PLAYERUNKNOWN reveals how Prologue: Go Wayback! lays the groundwork for an Earth-size game

The Amsterdam where PLAYERUNKNOWN Productions resides isn't the one you'd see on a postcard. Across the placid water of the North Sea Canal, away from the slender canal houses and whirring bicycle traffic of downtown, lies the NDSM neighbourhood. Named after the shipbuilding company whose yards sprawled across the district until 1984, NDSM fell into disrepair in the '90s before a regeneration project commenced post-millennium.

That project is ongoing, with construction yards and brownfield sites dotted between the trendy bars and modern office buildings that comprise NDSM today. On the wall of an old warehouse converted into a street art museum is a vast mural of Anne Frank, while moored in the wharf where the frequent ferry across the North Sea Canal docks is a floating hotel (called the "Botel") and an actual yellow submarine.

It's a slightly surreal space, separate from Amsterdam but also inescapably part of it. When I ask Brendan “PLAYERUNKNOWN” Greene why he selected the area, and Amsterdam in general, to found PLAYERUNKNOWN Productions, his answer is initially pragmatic: "I was travelling a lot with PUBG, initially. I was going all over the world. And here they have Schiphol Airport. It's super easy to go anywhere."

Then his response takes on a more personal quality. "I just like it here, especially this side of the river as well. It's quieter. Someone said it's like a mini-Berlin, almost."

NDSM's mini-Berlin has little in common with Prologue: Go Wayback!, PLAYERUNKNOWN Productions' debut survival game, which is set in a tempestuous and forested wilderness. But it does provide a handy frame of reference for thinking about the project. Go Wayback is at once a separate entity, a game in its own right, and also part of something much greater—a decade-long plan to create a virtual space the size and complexity of Earth itself.

Early DayZ

This grander project, which is codenamed Project Artemis and described by Greene as a blend of Minecraft Survival Mode and a 3D version of the internet, is difficult to hold in your head—which is perhaps why it seems to spill out of Greene with minimal prompting as we sit in his office overlooking the North Sea Canal. At 49, Greene both looks and sounds a decade younger, especially when speaking about his grand vision in an effusive and meandering flow.

The origins of both Go Wayback and Project Artemis stretch back to Greene's first encounter with the game that pulled him into the industry, the Arma 2 zombie-survival mod DayZ.

"I could see the possibility of the platform, and especially emergence. That's what made me fall in love," he says. "Seeing this kind of survival game where you tell your own story, rather than it being a path you're meant to follow, that filled me with a lot of excitement."

It was that excitement that led Greene to create DayZ: Battle Royale, which (via another mod designed for Arma 3) eventually led him to work with South Korean publisher Krafton to direct PLAYERUNKNOWN’s Battlegrounds—or PUBG for short, and now known as PUBG: Battlegrounds.

PUBG made "battle royale" the most popular multiplayer shooter format of the last decade. Its core tenets, with players scrounging resources and fighting it out in an ever-shrinking battlefield, inspired games like Fortnite, Apex Legends, and even Call of Duty's Warzone spinoff.

But for Greene, the appeal of PUBG was as much about the authentic landscapes and grounded survivalism inherited from DayZ as the rules of the game itself.

Following PUBG's success, Greene wanted to make a game with that kind of realistic world, but with the more open-ended survival loop and player-driven community of a game like Rust. "I always wanted to do a bigger survival game, because I loved what happens in Rust and this emergent space" he says. "I thought with 100x100 kilometers, you could do trade routes, or if there was a mountain where you knew there was iron, you could maybe become a salesperson for iron and become very wealthy that way, or a warlord."

Put another way, he wanted a survival game where "you could make a helicopter matter." Says Greene, "In DayZ, it takes you across the map in thirty seconds or a minute—but imagine cutting the time from, you know, 10-12 hours down to 10 minutes."

PLAYERUNKNOWN Productions initially started working on this project in 2019 with Krafton, but the two parted ways after a couple of years. All the while, Greene's idea grew more ambitious, expanding from a 100x100 kilometer survival game to an experience taking place on a map the size of Earth—and one that involves a lot more than mere survival.

