Ben Mattinson
The allegedly boring but important legal stuff about Gold Rush, one of the escape rooms developed by The Escape Game San Francisco, is the “very low” but “technically not zero percent” chance of getting incinerated, electrocuted, squished, or poisoned. That’s according to the waiver I have to sign before getting locked in a room with my friend Ben for one hour, which doesn’t strike me as the worst thing to be doing in the Financial District on a sunny Sunday afternoon.
“So,” I say to Ben at the check-in counter, “how do you feel about our very low but technically not zero percent chance of getting incinerated, electrocuted, squished, or poisoned?”
He snickers.
“Uh, good. I guess.”
The premise of the Gold Rush room is this: Clyde Hamilton (not to be confused with the Fourth Circuit federal judge) is our relative who got rich during the nineteenth-century gold rush in California. The riches he acquired are hidden somewhere in his log cabin and it can all be ours—if we can crack the brainy puzzles, obviously. It’s the fifth escape room I will have ever done but I have no idea what number this will be for Ben, who has not only been a dedicated fan of these high-pressure spatial puzzles for more than a decade but has for a while even been a developer of one. He knows a lot about what it takes to escape and he is good at escaping.
This explains his exuberance when, fifty nine minutes later, we escape the room at the very last moment. It’s rarely this tight when Ben plays, primarily because he does escape rooms with a group of friends, whereas today it’s just the two of us. All the more impressive, the Game Master tells us when she greets us at the door. Even big groups apparently ask for a lot of help in this escape room.
“What did you think?” I ask.
“Oh, this was a good one,” Ben says. “Well-crafted, immersive, and varied puzzles. Not impossible to solve but also not easy. I would say definitely among the top of the ones I’ve played.”
Ben’s interest in escape rooms began in 2014. He was a college student at MIT and was helping produce Next Haunt, a student-run haunted-house escape room that opens every year for Halloween. Handy by nature, Ben had already been involved in the builder community on campus, but the concept of an escape room was new, and he therefore needed to first do field research. The summer before his senior year, while he was interning in the Bay Area, he got to experience his first one: in Oakland, at one of the rooms developed by the company Omescape.
“Prior to escape rooms, puzzles were not something I did,” he notes. “The singular, one-off puzzles never appealed to me; what I was always more interested in was the intersection of world-building, storytelling, and immersive experiences. I usually got that through video games or fantasy books, but escape rooms unlocked something new for me, I think. They showed me that puzzles can exist in that intersection and be just as immersive and thematic.”
I suggested we do an escape room for Ben’s interview because, in a way, it’s a symbol of how our friendship began. I met him in 2017 when I moved to San Francisco and took over my friend’s room in a house in the Mission District, where Ben was one of the housemates. The first thing I heard about Ben was that he was developing an escape room with his friends. And by that, I don’t mean only the puzzles but the whole thing: the room, the experience, the logistics.
“I don’t remember when you guys stopped,” I say. “Was it when the pandemic hit?”
“Yeah. When Covid started, it just didn’t make sense to continue developing it. We were building it out in South San Francisco, and even before the pandemic, it was already hard work. Developing an escape room is building out a full business; you have to think about applying for permits and you have to understand the zoning rules and then you have to figure out how all that affects what you can develop creatively. I was doing this in addition to a full-time job, so it was a lot.”
The South San Francisco escape room didn’t happen—at least for now—but Ben still gets his puzzle-and-escape fix regularly. He and Clare, his fiancée, often compete in the MIT Mystery Hunt, one of the oldest and trickiest puzzle hunts in the world, which takes place every January. And in between, whenever he can, Ben does escape rooms. He tracks his escapes on Morty, an app that enthusiasts use to discover, book, and log their escape-room adventures. When I ask Ben about his favorite escapes, he gives me seven names off the bat, then pauses, and decides to extend the list with an eighth contender: the room we just escaped.
