I was thirteen, sitting in front of my family computer with minutes
left in my turn online, when I stumbled across an article
titled “Could a Machine Think?” It was the summer of 1990. A
few weeks later, the president of the United States would declare
the ’90s the Decade of the Brain, a decade that would see the
emergence of the internet and the world wide web as well as a
new generation of neuroscientists and computer scientists who
were going to change the world. Th e article mentioned a growing
research area: artifi cial neural networks, or computer programs
that imitated how our brain learns. Th e networks could automatically
learn from examples by modifying their synaptic weights.
I was mystified. How could they learn automatically? Self-learning?
Was this fi nal proof of Cartesian dualism, where the mind
arose from a magical realm outside our physical universe? It
seemed like sorcery.
My mom called from downstairs. “Alex, it’s your brother’s
turn!” I was allowed only two hours each day, to limit screen time
and get us outdoors. It was never enough. A revolution was in
progress. Telephone networks had given rise to something called
the internet. Suddenly, anyone with moderate technical prowess
could travel the digital world and access anything. Some things
that were meant to be accessed. Some that were not.
Th e walls had come down. It was the cyber version of the Wild
West. Th e new frontier was open. My parents couldn’t aff ord subscriptions
to paid content, but it didn’t matter. My brother, sister,
and I were able to access anything we wanted. Books, articles, schematics.
Searching, exploring, discovering, liberating content. Suddenly,
teenagers had power. We were able to do things our parents
couldn’t understand. Download technical manuals and build things
like blue boxes that could make free phone calls (illegally, of course).
Communicate online with other hackers, sharing tips, tricks, and
failures. A global community was born. For a socially awkward
teenager, it was paradise.
Th e computer fan started to vibrate, like it was about to overheat
and shut down. It was a sweltering summer day in my parents’
basement in Portland, Oregon. Without air-conditioning, the humidity
had fused my shirt to my skin. I barely noticed. I was plugged
in to the family computer in a race against time.
But my time was up. My younger brother approached with a
cruel grin, recognizing my anguish and relishing the moment as
he took control of the family computer. I knew the drill: if I resisted,
he would pull the power cord, causing me to lose all my work.
I was banished, cut off from the cyber world, until tomorrow. An
As I left the room, my mind returned to the article about thinking
machines. A seed had been planted. I headed for Portland’s
iconic bookstore, Powell’s, and picked up an introductory textbook
on neural networks. Th ey seemed magical, and I needed to intuitively
understand how they worked. I sat down and manually calculated
the fl ow of data through a neural network, how it would
learn from a mistake, calculating an error signal and updating the
artifi cial synaptic weights all the way back to the input layer of the
neural network (a process called back propagation). Th e numbers
all checked out. Magic.
I had been intrigued by the mind and brain for years, liberating
philosophy books from my grandparents’ bookshelves when I visited
them in New York. William James and the stream of consciousness.
David Hume and the rational basis of behavior. Oliver Sacks
and localized neurological disorders. Early attempts to understand
understanding. Understanding ourselves as a way to understand
the universe and its purpose, if there was one.
I embarked on an obsessive journey into the world of neural
networks. Or artifi cial neural networks, as they were called back
then. When I enrolled in the University of Oregon’s Clark Honors
College, I found every course a lens into the mind and brain. I double-
majored in computer science and mathematics, with an unrecognized
third major in neuroscience. Several professors observed
my obsession and invited me to do research at their laboratories. In
a biophysics lab, we conducted patch clamp recording to explore
how the fruit fl y (drosophila) nervous system worked. In a neurobiology
lab, we explored how tiny neural networks (with a few hundred
neurons) gave rise to complex behaviors. In a psychology lab,
we explored inhibitory mechanisms involved in memory and cognition.
In the summers, I returned to Portland to do research at a
private institute focused on biomedical diagnostics.
At graduation, my advisors were surprised to hear that I was
taking a leave of absence from academics to enter private industry.
I had huge respect for the academic world and the rigor of peer-reviewed
publishing and scientifi c debate. I had published fi ve
peer-reviewed publications in computational diagnostics, a fi eld
that used computing techniques to detect the early onset of biomedical
disorders, such as glaucoma so that doctors could inter4
vene before blindness set in. But I felt I needed to leave to give myself
time to build, to manifest, to work on solutions out in the messy
real world, tackling all the challenges that came with it.
I fi gured I would return to academics at some point. I decided
to take a day job, so I could pay off college debt and work on secret
technology projects at night. I moved to San Diego and became a
soft ware engineer doing fi nancial analytics at Teradata, a pioneer in
massively parallel processing (MPP) technology, working on the
largest databases in the world, from those in three-letter agencies to
those in large banks. Data is fuel that helped spark the rise of AI.
Th e world of tech entrepreneurship was intoxicating. I had
been inspired by the stories of mainstream leaders like Steve Jobs
and Bill Gates as well as lesser known ones like the Monk and the
Riddle. And combining tech entrepreneurship and neural networks
was the most addictive drug I had ever encountered. You could create
technology and put it out in the real world, with no layers in
At this point in the early 2000s, we were in the middle of a major
recession and an “AI winter.” An AI winter is a period when the
funding in artifi cial intelligence dries up, with university researchers
losing grants, companies going out of business, and the general
public losing interest. It’s like a recession that impacts one specifi c
area. I didn’t let any of that distract me and never lost faith in the
immense power of neural networks and AI.
Eve rything Is Connected
I would occasionally escape the world of technology to explore
nature on camping trips or during underwater diving adventures. It
was on one of these trips that I met Paul Rahilly. His background
was mechanical engineering and equipment reliability, and we talked
about all the problems big energy companies were having with
equipment breakdowns. I had never planned to work in the industrial
realm, but I was drawn by the gravity of the problems he described.
