Echo: Ex-Hitman devs bring machine learning to stealth games

The reference books that line the shelves of developer Ultra Ultra’s modest Copenhagen office offers insight into the aesthetic of its first game, Echo. There’s a book of 19th Century Interior Design and a book of Venetian photography. Prometheus: The Art of the Film and Star Wars provide sci-fi reference points, while Metal Gear Solid, Blame! and Neon Genesis Evangelion—all three of which are represented in some form on the shelf—provide the inspiration for character design.

Echo is made up of many familiar parts, but parts that are remixed in a way that makes them feel new. This is fundamental to Echo not just aesthetically, but also mechanically. Unsurprisingly, given Ultra Ultra’s staff of ex-IO Interactive Hitman developers, Echo is a stealth game. It learns from your actions to figure out how you play the game. This causes you to second guess your tactics regularly, almost like playing chess against yourself.

The third-person camera and helpfully placed walls and corners that act as cover are all familiar, then, but the overall atmosphere is different. This is much more of a reactionary stealth game than we’re used to. As such, this is not a game that—even if you’re familiar with the genre—you can simply sleepwalk through.

“How do you interact with your own play style? That’s the core of our gameplay, and that could be interpreted and done in a million different ways,” explains Martin Emborg, Ultra Ultra’s CEO and Echo‘s game director.

“People originally thought that we were doing a game that used some form of long-term machine learning in which, if you played the game for hours on end, it would eventually become you and copy exactly how you play. We didn’t want to do that, though. We had a few early prototypes in which the game became a representation of how you interacted with it on a macro scale, but we found that became extremely punishing.”

These iterations of the game weren’t only punishing, they were predictable and dull. If you performed a certain action at the very start then you could likely never perform it again, given that the machine would be able to see it coming. As such, your quota of potential moves reduced with each play session. By the end of the game, you barely have any strategies left to use at all.

To combat this lack of dynamism, Emborg and his team opted for a system in which the AI learns and unlearns from you. This means that you’re always aware that the game is watching you, because you see the enemies change their approach regularly. But it also forces you to pay more attention to these enemies, because you’re not always immediately sure what it is they know or don’t know at any given moment.

“From that idea of learning and unlearning we create the idea of cycles and blackouts so that the player is encouraged to have a very real sense of being aware of what they’re doing and when and why,” continues Emborg. “With each cycle you’re playing against the last cycle, and when the blackout happens you’ve got a chance to not have to worry about how what you’re doing is going to affect the next one.”

Echo is structured as a series of cycles, which move through two modes: lights on and blackout. Whatever you do when the lights are on is communicated to the AI on the next cycle. Once you perform a certain number of actions the memory of the AI becomes maxed out, and it switches the lights off to prepare for the next cycle. In darkness, the AI cannot see what you’re doing and therefore cannot learn from you.

Early on it’s impossible to know what the AI has picked up on, because it’s never made clear before play which actions are learnable. I played for about an hour, and at the end of that hour I still wasn’t sure about what the AI had learned.

The ability to sprint, crouch, and fire weapons are quite obvious abilities that can be learned, but more surprising examples exist. En, the protagonist, can perform sneak kills when approaching an unknowing enemy from behind, and the AI learns this from you. Dying in this way is a humbling experience, one that illustrates that you’re simply not paying enough attention to your surroundings.

The AI can also learn how to replenish its health from food in the environment, how to operate lifts that provide access to different levels, and that walking across shallow water is safe. You might decide never to walk across water in a bid to create more barriers between you and enemies in a future cycle, but doing so makes it harder for you in the present.

How you weigh up the risks and rewards between the immediate moment and a new future adds a huge degree of personalisation to your play style. Others that played Echo had a very different experience to me. Given that the only weapon I’ve tried is a gun that can only shoot twice before a lengthy cool down period, skill with a firearm is not something that you can fall back on regularly.

“Really, the core idea of Echo is quite simple, but when you experience everything interconnecting it gets very complicated,” says Christel Graabæk, Echo‘s scriptwriter and producer. “I do think that we’ve managed to keep things on a level that’s intuitive. You soon learn whether you might want to wait for the dark to come or not, and you learn that fairly quickly because you’re constantly encouraged to keep your wits about you through how important it is to watch yourself play and think about what you’re doing.”

“I tend to play in a way that, I suppose, could be described at going with the flow,” continues Graabæk. “I do what I think I want to do in the moment and then react to my decisions, for better or worse, down the line. Certainly, though, people do play it in a very meticulous way that’s all about stealth and trying to only do things when you’re in darkness. It’s up to you. There’s no right or wrong, just how you react to what’s presented.”

Echo is a game in which you’re encouraged to enter into a dialogue with yourself, one in which you perform constant self-criticism, engage with ideas about how you affect everything around you, and how your own actions can result in a positive or negative environment.

In this regard there’s a huge amount of philosophical thought that can be read into Echo, and it’s no surprise to hear Graabæk and Emborg reveal that the ideas of cycles and AI learning came from their initial thematic and narrative inspirations, rather than the other way around. Jorge Luis Borges’ short story The Library of Babel is mentioned by the two as an inspirational text—and it’s easy to see the link.

Combined, the books beset on shelves within The Library of Babel are believed to contain every possible ordering of the letters in the alphabet (plus the comma, full stop, and space) within the format of a 410-page book. As such, the library must contain within it every book ever written and that will ever be written—as well as a huge volume of meaningless text.

The biography of your life, for instance, as well as this article you’re reading, with altered punctuation, exists somewhere in the library. In Echo, every possible set of moves you make exists as a form of the game’s world and enemy design. No matter what set you apply, Echo has any and all sets available to present back to you.

“I’d also throw Dune into the inspiration list,” says Emborg. “When you read that book it’s so complete, but it’s also very difficult to fully grasp as it seems to always change as I’m reading it… the kind of science fiction that deals with very big, ethereal ideas is what we’re trying to do here. How those influences turned explicitly into Echo and the ideas we’ve got in there is a mystery to us and we’re happy to be honest about the mystery element.”