Announcing an ambitious research partnership is the easy part. Turning that ambition into a practical programme of work is considerably harder. Fenris Creations and Google DeepMind are working together to explore AI research using the technology and systems behind EVE Online, including areas such as long-horizon planning, memory and continual learning. In this talk, Hilmar Pétursson, CEO of Fenris Creations, will look at what happens after the announcement: how broad research ambitions are turned into practical workstreams, how teams decide which problems are worth pursuing, and what happens when AI research meets the realities of a mature game and development environment. The partnership is still at an early stage. By November, however, its first ideas will have moved through workshops, scoping and experimentation into tangible outputs and learnings. Rather than presenting a finished success story, Hilmar will give an honest account of the first phase of the collaboration: what the partners set out to do, what they discovered once the practical work began, which assumptions held up, and which had to change.
1. How to move from a broad AI research ambition to something that can actually be built and tested. Attendees will see how the partnership moved from high-level areas of interest into specific workstreams, experiments and definitions of success.
2. What happens when you use a real game rather than a purpose-built research environment. The talk will explore the trade-offs between control and repeatability and retaining the complexity that makes a mature game interesting for research.
3. How game-development and AI-research organisations work together in practice. Attendees will hear about the different assumptions, processes, timelines and definitions of progress that have to be reconciled.
4. How to distinguish between research value, prototype value and production value. A successful AI experiment does not automatically mean something should become part of a live game. Hilmar will explain how the team thinks about that boundary.
5. What changes once the initial theory encounters reality. By November, Hilmar will be able to share the first practical lessons from the collaboration — including assumptions that proved correct, areas that became more difficult than expected, and how the work changed as a result.