People were getting killed on a regular basis, along with the
environmental catastrophes we saw in the news.
He and I founded a company called Mtell, which applied neural
networks to industrial diagnostics, monitoring heavy machinery
in manufacturing and process industries. Our technology was
similar to the precogs in the movie Minority Report—we predicted
events (equipment failure) and then prevented them from happening
with intelligent interventions (the maintenance equivalent of
Tom Cruise, Chief of PreCrime).
We launched the company in the middle of an AI winter, and
investors thought we were crazy. Industrial soft ware wasn’t sexy. AI
and neural networks were bad words. Apparently, combining two
unpopular categories created something that was even more unsexy
But we were convinced we were doing something that would
last—and, more than that, impact the world. Our mission statement
was “creating a world that doesn’t break down”—a mission
that most industrial companies laughed at, until the Gulf oil spill of
2010 involving the Deepwater Horizon off shore rig. Eleven lives
lost, many of the bodies never found, bereaved families left to
search for closure. On the fi nancial front, there was $45 billion in
fi nancial losses for BP. We might never know the extent of the damage
to the Gulf underwater ecosystem. It was a wakeup call.
At Mtell headquarters, we had a somber gathering that day,
asking if we could have prevented the catastrophe. It was a watershed
moment for us. We decided to focus on upstream drilling
and see if we could make a difference. We knew from the
inside that these companies were using old-school maintenance
techniques and ignoring the deluge of sensor data that was now
available for their equipment. It was too much data for their operators
and maintenance techs (who were already overworked
and overwhelmed) to analyze. We figured our neural network
technology could help.
One of the biggest challenges in these environments is that everything
is connected. If one part fails, it can bring down a system
that could cost millions of dollars per day in lost work, fi nes, and
cleanup, not to mention loss of life and environmental damage.
What Mtell developed was a system that could predict failures in
these complex facilities up to six months before they would fall
apart and isolate exactly what would go wrong and where. So, you
could replace that part that would only cost fi ve to six fi gures, and
the business would be back to producing product in a matter of
days. Without this solution, a part could fail, no one would know
about it, and there’d be leaking oil or fl ammable gas, leading to all
kinds of environmental damage and loss of human lives.
Succes s Is About People
Like all start-ups experience, life became an unpredictable roller
coaster. On a typical week, I would ricochet from worrying about
mounting debt, to celebrating good news from a major customer, to
panicking about a press release by a gargantuan competitor. Th ere
were many days when our eff orts seemed futile, but we took a deep
breath and kept going.
Our journey culminated in Mtell being acquired by AspenTech
(NASDAQ: AZPN), a publicly traded company that spun out of
MIT in the 1980s and had grown into the world leader in process
optimization. With AspenTech’s depth in process and our depth in
machines used in the process, combining the two approaches was
an exciting prospect. We sold for approximately $40 million—two
years aft er raising $1 million in venture capital.
Th e decision to sell is never an easy one. We were starting to
gain traction with our soft ware and could have raised another
round of venture capital, taking more time to grow the company
independently. But having spent a decade in the industrial sector, I
was ready to embark on the next chapter. I stayed at AspenTech for
two years aft er the acquisition to integrate the companies, watching
their stock price double to an $8 billion market cap. It was amazing
to hear Wall Street analysts give a stamp of approval to AspenTech’s
AI strategy, with Barron’s highlighting AspenTech as one of the top
three AI investments outside of FAANG (Facebook, Apple, Amazon,
Netfl ix, Google) in an article published in June 2018.
Our success was ultimately about people. Everyone on our
team understood that our clients worked in extreme environments,
going to work every day not knowing if they were going to
make it home. We made an eff ort to fl y every new hire out to a
drilling rig to see, hear, feel what those extreme environments
were like. To feel a top drive on a rig start to vibrate out of control
and get a pit of fear in your stomach like during the worst turbulence
you’ve ever experienced on a plane. And to observe the intelligence
and creativity of the operators, some of whom had only
a high school education but knew how their rig worked better
than PhD chemical engineers.
Aft er a decade of my working with hundreds of customers on
AI adoption, a key insight emerged: that humans’ role was essential.
Humans are the unsung heroes of the AI revolution, unrecognized,
both in terms of media coverage and in terms of receiving research
and development . I saw fi rsthand the breakthroughs, the spark of
something great, when people started working with AI as a team
and when AI was able to incorporate the unique intuition and insights
of the human technicians who brought decades of experience
with a facility and its equipment and had developed instinctive
When people saw the AI as an ally and not as a foe, they began
to use it to have their own hero moments, leading to promotions
and discussing with pride the AI agents that they had created using
our soft ware. Th ere was a metals and mining site that avoided environmental
damage and associated $10 million environmental fi nes.
A rail company that avoided engine failures and track interruptions,
worth hundreds of millions of dollars of savings. Our clients
were saving lives and creating heroes at their facilities.
AI was back in a big way. Th e world had gone from an AI winter
to a global AI arms race. Th ere were some amazing breakthroughs
in AI in the last decade, including the deep-learning revolution.
Th ere were also tech luminaries talking the threat of
artifi cial superintelligence and the need for regulation. But with all
the recent advances and fears, the human element was lost in the
shuffl e. Th e unique and essential role that humans played in the actual
AI projects being used around the world. Real-world, operationalized,
applied AI deployments are messy. Unlike idealized
gaming environments, we live in world of complexity, of nonlinear
interactions that make things like prediction and causal analysis
diffi cult. Yet this was just laying the groundwork for the next phase: