by Andrew Watts
“Please wear this around your neck at all times. You’ll need the chip to access your office and computer.”
Colt thanked them and then stood by a cluster of empty chairs in the waiting area, studying his surroundings. Wide floating stairs led to the four floors above. The sweet smell of baked goods and coffee filled the air. He heard what sounded like an espresso machine whirring from an internal café around the corner. Several employees were laid out on beanbag chairs and couches, wireless headphones plugging their ears as they typed away on stickered laptops.
“Colt McShane?” A tall man wearing khakis and a white button-down shirt called out as he walked down the stairs.
The man walked over and stuck out his hand.
“Gerry Nader, chief technology officer here at Pax AI,” the man said, giving Colt a million-dollar smile.
“Good to meet you,” Colt said.
“Come on, let me show you around.” He gestured for Colt to follow him. They walked around the corner and up to the café. “Would you like anything? One thing we pride ourselves on here is great coffee.”
A barista smiled from behind the counter. Both men ordered lattes and made small talk as they waited for their drinks, sharing personal history and business contacts. Playing the game of “oh, do you know so-and-so?” The typical first conversation Colt encountered when starting an investor evaluation.
They discussed Colt’s plans for the next few weeks regarding financial analysis, and Nader provided recommendations on how to spend his free time in the city. Then the lattes arrived, and Nader continued their walk around the first floor. Colt saw a man wearing a visitor badge with the words Special Agent preceding his name.
“Is he with the FBI?” Colt asked after they walked out of earshot.
Nader nodded somberly. “Yes. By now you must have heard about the death of one of our AI researchers.”
“I did. I’m so sorry. Any word on what happened?”
Nader shook his head. “Well, these crazy Trinity people have taken credit for it. You saw the extra security at the front desk. Our head of security is doing a complete evaluation and we’ll be getting much stronger on that front.”
Colt said, “How is everyone holding up?”
“It’s been a mess, to be quite honest. Security is the main issue. And the HR issues that go along with it. We’ve asked most of our rank-and-file scientists and engineers to move into temporary housing so we can provide better security. Essentially, we bought up a floor of suites at a luxury hotel in Palo Alto. A logistical nightmare with all the families. Some of our employees quit, citing safety concerns. Others didn’t want to deal with the hassle. The executives all have personal bodyguards staying with us now.”
“Oh my. I didn’t even think of that. So, you think what happened to Kozlov . . . something like that could happen again?”
Nader shrugged. “Personally, no, I don’t think so. I think it was probably some deranged lunatic. A tragedy certainly, but our head of security is working closely with the FBI and they tell us we are all very safe. Still, we must be cautious.”
“Must be hard on all the employees.”
“It will be. It’s an enormous amount of work . . . and it’s quite challenging for a company like ours to replace personnel at that talent level . . .” Nader’s face reddened as he realized that wasn’t what Colt meant. “But obviously everyone feels just awful. Kozlov was a very nice man. It’s just such a shame. These conspiracy theories have gotten completely out of hand. Anyway, you saw the FBI agent. They’re wrapping up the last of the interviews with us. Such a tragedy.”
They reached the wide stairway, which was right next to an elevator. Nader said, “Hope you don’t mind, I’m trying to get my steps in.” He held up his smartwatch and they walked up to the second level.
The entire second floor was a combination of open office desks and glass-walled huddle rooms. Colt looked into one of the meeting rooms as they passed. Engineers and scientists gathered around a long whiteboard. They were young, most in their twenties and thirties. One was pointing at the board and speaking as another scribbled down a complex formula.
“This is where the sausage gets made,” Nader said. “These are some of the smartest people on the planet.” He turned to Colt. “How familiar are you with our work?”
“I’ve read all of the pre-reading your firm prepared. Artificial intelligence is a fascinating field. But I’m sure everything I know would be miniscule compared to the people who work here.”
Nader said, “No problem. We’ve set up a crash course tour for you this week. And Jeff has said to give you whatever you need, so just say the word.”
“Thanks.”
Nader stopped outside one of the glass-walled offices. “Here’s you. You can put your bag down inside if you like.”
It was a simple office. A bamboo standing desk with a push-button control on the side. A computer and a few office supplies. Colt placed his leather travel bag on the desk.
“Now we’ll head up to the third floor and say hi to Jeff real quick.”
As they walked up another flight of stairs, Colt caught a glimpse of the fourth floor. The stairway led to a door with what looked like a fingerprint and eye scanner. An armed guard sat on a chair outside it, looking back at Colt. Beyond him, the glass walls were opaque.
“Looks like a higher level of security up there?”
Nader said, “That’s where our development team works. Only a handful of our one hundred and fifty employees are allowed up to the fourth floor. We also hold some of our lead team meetings there, when the meeting covers those programs.”
“I assume I’ll be able to check it out?”
Nader gave an awkward grin. “We’ll have to check with Jeff on that.”
So much for whatever you need.
The third floor had less open floor workspace, and the offices were larger. Some of them had transparent glass while others were opaque. Colt watched as a glass office wall transformed from opaque to clear as someone walked out the door.