Imagining such a game is one thing. Realizing it another. Greene admits that as recently as 18 months ago, PLAYERUNKNOWN Productions didn't know how to make it. "We didn't have a serious plan," he explains. "The leadership team I had at the start, up until a year and a half ago, it wasn't the right team to lead the studio. They didn't have enough experience making games. I don't think they accepted the vision fully from my side."

Greene also admits that his own experience leading a studio was lacking. "I was a photographer, graphic designer, and then someone gave me a bunch of money and a studio," he says. "I had very little experience, so there's a deep learning curve."

This led to PLAYERUNKNOWN Productions recruiting a new leadership team, who hashed out a whole new plan for the project. "They sat down and went like 'Can we do this? Here's the vision. [Greene] wants to do millions of players, Earth-scale worlds—basically a digital place,'" he says. "After 10 minutes of hemming and hawing they were like 'Oh wow, we can do this,' and they came back with a plan."

 

Weather Permitting

That plan began with the creation of two products. The first is Preface, a playable showcase of the planet-generating engine technology, named Melba, that will ultimately power Project Artemis. Preface was released for free late last year. The second is Prologue: Go Wayback!, a survival game built in Unreal Engine that embodies the mechanical principles that will, one day, form the systemic foundation for Project Artemis.

Despite all the ideas and ambition swirling around it, the premise of Go Wayback is surprisingly simple. You play as a woman named Lucy, and you spawn in the relative safety of a cozy (if slightly rundown) woodland cabin. Your goal is to cross a stretch of terrain several kilometers across to reach a weather tower on the far side of the map. As you hike across the game world, you'll need to manage your food, hydration, and temperature, building fires out of burnable materials and taking shelter where you can.

While Greene oversees the studio as a whole, Go Wayback’s Creative Director is Scott Davidson, a veteran British game developer who worked at Blitz Games and Rebellion before spending five years at Facepunch Studios as Art Director on Rust. “I basically got Rust’s look to the point where it was very early access," he says. It was this experience that landed him the job at PLAYERUNKNOWN Productions. "When I looked at Brendan's vision document, it said Rust about 15 times," Davidson says.

Although Go Wayback is a singleplayer game (for now), like Rust it places heavy emphasis on depicting a survival situation with uncompromising authenticity. Your starting cabin contains some basic survival equipment, including a fire-striker, a torch, a map, and a compass. But you can only take with you what you can carry in your hands and in a drawstring backpack, and your clothes are unsuited to the inclement weather that frequently and capriciously batters Go Wayback’s terrain.

Unlike most survival games, your biggest concern in Go Wayback is not finding food or water, but staying warm.

"We want to prioritise temperature as the metric," says Davidson. "You very rarely die of dehydration in survival circumstances. You very rarely die of malnutrition, because it takes days and days and days for you to burn through all the calories you've got left." While you'll still have to manage your food and water intake in Go Wayback, PLAYERUNKNOWN Productions ultimately plans for these metrics to feed into the temperature mechanic. "When your hunger and thirst go down, you lose the ability to regulate your temperature, so your temperature is going to drop faster," Davidson says.

Because of this, survival in Go Wayback is primarily governed by two systems.

The first is fire. Go Wayback features an elaborate fire simulation, whereby instead of creating the Platonic ideal of a campfire using a blueprint from a crafting menu, you simply assemble a pile of burnable items on a surface and then ignite them.

"It's tinder, it's kindling, and it's fuel," Davidson says. "You have to start with pieces of paper and cardboard, you light those with some sticks on top. Then the sticks catch light, and then you can put logs on top, and then the logs will burn over time. And then you just keep feeding the fire with a couple of logs, and it will keep going and keep giving heat, and the heat is retained within structures."

The reason the system works like this is partly for authenticity, but also to encourage the player's ingenuity. You can create a fire out of most things that are intuitively burnable in Go Wayback, like the hardback books you find in your cabin, or the drink coasters lying on the table. Likewise, fires can be lit using any source of ignition. Typically, this will be the fire striker that you can carry around, as it's easily portable. But you can also use the hobs on the cabin's cooking stove—provided the electricity is working.

There are some limitations. You can't light an entire cabin on fire, for example. "That would be a structural integrity system, which would require a lot of physics," Davidson says. That said, there is a "physical feedback loop" to objects that should logically burn but don't. "If you were to set a fire next to a cabin wall, the cabin wall will turn black," Davidson continues. It’s something they hope to implement in the future.

