstone: I don't think people in software can fully grasp what it means to bring together electrical engineers, software engineers, embedded engineers, process engineers, and mechanical engineers and dovetail all those processes to release a physical product. And now with IoT and telemetry, the hardware is populating data up to the cloud for insights on predictive and preventative maintenance. The amount of data Stanley Black and Decker produces must be staggering.
Mark Maybury: We're absolutely in this transformation from a completely analog physical world to one where it's mixed reality - physical and digital integrated. I'm leading a congressionally directed study for the Defense Science Board focused on digital engineering, looking at both commercial and government sectors - aircraft manufacturers, car manufacturers, submarine builders - and what we find is that in leading-edge companies things are born digital. At Stanley Black and Decker, we digitally design all our tools and then 3D print them to look at ergonomics and share with channel partners like Lowe's and Home Depot to ask what their customers will like. The pros will try it and say "I love this one" or "that's terrible." Then we manufacture to that specification.
The factory of the future we were building aggregated all the different disciplines - industrial, tools, outdoor, security - into a vision for modular, programmable robots, both cage and free-range. You can take the digital designs for those products and pour them into the robots for cutting, drilling, screwing, labeling. Then use those same digital designs to redesign the factory so it's optimized, and use them for marketing - flying through the product visualization, showing it in operation. Digital is absolutely critical, and that's exactly why Ready Robotics excites me. They've developed Forge OS, a common operating system across a multiplicity of robots. Instead of training one person on one robot and another person on another robot, you have an intuitive interface that lets a single operator control not just one but many different robots, regardless of manufacturer.
Seth Earley: Let's go back to your early AI work at Cambridge. What was it like doing foundational AI research in that era?
Mark Maybury: I did my Master's in 1986 and 1987 in computer speech and language. I was awarded my doctorate in 1991 - I invented the first computer program to write multi-paragraph, multilingual texts. Essentially I created a digital author. Unlike today where AI is ubiquitous and everyone wants to work in it, when I graduated I was one of four Air Force officers who had any kind of degree in AI. It was a very select club.
At Cambridge we had the only hundred-thousand-word, speaker-independent, one-thousand-word vocabulary, speaker-dependent speech recognition system in the world at the time. The only comparable system was at ATR in Japan. I had a bank of workstations - a major investment. Then I went to the Air Force and they said "run our AI lab" and I had half a dozen Lisp machines, each worth roughly a million dollars in today's money, with object-oriented operating systems and knowledge-based simulators.
The key thing, though, was: how did you do it without modern machine learning frameworks?
Chris Featherstone: Was it more concatenative? Search patterns and things?
Mark Maybury: We did have neural networks conceptually - they go back to the fifties, Minsky and perceptrons. But there were no AWS instances, no GPT-3, no machine learning systems to plug into. So you become the learning system. And what's interesting is that methodology was actually more challenging in some ways, and richer in others. When I was an undergraduate I worked with neuropsychologists and we built an expert system to diagnose neuropsychological conditions. I asked the doctors: "How do you do this?" And they described having a Christmas tree visual model of the brain where all these reports light up - some bright, some dim - and then you say, "This person has an occipital lobe problem." So I built a Bayesian network. I looked at the probabilities associated with each frame element and created essentially a statistical visual model without high-tech visualization.
At Cambridge, when they said "no one has ever written a computer program that can author text - what would that look like?", I went to English professors and asked: "What textbooks do you use? How do you teach freshmen and graduates to write?" They told me the four major forms: description, comparison, exposition, and argument. I tied those forms to knowledge structures. I also added multiple levels of representation that GPT-3 still can't replicate in the same way - syntax, semantics, and critically pragmatics: the intent. Why is somebody writing that? Are they trying to convince you? Educate you? Deceive you? I represented what rhetoricians call communicative acts - the actions we perform with words - and I used AI planning technology to capture all of that in a computational paradigm. I could go to a knowledge-based battle simulator in the Air Force, ask what happened in the battle, and write a story about it - or an argument about whether it was fought well or poorly.
Today I would do it totally differently. I'd take GPT-3 and throw every text in the world at it. But what I could add that GPT-3 couldn't is those multiple structured levels of representation - and a genuine understanding of what the author is trying to accomplish.
