“If Walt Black wanted to join the Air Force, he should have done so. Now he operates his private air force from Langley.” Langley, Virginia was the headquarters of the CIA. It was about a one hour drive from Fort Meade.
The General took a deep breath and recomposed himself. “Alright, I’m ready. Let me show you the Accelerator Line of what we call the, ‘Underground Railroad.’” Yesterday, Shields had read Lin into the special access program that contained details about the clandestine NSA tunnel complex.
“It’s a lot faster than driving. In the basement of this building, there’s a mile and a half tunnel. It runs under Highway 32, straight to Defense Innovations.”
“That’s incredible! No wonder you get there so quickly,” said Lin.
They left his office and walked towards DIRNSA’s private elevator.
The NSA was a component of the US Department of Defense—DoD. In 1952, President Truman initiated the NSA, in response to the escalating Cold War. After 9/11, the NSA became the most powerful spy agency in the world. It employed over 40,000 military and civilian workers. Also, an even larger cadre of government contractors worked to support the NSA.
General Shields could have chosen to house Defense Innovations Accelerator at NSA headquarters, within the confines of Fort Meade. It wasn’t like the Fort lacked space.
Fort George G. Meade was a US Army base spread over eight square miles. The 660-acre campus of the NSA sat on the western side of the base. Located 25 miles northeast of Washington DC, it was a gargantuan complex. The Fort also housed US Cyber Command—CYBERCOM—headquarters and the Defense Information Systems Agency. DISA acted like the DoD’s phone company and Internet service provider.
Not including the Underground Railroad, the Fort contained more than 32 miles of roads, its own fire department, post office, police force, and a SWAT team. Before CYBERCOM and DISA arrived, the NSA had two golf courses. Fort Meade housed more than 50 offices buildings totaling over seven million square feet. And that was just the Fort. The NSA maintained facilities all over the world.
Shields didn’t want to house the Accelerator on base. He wanted to keep the government-types away from the entrepreneurs. He knew that many govies would pour energy into throwing rocks at unproven technologies.
In addition, it took time to get into and out of the Fort. Traffic jams during rush hour were standard. Also, if the General wanted to house the Accelerator within one of the buildings at Headquarters, no cell phones were allowed. General Shields knew that mobile phones were the lifeblood of entrepreneurial companies trying to quickly ramp up revenue.
Shields pushed the basement floor button.
*
General Shields and Lin hopped into the back seat of a golf-cart sized, hovercraft. From their position in the hovercraft, tunnels extended in many different directions. A heavily armed NSA police officer drove the vehicle. They sped towards the Accelerator. “I love this thing,” squealed Lin in delight. “I want one!”
Shields laughed. “They’re fun. If the Fort still had golf courses, they’d be perfect. They glide right over water and don’t leave any tracks on the greens.”
“How far does the Underground Railroad go? These tunnels are huge. You could drive two semi-trucks in either direction, and still not hit pedestrians.”
“That, I can’t tell you, Lin. Your clearance only covers the Accelerator tunnel. But I could use the word, far.”
“General, before this meeting, I was hoping you could tell me more about how the Accelerator works? It’s still a little fuzzy to me.”
“Sure,” he replied, as they passed another guard station in the tunnel.
“Defense Innovations Accelerator is a combination of a VC firm and a startup accelerator. VC firms provide money—capital—to promising early stage, high-growth potential companies. Lots of famous companies you’ve heard of—Facebook, Twitter, Tesla Motors, Uber—were started with VC capital. I wanted the NSA to have its own VC firm, like the CIA. They have In-Q-Tel. I desired a VC that the NSA Director could more easily influence.
“Now the entire IC has benefited from In-Q-Tel. For example, they were an early investor in Palantir Technologies. We’ve used their software to track terrorist money laundering.”
“Don’t you have DARPA?”
DARPA stood for the Defense Advanced Research Projects Agency. DARPA was initiated after the shock of the Russian’s launch of Sputnik in 1957. DARPA used to be an agency that looked 10 to 20 years ahead of the curve. Now, their timeframe was much more compact. DARPA’s programs helped create the Internet, stealth aircraft, the global positioning systems (GPS), and Apple’s Siri.
“Not only do we have DARPA, but we also have Intelligence Advanced Research Projects Agency—IARPA. And the NSA has its own research and engineering labs. They all have their place, but they can’t match the passion, creativity, and innovation of startups.
“I’m convinced that America’s greatest strength is its innovators. God wired entrepreneurs differently. Most government leadership plays not to lose. Entrepreneurs must play to win—or they starve.”
“Ok, so I understand the VC aspect. Why did you combine the VC with an accelerator? What exactly does a startup accelerator do?”
“The fundamental idea is to find the best companies for your accelerator, provide them with a curriculum, mentors, office space, and technology to help them to succeed quickly. Accelerators give their portfolio companies a template—a recipe—for success. And they facilitate access to business leaders that have gone through similar startup challenges.