The other key survival-related system is the weather. Go Wayback features a detailed simulation of rain, wind, snow, and hail, all of which can (or will be able to) affect both the environment and your character's personal situation. The simplest example is that rain makes your character and the wider environment wet. This makes your temperature drop faster and makes burnable items harder to ignite. Hence, when carrying fuel or tinder, it's important to keep it in your backpack so it stays dry.

But weather also affects your situation in other ways. When you open the door of your starting cabin, you can see the wind blowing through the aperture and track how it lowers the temperature of the interior. Closing the door will obviously stop this one incursion—but to stop the wind blowing through the cabin's shattered windows, you'd need to find some wood to nail across the broken glass.

"The weather is dynamic, and we don't control it," Davidson says. "It means that things like shelters just appear from the objects in the world. So you can walk into a big section of rocks, and one of the rocks has fallen over against another one, and you'll notice that there's no rain falling below it."

There are further plans for Go Wayback's weather system. As you wander through its world, you'll regularly encounter patches of mud that slow your character down. Eventually, Greene wants these patches to be dynamically created by the game's rainstorms.

Moreover, the team wants Go Wayback's hailstorms to include giant, apple-sized hailstones that will damage you. Even though these aren't physically present in the game yet, their audio is, so you can hear them thudding into the roofs of Go Wayback's cabins. "We've got to work out how you can get the hailstones to hit the ground around you and the environment, and then when one hits you, you correlate that with the things hitting around you, and don't go, 'Why am I taking damage?'"

What this amounts to, at present, is a particularly harsh hiking simulator, with you trying to wend your way toward that weather tower, seeking out shelter and useful items as you go. In this task, navigation plays a big role. Your map shows you key locations like your starting cabin, as well as other cabins you can shelter in. But it doesn't track where you are, meaning you need to deduce your location and heading by using a compass and studying the terrain.

Due to the size of the landscape, finding points of interest can be surprisingly challenging, and your situation can change drastically while trudging to the wilderness. The onset of night can leave you hopelessly lost as you fumble around in the darkness, while a sudden blizzard or rainstorm can freeze you in a matter of minutes. Even something as simple as a slope can be treacherous, with a slight increase in incline threatening to tumble your character to their death.

Only when you die (and die you almost certainly shall) will you see Go Wayback's biggest trick of all. Unlike DayZ and PUBG, Go Wayback's map is not hand-crafted. It is generated on the fly for every playthrough. Moreover, these maps are not created using conventional procedural generation. Instead, they are created using large language model-based machine learning—or as it is now more commonly known, Machine Learning Generation.

 

Generation Hill

Although Go Wayback doesn't run on the same graphics engine that will ultimately power Project Artemis, it does make use of PLAYERUNKNOWN Productions' same bespoke level generation tech, which forms part of the engine and is built upon machine learning techniques. Greene was drawn to using machine learning due to the prohibitive size of the maps he wanted to create.

"When we tried to do 100x100 kilometres, we found you can't do that using the traditional way of doing it. Even storing that amount of data, you're shipping hard drives to people," Greene says. "And I thought, well, isn't there a way to use a low-res map to inform what the high-res looks like, so we can generate the whole, massive map? The researchers at the time said 'Yes, machine learning can do that.'"

Prologue: Go Wayback! doesn't require maps this size, but according to PLAYERUNKNOWN's Senior Machine Learning Research Engineer Joey Faulkner, there are other advantages to using machine learning for landscape generation—namely, greater variety. Conventional procedural generation algorithms, as Faulkner explains to me, are based on rules that produce patterns that become easily recognizable over time. "The opportunity we have with machine learning is to go from setting up rules to this black box that can produce almost anything."

Before delving any further, it's worth addressing the elephant in the room. Use of ML generation technology (both within game development and more generally) is a controversial topic, particularly regarding general-purpose AI tools like ChatGPT and Midjourney. Problems range from copyright issues regarding the data these tools were trained upon, to the environmental cost of using them, the quality of their output, and how they might be used to replace humans in the workforce.

Regarding PLAYERUNKNOWN's machine learning tech, however, there are some important caveats. The tech the studio uses is its own, in-house technology. It is designed specifically for the purpose of generating landscapes for its games. It generates those landscapes locally on the machine the game is being played on, so it does not require a vast, energy-ravenous server infrastructure. The ML is trained on publicly available earth science data from sources such as NASA. And finally, Greene says that PLAYERUNKNOWN Productions avoids using AI elsewhere.