Colt nodded toward it. “That’s a cool party trick.”
Nader said, “Yeah, all the glass on the third and fourth floors has a polarization mechanism so we can make meetings visually private. Seems high tech at first, but you get used to it.”
Kim’s corner office was transparent, and Colt saw the famed CEO kneeling on a yoga mat, his eyes closed, hands on his knees.
Nader said, “It looks like he’s meditating. He’s really into that. But I think he’ll be done soon.” Nader tapped the touch screen outside Jeff Kim’s office, swiping to check the CEO’s schedule. “Yeah, he’s only got two more minutes of this.”
They waited in silence, which Colt found weird. As the clock hit the top of the hour, Nader tapped on the glass door and Jeff Kim’s eyes opened. He gestured for them to enter and stood to greet Colt.
“Mr. Kim, thank you for having me,” Colt said.
Kim said, “Absolutely. Given that one of our senior researchers passed away unexpectedly, I understand the perceived risk among our investors. It is only prudent to have a trusted third party conduct a financial prospective analysis.”
Nader remained quiet while Kim spoke, assuming a deferential posture a few feet to the side. Colt had the distinct impression that Jeff Kim’s employees worshiped the ground he walked on.
Colt said, “I appreciate your honest perspective.”
“Of course.” Kim turned to his CTO. “You’ll show Mr. McShane around today?”
“We have a few days of presentations showing him everything about our company.”
“Including the fourth floor, I hope?” Colt said.
Nader shot him a look, but Colt didn’t blink. He was, as far as they were concerned, there on behalf of investors. It was expected that Colt would want to get his eyes on all aspects of Pax AI’s business, especially their most tightly kept secrets.
Kim gave a thin smile, his voice soft. “I’m sure we can figure something out. We’ll just
need to take security precautions.” He looked at Nader. “Let’s make sure Colt gets a tour of the fourth floor soon. He should have full visibility.”
“Will do, Jeff.”
A dull knock on the glass door behind them caused Colt to turn. Two men stood at the door, one holding a tablet computer with a stylus.
“Does Jeff have a moment?”
Nader said, “Guys, we’re in a meeting.”
Colt said, “I don’t mind.”
“Thank you.” Kim walked over to the door.
Nader said, “You will have to excuse my engineers. These two are working on our new language algorithm we just announced.”
The room went silent as Kim held the tablet, his eyes scanning the screen.
“What is this?” he asked, pointing at the tablet. Colt snuck a peek, but to him it just looked like random streams of text and numbers. “This isn’t right.”
One of the engineers said, “We double-checked it.”
“Then you got it wrong twice.” Kim walked over to a dry erase board near his desk. “Look, I’ll write it out.”
Colt watched as Kim began scribbling what looked like multi-line calculus equations on the board. He handed the tablet back to one of the engineers with his free hand while using the other to finish the math. “See? Take a look at the second page of code. Line 32. Then look at this. If you release it like that, you’ll still have grammar induction errors in the linguistic model.”
The two engineers looked at the board and then back at the tablet, understanding dawning on them.
One of them said, “Oh shit. He’s right. We missed that.” He looked up. “Sorry, Jeff, we’ll get it fixed.”
“Take a picture of this, please.” Kim gestured to his white board, and the engineer holding the tablet took a picture. Kim’s voice was patient as he said, “Please take your time and get it right.”
The two men nodded and walked out of the office. As they left, Colt heard the second engineer whisper, “How the hell did he catch that?”
After they left, Kim turned back to Colt. “Do you know why you are here?”
“Given the stage in your company’s development, and the amount of financing that could be injected into your balance sheet by the investors I represent, I am here to provide a risk assessment.”
Kim said, “That pretty much sums it up. And it’s part of our job to wine and dine you in hopes that it improves your evaluation.”
Colt gave a polite laugh.
“In all seriousness, if you need anything, let me know. This is very important to us.”
“Thank you,” said Colt.
Kim inclined his head to Nader, who said, “Thanks, Jeff.”
Nader and Colt departed, heading down the hall as Kim’s door slid shut. With his peripheral vision, Colt could see Kim standing behind the glass door.
He was still watching them when the glass walls of his office became opaque.
11
Colt followed Nader into a third-floor conference room, thinking about Kim’s ability to pick out flaws in the AI language algorithm just by glancing at his engineer’s tablet. He realized that some of these people were scary smart. That level of brainpower was off the charts. Even if Kim had something to hide, would a man like that make a mistake?
Nader gestured for Colt to sit down. “We’ll spend the morning in here. We have a few short presentations, followed by lunch at a nice place downtown, and then we’ll be back here in the afternoon to let you pore over financials.”
Colt smiled. “Sounds like fun.”
A short, bearded man entered the room carrying a laptop, a pen tucked above his left ear.
“Colt, this is Luke Pace.”
Pace nodded without saying hello, then sat down and began furiously typing on his computer.
Nader said, “Despite his manners, he’s an exceptionally talented chief scientist, and unfortunately he’s all we can afford at the moment.”
