by FC Rattati4:00pm on Thursday 20th August 2026
Curious capsuleers,
When we announced becoming Fenris Creations back in May, we also announced a research partnership with Google DeepMind. We've had some great conversations and discussions with them since then, and two weeks ago our teams met for a knowledge share session to really dive deep into the challenges and the opportunities on both sides. A good part of that session went on why EVE is one of the most challenging environments for their tech, and why they're so excited about it.
Our work with Google DeepMind begins with an offline version of EVE Online running on a local server. This is research conducted in a controlled environment, using simulation and historical information. So why are we doing it?
EVE poses an unusually difficult problem for AI research, and therefore a good challenge. Most games that are chosen have a defined match, a clear objective and a point at which everything resets. New Eden does not reset. It is one persistent universe with changing markets, scarce resources, incomplete information and hundreds of thousands of people pursuing their own agendas. It is vast and complex in a way that no other game is.
An agent trying to understand that world cannot simply learn how to win a battle. It has to plan without knowing everything, adapt when somebody interferes, understand that other actors may cooperate or betray it, and deal with decisions whose consequences can last for months or years. That is what makes EVE interesting to researchers.
Games have played an important role in the history of AI research. Each has introduced a harder problem involving learning, reasoning or planning. EVE is different again. The challenge is not mastering a fixed game. It is understanding and adapting to a living system that never really stops. For us, this is also an opportunity to understand EVE better. Whenever we look at a new technology, the first question should not be whether we can put it into the game. The question is what purpose it serves for players.
We are not interested in building systems that play EVE on your behalf or sit above its rules. If something removes risk, bypasses consequence or replaces player decisions with automation, then it works against the things that make EVE compelling. For example, when we give players tools to deal with their own problems and then see what they do with them. Any intelligent system we might consider in the future would have to follow thesame principle. It should operate within the world, respect its constraints and create more meaningful decisions for players rather than removing them.
There are several practical areas where this research could eventually prove useful. It may help us test complex systems more effectively, understand how changes spread through a simulated economy, or find better ways to help new pilots navigate a game with an enormous number of possible paths. Over time, it may also help us explore new kinds of semi-autonomous governance systems capable of creating pressure, opportunity and interaction inside the sandbox.
But that work is at an early stage. We are not making promises about where it will lead.
There is a useful precedent for games contributing to research without losing what makes them games. Through Project Discovery, EVE players have made meaningful contributions to scientific research while continuing to play EVE. This is not the same model, but the principle is familiar: researchers gain access to something unusually valuable, while what we learn can also improve our understanding of how to act as custodians of our shared universe.
This work is also broader than one experiment or one partner. The EVE Universe gives us several different environments in which to explore these questions, and those environments do not all have the same threshold for experimentation. Tranquility contains 23 years of player history, relationships, institutions, wars and economic decisions. We have a responsibility to treat that history carefully. Other environments give us more freedom to test ideas, observe what happens and pull them back if they do not work. For example, a test server dedicated to a specific AI research project would allow players to opt-in to take a look.
I also understand why players approach any conversation about AI with scepticism. The technology is moving quickly, the terminology is often vague, and people have seen plenty of examples of it being used without a clear purpose. Our responsibility is to be specific about what we are actually doing, distinguish research from future possibilities, and explain the value before asking anyone to accept a new system. This is the beginning of that work.If something we explore cannot provide clear value to players, or if it undermines the agency and consequences at the heart of EVE, then it does not belong in Tranquility.
New Eden belongs to its players. The purpose of this research is not to change that. It is to understand complex virtual worlds better, give players better tools and explore whether intelligent systems can help create more of the conflict, cooperation and unexpected stories that no development team could ever script.
See you in space.
FC Rattati