"We try not to use generative AI for the artistic process. I do see there's advantages for ML and generative AI with, for example, textures. For an earth-scale world, having an agent that can generate a fresh texture or a unique texture for every tree in the world, that to me is interesting because it gives variety," he explains. "But we're very careful about how we use it. We spoke at one point about using AI for voices in certain instances, but we shut it down and we want to use voice artists."

Even in the specific area the studio uses machine learning, it does not replace the existing creative pipeline. Indeed, as PLAYERUNKNOWN Productions has discovered, machine learning generation left to run of its own accord won't produce interesting playscapes.

"If you let the machine learning do what it wanted, it would encounter this '10,000 bowls of oatmeal' problem where it would just try and produce the most normal looking stuff," Faulkner says. "So we were like 'How do we encode interesting gameplay features by game designers, or procedural stuff?'"

The system Go Wayback uses is referred to by the studio as "Guided Generation," and it's a mixture of procedural generation, machine learning, and good old-fashioned handcrafted art.

"Everything foundationally builds off the ML generation of the heightmap every time the map builds," says Alexander Helliwell, Senior Environment Artist at PLAYERUNKNOWN Productions. "On top of that, we'll have spawning of different biomes where they're appearing. And within the biomes themselves, they're built up of tiles, and we build [the] tiles."

The landscape players explore in Go Wayback is based off Czechia's Bohemia Switzerland national park, an area of rugged terrain and sandstone cliffs that also inspired Arma 2/DayZ's original map, Chernarus. Helliwell says this area was chosen not for its spiritual significance, but because it provides a large amount of geographical variety in a small area.  

"We want to expand the path for the player, make it more of a maze for them," he says. "Tonally, having a forest with rugged terrain and trying to hike through that implies hard survival."

PLAYERUNKNOWN's ML algorithm is trained on publicly available real-world data from this area, which Faulkner says is "augmented" at small and medium scales with procedural generation, a process which is led by the art team. The difference between this and conventional procedural generation is that the ML can make assumptions about how a landscape should look, something that more conventional rules-based algorithms would not be able to do.

For example, Go Wayback’s Guided Generation system is designed to imbue maps with verticality, which produces dramatic environments that are more engaging for players to navigate. Since Go Wayback is designed to take place in natural-looking landscapes, the algorithms that define this verticality are based upon natural drainage networks—how rainfall drains into streams and rivers, eroding terrain into mountains and valleys in the process.

In this manner, Go Wayback’s map generation is "guided" by top-down artist diagrams of rivers and mountains. From this simple schematic, the ML can generate millions of unique heightmaps. "We find that there are sub-rivers popping up in Go Wayback that we don't even ask for," Faulkner says. "We draw a main river, and then the ML model is like 'Well, a river this wide should have tributaries,' and you start to see that in the actual outputs.'"

Go Wayback's guided generation system is beneficial—but it also brings with it the challenges of both working with a machine learning system and a traditional art pipeline. Regarding the former, Go Wayback's ML is not exempt from the glitches experienced by other AI programs. "There was a part of the map where there was just a square," Faulkner says. "Like the map was completely normal, but then there was a cube of land above everything else, and I still have no idea where that came from."

At the same time, Helliwell is faced with the challenge of ensuring Go Wayback's landscapes are artfully composed no matter what the generation system comes up with. "We're finding those moments in the game where we're like 'Ah, that looks great in this place, this looks great at this time. Why does that work? Why does this place not work?' It's a constant back-and-forth, testing what will work for the next billion maps."

Because Go Wayback's setting is familiar and its landscapes are heavily forested, it's hard for a layman to get a sense of what exactly the tech is doing. That said, the contours and undulations of Go Wayback's maps seem closer in appearance to the handcrafted landscapes of Arma 2 than, say, No Man's Sky's planets—which is significant when the end goal is to produce similarly planet-sized maps.

