Game AI Meetup

The next Game AI meetup will take place on the 16th of February 2024! Attendance is free, but we do have limited spaces - please update your RSVP if you can no longer make it, to allow others to take your place.

We will have a series of short talks (more detailed to be announced soon), followed by networking with pizza and drinks.

Partner event: Tabletop R&D Playtest Series (February 17):
Come join us at Tabletop R&D: Playtest, where we'll gather in person to test out and have a blast with the latest board game prototypes!
https://www.eventbrite.co.uk/e/tabletop-rd-playtest-series-registration-798933420367

* If you would like to give a short informal talk at this or future events (on any Game AI topic), please contact one of the organisers with a title and short abstract! (Limited spaces available!)

EVENT SCHEDULE
17:30 - 18:00: Arrival
18:00 - 18:10: Welcome
18:10 - 19:10: Game AI Talks
19:10 - 19:15: Closing
19:15 - 20:00: Pizza & drinks social

--- Game AI Talks ---

Steve Gaffney: Can ML help Game Development? - Problems & Opportunities
Steve Gaffney is Technical Production Director at Lucid, and helps keep the engineering and production teams across Lucid's projects organised. Prior to Lucid, Steve worked at DeepMind for 7 years as a program manager, helping to organise fundamental ML research, and prior to this was a Director at Splash Damage, making multiplayer games for PC and console.

James Goodman: Using AI techniques to design and play-test modern tabletop board games
Automated testing of computer games has been around for a while; this talk is on work to extend this to analogue modern tabletop board games ('Eurogames'). We have been working with game designers to integrate AI testing into their design process, and complement human play-testing with data that provides actionable insights on tweaking the rules for the next cycle of the design loop.

Bio: James is currently finishing a PhD in the use of AI planning techniques in tabletop board games at Queen Mary University of London. His research is also being used as part of a university spin-out company, TabletopR&D.

-- Lightning Talks --

Dominik Jeurissen: Playing NetHack with Language Models - Potential & Limitations
Recently, a plethora of LLM-based agents have emerged,
aiming to leverage the planning capabilities of language models. However, existing research fails to push these agents to their limits. To fill this literature gap, we applied an LLM-based agent to Nethack, a complex rogue-like with many items, monsters, and ways to die. In this presentation, we will present our LLM-based NetHack agent, showcase its abilities, and highlight its weaknesses likely caused by the limitations of current LLMs.

Bio: Dominik Jeurissen is a PhD student at Queen Mary University who is exploring new ways of playtesting complex games using large language models (LLMs). Dominik collaborates with Creative Assembly to develop playtesting methods that can be applied to games such as Total War.

Fandi Meng: Solve the deduction games with Information Set Entropy Search(ISES)
Deduction games, such as Wordle and Mastermind, require players to solve puzzles or uncover hidden information through logical reasoning. We introduce Information Set Entropy Search, a new and efficient general algorithm for single-player deduction games, delivering state-of-the-art performance. This innovative method quantifies the number of possibilities in the information set as entropy values to evaluate the amount of information gained from different actions, theoretically yielding the optimal solution with the fewest average actions. Some games are solvable directly in this way, while for others we use sampling to make the approach tractable while still offering close to optimal performance. Furthermore, our proposed deduction game framework is not only useful for automatically testing new games but also provides insights into their appeal. By analyzing the entropy in information sets, we can uncover effective strategies and gain a deeper understanding of what makes deduction games enjoyable.

Bio: Fandi Meng is a PhD student at Queen Mary University who is doing the research on deduction games. Fandi’s research is presently oriented towards investigating the characteristics of deduction games and devising innovative AI approaches.