The list, grouped by city, is:
San Francisco Bay Area, California, USA
The Roosevelt Escape Room, developed by Palace Games, in San Francisco, California, USA
The Attraction, developed by Palace Games, in San Francisco, California, USA
Popstar’s Room of Doom, developed by SCRAP/Real Escape Game, in San Francisco, California, USA
Gold Rush, developed by The Escape Game, in San Francisco, California, USA
Ghost Patrol, developed by Trivium Games, in Emeryville, California, USA
Boston, Massachusetts, USA
The Storyteller’s Secret, developed by Boxaroo, in Boston, Massachusetts, USA
Level99 Rooms, developed by Level99, in Natick (Boston area), Massachusetts, USA
Seattle, Washington, USA
The Storykeeper, developed by Locurio, in Seattle, Washington, USA
That Ben was drawn to meticulously crafted and often monumental stories did not become apparent to me until the two of us moved out of the five-bedroom house in 2019—by which point we had become friends—and moved two blocks away into a two-bedroom apartment.
He talked a lot about the quiet but epic films by Denis Villeneuve, the psychological intensity of Alex Garland’s movies, the atmospheric grandeur in Christopher Nolan’s work. We started going to the movies together and I found it amusing how quickly we grew to know each other’s tastes: I had no interest in Marvel films and Ben would not be seen at screenings of coming-of-age dramas à la Call Me By Your Name. He wanted great and ambitious stories that stretched the boundaries of human ingenuity, stories that engrossed the mind.
Living together, I learned that this passion extended to video games as well. I always hesitated, however, to call Ben a gamer because the games he played and loved were not the ones I associated with the gamer community, like Dota or League of Legends. His favorites are instead either hyper cinematic or hyper brainy. In order of their release year, they are:
The Last of Us, developed by Naughty Dog (2013) for PlayStation 3
Outer Wilds, developed by Mobius Digital (2019) for PC and consoles
Baba Is You, developed by Hempuli Oy (2019) for PC and consoles
The Last of Us Part II, developed by Naughty Dog (2020) for PlayStation 4
Elden Ring, developed by FromSoftware (2022) for PC and consoles
I ask Ben, as we make our way over from the Financial District to his new place in the Mission, whether there is a correlation between the video games and the movies he loves. He thinks about this one for a bit.
“I’m not sure. Outer Wilds, which is probably my all-time favorite, could not be told as a movie. It’s a puzzle exploration game that feels like an escape room set in an open world. Then there is The Last of Us, both parts, which has actually become a TV show, and I can see why—at its core, the game is about the relationship between the two main characters and it’s a really well told story. Then you have something like Baba Is You, a super clever game in which you can change the rules of the game by moving these word blocks.”
“Yeah,” I add, “they’re all unique. Maybe it’s the world-building aspect? I feel like all these games have their own worlds in which you’re immersed as a player, and that’s something I think you like in movies too.”
“Definitely. World-building, especially when it’s done through big universes which share characters, is something I like across games, movies, and books. Clare and I both love this sci-fi fantasy author, Brandon Sanderson, because he does this in his books—characters from one universe will be referenced or will even appear in another. That’s why I also love the Marvel Cinematic Universe, like how you had Tom Holland and Tobey Maguire and Andrew Garfield all in one Spiderman movie.”
Ben’s new place is only a block away from the two-bedroom apartment we lived in. Structurally, the apartment is different on many levels; it’s newer, brighter, higher, and more modern. But it feels familiar, particularly in the living room, where Ben’s keyboard and electric guitar catch my attention the same way they did when we lived together. Ben is deeply musical: he plays and composes, he sings, and he is unsurprisingly particular about the music he listens to. His genre of choice? Heavy metal.
“It’s funny in a way,” I say, “because you’re very chill and quite pragmatic. Metal, at least to me, has always felt explosive, maybe? Aggressive?”
“I don’t necessarily agree with that,” he notes. “For sure it is more abrasive in terms of sound, but I don’t think that’s true in terms of lyrics. In my case, I like pop music a lot as well, but metal albums have always worked better for me. I feel like they take you on a musical journey.”