Ultimately, PLAYERUNKNOWN Productions uses machine learning because the team believes it is the right tool for the job they want to achieve. But Faulkner also hopes it can serve as an example of how machine learning should be used, compared to how it is being deployed elsewhere. "Generative AI is thought of now as the nuclear power plant, powering these massive data centres, which are telling you a tomato soup recipe. But everything that we're doing creating these worlds with Guided Generation, it's all running on your hardware," he says.

"When I think of the future of machine learning, even at this moment where people are saying, 'Oh, AGI [Artificial General Intelligence] is around the corner,' I think what we're doing here is a much more realistic version of how machine learning can fit into things in the future," Faulkner continues.

 

Wayback Machine

As someone who spent dozens of hours hiking through DayZ's Chernarus, the idea of braving the elements in a landscape that's similar but also renews itself with every playthrough certainly has appeal.

Compared to existing survival games, however, Go Wayback is undeniably thin in terms of things the player can do—at least in its current state. There are no animals to hunt or be hunted by. There is no combat or puzzling beyond navigation. There isn't even any crafting, insofar as you would recognise it in other survival games. This is something Greene himself acknowledges. "We've got some feedback that is like 'There's not much to do,' and yes, there isn't much to do in the world right now."

But this version of Go Wayback is far from the finished article. When Go Wayback launches, it will do so in Early Access, with the team estimating a two-year alpha period before the game hits 1.0.

Davidson is cagey about what exactly will be coming to Go Wayback post-Early Access launch, partly because the launch version isn't fully nailed down yet, but also because it will depend on what players want added. "Our roadmap is going to be a little bit more open-ended and driven by the community," he says. "It's then up to how we prioritize those things and make sure we get to them in a timely fashion."

Broadly, though, the plan is to considerably flesh out the existing survival skeleton. "You noticed there's an electricity system in cabins, and there's fuses? Eventually you'll be able to fill the generator up. Eventually, we'd have other things like plugs, so you can plug things in," says Davidson. "We want the player to get to a cabin and go, 'This is pretty good cabin, I'm gonna try and make this good, and I found a nice table, and I'm gonna bring the table from that other cabin, and I'm gonna make it my little home, and I can fish, hunt, and do all the things you would do.'"

Then there are features the team wants to bring or is at least deliberating bringing to Go Wayback. Davidson says that Greene would like to see the ability to chop down trees in the game, but he isn't sure that it fits with Go Wayback's particular style of survival game. "Have you tried chopping down a tree? It's really, really hard."

Greene also wants to add multiplayer at some point. "I would like it to be co-op multiplayer, if you want to play that way," says Greene.

One thing definitely coming to Go Wayback, though, is more of a story. It's written by Greene—the first time he's authored a story for one of his games, a process he describes as a "great difficulty." He hasn't decided how Go Wayback’s story will be implemented yet, but says it will have a "light touch" to it. "I was even talking about maybe doing a comic book. Like I'm doing a three-volume comic book that tells the story, and doing it that way, that these games take part within our universe."

Indeed, one of the key elements of Go Wayback’s story is that the game doesn't take place in the real world. "We're gonna make it clear this is actually in a virtual space," Greene says. "We have this idea where at the edge of the world, we're going to do data washing as an effect to make it clear that this is not a real space."

Greene also unveils some of the more specific plot points he is working with. "The story is about a father trying to deliver a message to his daughter," he says. "It is a hacked version of her more complete game that he is using to deliver a message." He doesn't specify whether Lucy, the game's playable character, is the daughter in this scenario.

Greene is reticent about sharing some of these details, but he has publicly hinted about elements of the narrative multiple times already. "The reason it's called Go Wayback is because I hid stuff in the header and in the first tweets of the Go Wayback account that was pointing you to the Wayback Machine archive of the Twitter account," he says.

"No one got that. I did a bad ARG that no one got. I even had Morse code down the bottom of the header image of Go Wayback," Despite this, Greene stuck with the name anyway. "In our universe, the name Go Wayback just worked. It's meant to be a vision of a world in the '80s, as imagined sometime in the future."

 

Meta-versus

Go Wayback’s story is also not entirely self-contained. Its narrative threads will, eventually, connect Go Wayback to Project Artemis. But PLAYERUNKNOWN Productions will not jump from the former straight to the latter. In between these projects will be what Greene refers to as "Game Two."