“A typical startup accelerator runs two classes every year. The classes last for about four months. During that four month period, business mentors provide intensive instruction—like a mini MBA course—on all aspects of startups; from sales, to product development, to finance.
“The selection process is intense and competitive. The accelerator typically receives a small portion of equity in exchange for a small amount of cash, usually around $25,000 to $50,000. After the four months, the startups graduate from the accelerator.
“Now, I wanted Defense Innovations Accelerator to operate a bit differently than that standard model. We conduct an eight-week introductory course for 12 competitively selected startups. After the introductory course, the 12 startups go through a down-selection process.
“The down-selected startups receive funding from the VC. The startups remain a part of our Accelerator for at least one year. If needed, I can offer to extend this period. I can also negotiate with the startups to provide additional funding. They almost always need more money.
“In our first Accelerator class, we down-selected five companies. This includes Gamification Systems, who you met yesterday; and CyberAI, who you’ll meet shortly. A strength of our Accelerator is that we help our portfolio companies with the intricacies of government contracts. The NSA also works to expedite clearances for each startup. Clearances are a huge benefit. TS/SCI clearances are difficult to acquire. And, our Defense Innovations’ office complex contains two SCIFs.”
SCIF—pronounced, ‘skiff’—stood for Sensitive Compartmented Information Facilities. The SCIFs were secure rooms on the second and sixth floors of the building. Both SCIFs allowed for TS/SCI communications.
Lin asked, “So, who are the other companies in the Accelerator?”
“In addition to Gamification Systems and CyberAI; Flashcharge and Prosthetic Thought went through the introductory course. Flashcharge markets a sophisticated, wireless charging system. Prosthetic Thought makes neural VR controllers. The idea is to use thoughts to control video game movement.
“Defense Innovations invested in another company, Swarmbot Corporation. They didn’t go through the introductory course. I worked with the Co-Founders of Swarmbot before the NSA, when I was Commander of the 24th Air Force. So these five companies—Gamification, CyberAI, Flashcharge, Prosthetic Thought, and Swarmbot are the current portfolio companies of the Accelerator. I’m confident that they’ll be able
to sell technology to the NSA in the next year.”
The 24th Air Force was located in San Antonio, Texas. The 24th provided defensive and offensive cyberwar capabilities to the Air Force. In San Antonio, General Shields got very familiar with the NSA because the 24th contributed the Airman for CYBERCOM.
CYBERCOM was led by the NSA, but comprised of servicemen and women from all the armed forces. In other words, CYBERCOM was composed of cyber-warriors from all four services: Air Force, Army, Marines, and Navy.
“You’re going like the kid who founded CyberAI, Josh Adler. He’s a genius with AI, and he comes from good stock. His dad is the billionaire hedge fund manager, Jared Adler.”
Chapter 6 – Josh Adler
4:50 p.m. (EDT), Monday, July 27, 2020 - Columbia, MD
Suite 602, Conference Room, Defense Innovations Accelerator
Josh Adler, the Founder and CEO of CyberAI Defense, Inc., shot up from the conference room table. Josh removed his navy blue, Hugo Boss blazer and draped it over the chair. He wore a blue dress shirt with no tie.
Then he noticed his armpits. Colossal sweat stains transformed his shirt into a blue Rorschach test. The CEO recognized stress in the Rorschach blot. He quickly reapplied his coat. Josh took a series of deep breaths, attempting to reduce his heart rate. He didn’t bother to bring a laptop. All the flat screens were black. He had no presentation. He had no demo. This meeting was going to disappoint General Shields.
Josh was an athletic looking, five foot ten, 23-year-old. He had an olive complexion, with curly brown hair and hazel eyes. On any day except today, Adler exuded a quiet confidence that naturally attracted followers. His enthusiasm for AI was as contagious, as his dimpled smile. Josh won the gene-pool lottery. He inherited his mother’s good looks and his father’s analytical mind. The stereotypical bookish AI expert looked nothing like Josh.
In order to start CyberAI, Josh dropped out of MIT during his junior year. In response, Jared Adler broke all contact with his son. Josh’s greatest regret about quitting school was leaving the varsity MIT Crew team.
The CEO recognized opportunity when he saw it. He knew that automating cybersecurity with AI would lessen the need for human cybersecurity administrators. This would save money, while at the same time improving an enterprise’s security posture.
With the growth of connected devices—phones, tablets, wearables, home automation products, sensors, and the like—hooking up to the ‘Internet of Things,’ the need for security administrators was skyrocketing. There was so much more Internet traffic to watch. Reducing companies’ need for human cybersecurity engineers was a game changer.
Josh also felt that AI applied to cyber was just the beginning. His grand strategy was to use narrow AI for cybersecurity as a beachhead. In time, he could expand to a stronger form of AI. With this more general purpose AI, Josh could pivot to other markets, especially discovery.
The genesis of CyberAI Defense was a class assignment for a machine learning course at MIT. Some Fortune 1000 companies and government agencies gave MIT access to their large datasets. The goal was to see if the students could derive innovative insights from the data.