For the record, his favorite full-length albums, in order of their release year, are:
Awake (Dream Theater, 1994)
Metropolis Pt. 2: Scenes from a Memory (Dream Theater, 1999)
The Mountain (Haken, 2013)
The End of Everything (Plini, 2015)
Senpai III (Sithu Aye, 2021)
“The technicality of the musicianship is also something I appreciate about metal,” Ben adds. “It’s complex and there is a lot of music theory behind it. I also think the metal style of singing is underappreciated; it takes a great level of craftsmanship to sing that way without hurting your voice while sustaining the grittiness that is characteristic of the genre.”
I suggest Ben play a few pieces for us. On the guitar, he plays one of his favorites, “In the Presence of Enemies, Part 1” by Dream Theater, a song he grew to love because of its many movements. On the keyboard, however, he goes for something different—a majestic piece, at times triumphant and at times retreating, what sounds like a soundtrack to a cathartic epilogue. Ben tells me it’s his own composition, which he had finished during the pandemic after working on it for almost five years.
“You’re kidding. I had no idea you’ve been working on this. What is it called?”
“Great Beasts,” he says.
𐫱
In the afternoon, Ben and I go to Glen Canyon Park. The park, known for its dramatic terrain, is one of the city’s lesser known natural wonders whose resplendent but temperamental canyon walls look like something out of Lord of the Rings. It’s an imposing work of nature and, come to think of it, a place I am not surprised Ben wanted us to visit as part of his interview. Possibly it is a source of inspiration and possibly a treasure of visual data points which he is unconsciously conditioned to make sense of as someone in a job that might seem entirely incongruent to his proclivity for the grand and the experiential. A job that instead epitomizes rationality and precision: software engineering in the autonomous-vehicle space.
Ben came into this line of work steadily. He started programming toward the end of middle school as part of FIRST, fully known as For Inspiration and Recognition of Science and Technology, a non-profit that organizes many youth robotics and tech competitions across the US. Though robots were his first love—he especially loved working on robots’ automated routines—he took a bit of a detour in college and focused on web development for a while.
In 2017, when I met him, he was a full-stack software engineer at Google, developing internal features for the company. I didn’t know much about what Ben was interested in professionally back then but I started noticing he was spending a lot of time after work studying, as if we were in school again. Once I asked him what that was all about and he said he wanted to switch to deep learning—a concept foreign to me and many others at the time but one that soon became the harbinger of the AI boom.
“Oh yeah,” he notes, laughing. “It’s crazy to think that was only eight years ago; I don’t even hear it being called ‘deep learning’ so much anymore. But back then, it was just starting to hit the mainstream. A few friends were already in this field and through my conversations with them, my interest in AI deepened. I was also realizing how much I missed working with physical systems—things that interact with the real world. To me, there is something satisfying about writing code that controls physical objects, rather than just screens or apps.”
“I also remember,” I add, “it wasn’t just any artificial intelligence you were interested in. Even then, you were determined to work on self-driving cars.”
“Yeah, because self-driving cars felt like the most consumer-facing application of robotics out there—and I think they still are, at least to me. And importantly, it’s a domain where technology has the potential to really save lives. So many people are injured or killed on the roads every year. That felt meaningful to me.”
Ben joined Zoox, one of San Francisco’s leading autonomous-vehicles companies, around the time he and I started living together. The city was in its boom period, investments abounded and bolstered tech-industry ventures, and there was a palpable feeling in the air that San Francisco was the place to be. And then—the COVID-19 pandemic struck.
San Francisco withered during these years and the city’s tech landscape cracked. News of layoffs permeated every conversation. Burnout from technology, driven by unrelenting Zooms and Hangouts and Facetimes, festered. What was a technopolis dense with energy a few years ago was now a vacuum, hollowed by a malaise. The self-driving industry was not spared either; while the city’s prominent companies deployed their fleets during these years for training purposes, the momentum was decelerated by the public’s growing distrust of this innovative technology.