Greene doesn't provide many details on what kind of game this will be, but his rough idea is an FPS or RTS inspired by Command & Conquer that will involve tens or hundreds of players—like "Command & Conquer but first-person." He says it will be "in our engine" and that he hopes it will be used to test "marketplace and trading and digital ownership."

At the same time, Greene also has a notion about nesting the previous game from the project into the next one. "Game Two should be playable in Game Three, right? These are just layers to what we're building on top," he says. "I'd really love to be able to play Go Wayback within our own engine at some stage. Maybe in Game Two, I don't know."

It's possible, then, that Game Two might pan out as a blend of Rust and Command & Conquer, where players work their way up from subsistence-level survival to coordinating mechanised base sieges with tanks and helicopters they've built themselves.

Whatever Game Two ends up being, it will set the stage for Game Three, a.k.a. Project Artemis. Greene envisions Project Artemis as a 3D internet that players can explore in first-person, using the Melba engine to create their own worlds to play on.

"The ultimate platform of Melba is that everyone creates their own digital worlds, or exists in one like a Minecraft Survival layer," he says. "That's the end goal here, is to build a world creation space so you can have the internet in 3D, but every world is a page."

As for what players will be able to do in this 3D internet, the ideal is that you'll be able to build your own experiences much like Minecraft players currently do, with the survival systems as the foundational layer that players could opt in or out of. "I'd like to have a Civilization MMO-type structure where you can just make the world yours," Greene says. "You can create towns and cities, almost like Cities: Skylines." This will, naturally, involve an elaborate building system. "I want people to be able to build a Millennium Falcon using wood, like they did in Valheim."

This multifaceted experience would be massively multiplayer, with thousands (if not millions) of people coexisting in the world. But Greene would also like to include AI companions to help further populate it, assisting players with busywork. "I don't want to have, like most survival games, a loop where you just spend a lot of time gathering stuff."

These AI companions would possibly be powered by "tiny LLMs," so that "you can speak with them and they get to know you," though Greene isn't certain about this idea. In any case, as these player and non-player communities expand, Greene envisions they'd be able to create their own activities, "allowing you to transform your part of the world into dirt bike races and an FPS game in an abandoned skyscraper."

It's a concept that could be summarised as a metaverse, although that's a word Greene prefers to avoid. "Metaverse is a very loaded term, so I try not to use it as much as I can," he says. This is largely because the concept of the metaverse has become associated with dubious Web3 and blockchain projects, the latter of which Greene says is "interesting as a decentralized ledger," but isn't something he plans to incorporate into Project Artemis.

Project Artemis will have a financial layer to it, however—one which will be the ultimate source of revenue for PLAYERUNKNOWN Productions once the game launches. Like much else about Project Artemis, the plan for this is somewhat vague, but Greene imagines PLAYERUNKNOWN Productions would eventually become Project Artemis’s equivalent of MasterCard.

"This is all very dreamy, but I eventually see the engine being shifted into a nonprofit foundation, much like the World Wide Web Consortium," he says. "Then we will become a content company and a platform company, managing the transaction layer and the marketplace."

 

Boots on the ground

All of this is far off in the future. Go Wayback itself has, essentially, a full development cycle to go yet. "This is a long, one-plus-year Early Access," Davidson says. "Then a couple of years of post-full-launch support. We haven't figured it out fully."

The way PLAYERUNKNOWN Productions goes about developing Game Two and Project Artemis rests on the commercial performance of Go Wayback. Indeed, Greene says part of the reason for doing three games is to "derisk as much as possible."

"We have Go Wayback, which will fund us into Game Two, hopefully. We're also looking at raising funding at the moment," he says. "I want to insulate the team from having to worry about making products to sell, because that often destroys some really good ideas."

In the shorter term, Greene's goals are comparatively humble. Prologue: Go Wayback! is currently undergoing rounds of pre-alpha testing. "I'm hoping right now that with the data we get from this, that the game loop is stable when we do launch into Early Access," he says.

"That's what I hope to achieve from the Early Access launch at least. I just want to deliver something that is stable and enjoyable," he continues. "The gameplay is there. It still needs polish. It still needs balance. But I enjoy playing it, and there's not a whole lot of games I enjoy playing."

"There was a guy on our Discord that summed it up perfectly. He said, 'You know, I never got why people played Euro Truck Simulator, but I'm a hiker, and I get it now.' I think there are enough players out there that get that."