Josh choose to work with three petabytes of cybersecurity data supplied by the DHS and FBI. The data consisted of text-based, log files from hacked systems. Josh tuned his machine learning algorithms to perform natural language processing—NLP—on files from servers, hosts, intrusion detection systems, e-mail scanners, malware filters, and other enterprise infrastructure. In this manner, he trained the computer to reliably recognize patterns left by cyber-intruders.
Josh astutely perceived that he could use his algorithm to perform a job that took many human beings. Securing enterprise infrastructure was a labor intensive task that caused burgeoning payrolls for IT departments. And good security administrators were in short supply. They commanded high salaries. Josh felt if he could improve enterprise cybersecurity by 10 times, or 10X, CyberAI would be an incredibly valuable company.
Within the discipline of computer science, AI had a long history. First coined as a term in 1956 by Dartmouth professor John McCarthy, the goal of AI was to get computers to think like human beings.
To accomplish this task, over the years, AI developed a number of sub-disciplines that touched upon what it meant to be human. How does a computer learn? How does a computer process language? How does a computer see?
Due to the low processing power of computers, progress in these areas was painstakingly slow. A noteworthy milestone occurred in 1997, when the Deep Blue chess computer made by IBM, beat world champion chess player, Garry Kasparov.
After that event, many observers stated that it was merely an example of weak AI. Weak AI taught a computer a given task, like chess. Naysayers rightly said that there was a significant gap between playing chess and truly thinking like a human being. Strong AI was defined as human level thought.
The year 2010 was a watershed in AI. IBM’s Watson supercomputer beat a group of former champions on the TV game show, Jeopardy. This feat was more impressive than winning a game of chess. Watson had to listen to Alex Trebek’s questions and then process the language to find the answer. Watson married speech recognition with NLP, performed on vast libraries of information, like Wikipedia.
At the age of 13, Josh watched the Jeopardy contest. Watson completely mesmerized him. From that day forward, Josh wanted to make machines think. Josh envisioned AI algorithms that changed the world.
Algorithms dictated the sequential steps computers followed to perform a task. Step-by-step, the algorithm told the computer what to do. Algorithms worked no differently than a recipe to bake a cake. Line-by-line, the computer was told what step to execute next.
In a little over ten years, AI had progressed from beating the best player chess player in the world, to beating the best Jeopardy players in the world. AI was moving closer and closer towards strong AI—towards thinking like a human.
But, many pundits remained skeptical as to whether computers would ever match the intelligence of human beings. They said there’d never be a genuinely strong AI. In these prognosticator’s eyes, computer and human intelligence would never achieve parity.
Josh marveled at these doubters. AI was deployed so widely that if it disappeared, the world economy would cease to function. Without AI, no one could use a search engine, trade stocks online, or ask Apple’s Siri a question. Planes would fall out of the sky. War would be fought as it was during World War II.
While it was true that AI did not match the intelligence of human beings in 2020, Josh reasoned that technology was progressing exponentially. Even if AI technology progressed more linearly, Josh thought strong AI would occur well within his lifetime.
Dystopian predictions for AI were at the other end of the spectrum. With this mindset, once strong AI happened, the next logical step was a Terminator-like, apocalyptic world. Computers and robots would enslave humanity to work for the good of the machines, plugged into the Matrix. Josh felt that while strong AI had downside risks, the more logical progression was that AI would augment human capability.
For example, think how much smarter a search engine on a smartphone made everyone. In fact, were Josh made king for a day, he would change the name of AI to extended intelligence. Josh viewed this scenario as the most likely. He wasn’t naïve. AI would drastically hurt some forms of employment like taxi cab drivers, delivery men, and long-haul truckers. The pain would be real. And he did worry about the consequences to the economy. But, he felt that humanity would benefit from AI offerings that collectively improved the human experience. Josh expected the Bionic Man instead of Skynet. At least that was his hope.
Machine learning and NLP were the fields within AI that intrigued Josh the most.
Machine learning algorithms statistically sliced and diced data by classifying, clustering, and making recommendations. Movie recommendation engines and online shopping carts that suggested what you might like, based on past purchases, often used machine learning
algorithms.
NLP attempted to understand text—characters, words, sentences, paragraphs, and the like. The lines often blurred between these fields. For example, CyberAI employed machine learning algorithms to improve the NLP capabilities of Josh’s AI.
In the early 2010’s some trends mingled to rapidly increase the effectiveness of AI. There was an explosion of large data sets created by the likes Facebook, Twitter, and Instagram. This was dubbed, ‘Big Data,’ by the press. In the history of humanity, data of this size and variety had never been created so fast.
Larger datasets gave machine learning more information from which to learn. Computers could learn to recognize cats, from the millions of cat photos people posted on social media. Also, as predicted by Moore’s Law, computer processing power kept increasing exponentially. And now, you could cheaply rent supercomputers over the Internet cloud from Amazon and Microsoft.
Chasm Waxing: A Startup, Cyber-Thriller Page 4