Chris Featherstone: If you could canonize that into a product, you could score every article for its communicative intent - education score, confusion score, convincing score. Especially now with adversarial AI and re-fabricated speech and video.
Mark Maybury: That's a great thought. As we get more sophisticated with adversarial AI and deepfakes, understanding intent becomes critically important for information security. And interestingly, companies on Wall Street already do sentiment analytics of investor discourse - what do people think about Boeing stock, what do they think about the Department of Defense? I was very fortunate mid-career at Mitre - the government hired me to run a program with a bunch of universities and commercial companies to create what we called the Opinion Bank. It was a collection of annotated data regarding sentiment. That was the foundation for the whole field now known as sentiment analytics - and this was well before hate speech detection and election influence campaigns became major concerns. We recognized early that you'd need to detect sentiment and understand what populations were unhappy about. When you mix topic detection with sentiment analysis, and then get to intent - can you correlate this with authoritative factual sources? If you can't, well, as we've seen in many legal cases, a judge looks at something and asks: "What's the basis for this? Where's the evidence?" The legal community's argumentation language maps directly onto the argumentation strategies I codified in my PhD work.
I also teach my children - and have forced them at some point in junior high or high school to write an analytical essay using what the rhetoricians taught: ethos, logos, and pathos. Logos - make sure you have a logical argument grounded in fact. Ethos - make sure your argument is ethically based, that your purpose is not simply to empty someone's pockets or accumulate power. Pathos - we know humans are influenced by passion. You can watch an advertisement with a beautiful grandmother-granddaughter interaction and not learn a single thing about the product being advertised. All three have to be in balance if you're going to be a good leader and a good communicator.
Seth Earley: What else is on your agenda moving forward?
Mark Maybury: I love diversity of experience and challenge. I'm serving as a Special Government Employee for the Defense Science Board on a national study focused on digital engineering - both commercial and government sectors. And I just completed, with General Mark Russell, a classified study on ensuring America remains safe from particular aerial threats. We proposed what we're calling SAGE II - a modern version of the Semi Automated Ground Environment that Mitre was originally created to build back in the 1950s.
I also have a video in production with a Hollywood producer, focused on getting high school and college students excited about potential careers in artificial intelligence in the public interest - federal service, the Department of Defense, the national labs. I'm showcasing applications like the automated F-16 collision avoidance system that has already saved ten pilots from crashing, and systems that improve situational awareness for commanders in the field.
Seth Earley: Tell us more about the COVID deactivation project.
Mark Maybury: At the very beginning of January 2020, we knew we had a problem because our Chinese plants were shutting down. My CEO called me with a call he'd had from a former colleague who had an idea about deactivating COVID. My first reaction was "that's crazy, that's impossible." But I started digging into the literature and discovered that while COVID is not a living organism, you can deactivate it with heat - but it took about thirty minutes. A colleague of mine, the CEO of Patel, told me they'd invented a vaporization process using hydrogen peroxide that could decontaminate masks, but it required large containers.
What we discovered is that COVID-19 - SARS-CoV-2 - is electrostatically charged. So we invented a protocol combining electrostatic attraction and heat in a sinuous design. We tested it at MIT Lincoln Laboratory in a biosafety level 2 facility in the middle of COVID, working remotely and at night because it was a special project. We basically invented a mask and a room system that could deactivate SARS-CoV-2 essentially immediately - within seconds. Then along came the vaccines and the market dissipated - which is a great lesson in commercialization. But we have the capability, and if we need it in the future, we can manufacture it.
Seth Earley: It has been a real pleasure. Thank you so much for your time today, Mark. Truly enjoyed it.
Mark Maybury: My pleasure, Seth and Chris. Best to both of you. Wish all your listeners to be well and be safe. I think folks will be blown away by two things: first, the scheduling. But more importantly, the passion for life and for the things that drive you - that was palpable today. Thank you.
Chris Featherstone: We'll have to have you to our next Knowledge Salon. Hope to do another one before the end of the year.
Mark Maybury: Happy to do so. Wonderful to see you both.