All to say, for anyone working in this space, it was an ideal time to pull the brakes and jump ship. But Ben did not—seven years have passed and he is still at Zoox, working diligently to put the next-generation self-driving robotaxis on the road. By the time the AI craze overtook every corner of the tech industry in 2024, he was already a veteran in the space.
“So, after all these years,” I ask him, “what would you say is the biggest challenge the self-driving industry needs to solve now?”
“Good question,” he says. “I think the industry’s at a point where we understand a lot of the basics—both the core problems and many of the solutions. But scaling the systems to handle more complex, real-world edge cases is still a huge challenge.”
“Can you give me an example of an edge case?”
“Sure. Take something like a vehicle cutting in front of you while you are in a self-driving car. Right now, we can handle that pretty well, something like 9,999 times out of 10,000. But that one rare instance, in which the other driver is especially aggressive, might not go as smoothly. It’s not that we don’t know how to handle cut-ins, it’s that there are weird variants of that scenario in which the system struggles. So edge cases aren’t always one-off anomalies, they can be variations of something we already mostly handle.”
It’s worth stepping back a bit to provide color to what Ben is saying. The autonomous-vehicle industry stands on three intertwined challenges: perception, or “seeing” what’s happening in the environment, prediction, or “anticipating” how other objects in the environment will behave, and planning, or “figuring out” what to do based on this information. Ben started out in the industry through working on prediction but has since moved to the Planning Machine Learning team.
In a nutshell—and this is admittedly an oversimplification—Ben builds models that take in a bunch of information perceived by the self-driving car, like cones and barriers and other vehicles, and predict what will happen in the environment and then make decisions based on those predictions. The models can do this because engineers, like Ben, train them through real-life data points, collected to record what happened in the environment in similar situations.
Here’s why this matters for Ben’s point about edge cases. Those scenarios, like the one-in-ten-thousand chance of a really aggressive driver cutting in, are too specific and too rare to be accounted for individually. It’s also not a sustainable way of building artificial intelligence because there are too many edge cases, many of which are “unknown unknowns,” so there is no way a human could bruteforce a solution to each. What Ben therefore has to do is teach computers to generalize behavior across situations, to create systems that understand what perceived objects mean across infinite contexts. It’s about building systems flexible enough to adapt to new situations while being rigid enough to recognize the tiniest possibility of a risky situation.
“Obviously,” I add, “from a technical point of view, this makes sense. But I also understand why, whenever I explain this to people in my life who are skeptical of self-driving cars, there is resistance to this. We don’t think of humans thinking this way, even though they do, so it’s easy to get the perception that self-driving cars are less safe. What do you think about that?”
“It’s definitely a tension,” he says. “It’s hard to balance objective safety with human trust. But I also think it’s a problem that gets solved through exposure. As people see self-driving cars more regularly, I think they grow more comfortable with the idea of it. Adoption will take time, and that’s why a methodical rollout is important. If you rush it and problems show up, it undermines trust. But if you introduce AVs gradually and let people adapt, trust will follow. That’s what we’re already seeing here in San Francisco—people are used to them.”
An hour into our hike across the park, we reach the Glen Park Playground and it occurs to me that I have not paid any attention to the path we took to get here. My attention, undivided and unperturbed, has been immersed in the ever-expanding universe of self-driving cars and the infinite amount of questions that demand answers: could self-driving cars ever become personal companions? What if self-driving cars ever become sentient? Who makes the rules that govern our relationships to this technology, once entirely fictional, now undeniably real? In the end, what occurs to me is that this too is a universe of stories, a universe with worlds and rules that stand independent of the reality we’re in.
“How long will this last for you?” I ask Ben. “Is there a problem that you are waiting to solve? And when you do, will you know it’s time to move on to something new?”
“I don’t think it’s about a singular problem or project,” he says. “For me, the most important thing is that I keep working on self-driving cars until they fulfill their promise—reducing injuries and fatalities on the road. That was always my motivation.”