Pace snorted. After a few more seconds of typing, he slapped the laptop shut and made brief eye contact with Colt. “Sorry, I was just finishing something up. Lots to do.”
“No problem at all. A pleasure to meet you,” Colt said.
“Cold-pressed coffee?” Nader asked. “Cold latte? It’s the good stuff. La Colombe. We keep them chilled in this mini-fridge if you want any. For hot coffee, you’ll need to go back down the stairs.”
“I’m good.” Colt held up his still-full cup from downstairs. These people wanted for naught.
Nader gave Pace a look. “Ready when you are.”
“Oh, I’m presenting, aren’t I?” Pace said, re-opening his computer. He tapped a few keys and the projection screen on the conference table lit up.
A video began playing, the narrator’s voice a middle-aged woman with a British accent.
“What is learning?” said the narrator.
On the screen, a toddler walked on a flat lawn, blurred green trees and water in the background.
“What is intelligence?”
The video showed high school students at their desks, taking a test.
“What is our goal, when we train machines to learn and become more intelligent?”
Now the video showed Pax AI employees animatedly talking to each other in front of white boards or typing on their computers while sitting in beanbag chairs. The video cut to images of robot hands manipulating dinner plates.
“At Pax AI, we believe that machine learning needs to be done in a way that is safe and beneficial for all mankind.”
Now the robot hand was solving a Rubik’s Cube. A smiling researcher observed, the Pax AI logo emblazoned on his white lab coat.
“As our technology progresses, so does learning. What once took our species years to learn can now be completed in seconds. At Pax AI, we can run thousands of simulations simultaneously, teaching our machines faster than ever.”
Now the video showed four-legged robots hiking up rocky terrain, assisting miners in what looked like a third-world country. A graphic of a climbing stock chart overlayed the screen.
“Using reinforcement learning, our AI programs develop increasingly complex strategies for solving problems. Using multiagent game play to allow our bots to interact with each other, we’ve been able to improve our AI system’s ability to learn. Reinforcement learning is the same way that human beings have become generally intelligent. But that took countless generations and a lot of time. Computers are not bound by the same speed limitations. A generation of learning, in a simulation, can take a nanosecond.”
Colt watched as video game competitions pitting human teams against Pax AI teams flashed on the screen. Then a separate video showed red and blue bot-on-bot simulations.
The video went on for another ten minutes, showing the separate projects the company was working on, AI programs that could learn to anticipate and fulfill human requests. Ordering groceries when running low or scheduling a meeting after being introduced to a new client.
“What is that?” Colt asked.
The last part of the video showed a woman putting on headgear as scientists helped her.
Pace paused the video. “One of our projects has a cranial-mounted device that enables a neural interface.”
Colt looked at Nader, who translated. “It’s a type of headgear that allows primate animals and human beings to control machinery without physical inputs.”
Colt snorted. “You’re kidding, right?”
Pace said, “I’m surprised you haven’t read about it. Our competitors are working on similar devices.”
He unpaused the video, and the narrator began speaking again.
“With our market-leading prediction algorithms, Pax AI’s goal is to safely introduce our AI systems to an increasingly complex set of environments and—”
Nader smirked at Colt. “Sorry, I think our engineers are writing this copy.”
“With increased computing power, AI-augmented code development, and breakthroughs in AI techniques, we hope to bring about the first
real artificial general intelligence to the world, improving lives through science.”
Nader and Pace both looked at Colt when the video was over, a sense of pride in their eyes.
“It’s very impressive,” said Colt. “And we can really do all those things?”
Pace nodded. “Oh yes.”
“What was that simulation with the red and blue bots? In that video game thing?” Colt asked.
Pace’s eyes lit up. “That’s one of my favorites. We trained two different AI teams to play children’s games in 3D environments. Games like tag, capture the flag, hide-and-seek. So, the AI teams are each comprised of separate agents.”
“Agents?” Colt asked.
Nader said, “Think of each agent like its own unique machine.”
Pace nodded. “So the AI agents, or teams, work together, communicating and learning as we make the games increasingly more complex.”
Pace typed on his computer and the display screen changed to a YouTube video with Pax AI branding. It showed little red and blue human-shaped characters moving around blocks and walls, tagging each other and capturing digital flags.
“The agents on the same team actually talk to each other in their own language. We ran thousands of simulations, and over time, they get better and better at overcoming the opposing team. It’s incredible what they’ve learned to do.”
The video showed one team moving the blocks around to form barriers, while the other team used blocks to break through one of the barriers and capture the opposition’s flag.
Pace said, “Each of the agents has its own combination of neural networks. Some are for vision sensing. Others are for . . . well, thinking and learning. You fast-forward these thousands of simulations and the bots learn new strategies and tactics. Each team overcomes the other and introduces new obstacles until there is no more progress.”
Colt said, “What does that mean, no more progress?”
Nader said, “Until one team of bots has a strategic advantage over the other.”
Colt said, “Sounds like business. Or war . . .”