Mobile Dev Memo Podcast Season 7, Episode 16: AI and video games (with Julian Togelius)
May 13, 2026
Julian Togelius, NYU professor and director of the Game Innovation Lab, explores AI for games and games for AI. He discusses why games are perfect AI sandboxes. He weighs generative tools, consumer backlash, and faster production. He imagines infinite, procedurally generated worlds and how designers shift toward system architecture and human-in-the-loop pipelines.
AI Snips
Chapters
Books
Transcript
Episode notes
Mario Became AI's Model Organism
- Super Mario Bros became a model organism for game-AI after a Java remake sparked an academic competition in 2009.
- Despite heavy study, agents still struggle to reliably win unseen Mario levels without a simulator.
Finance Is A Nonstationary Game
- Trading systems resemble games with non-stationary dynamics and enormous action/observation spaces, so finance uses small, retrainable subgame models.
- Deep-learning practitioners entering finance often face surprise at continued reliance on simple models and rapid retraining practices.
Don't Hide Game AI Experimentation
- Game studios should experiment with AI despite public sensitivity and polarization, because it enables efficiency and new interactive possibilities.
- Julian warns many companies avoid public association with AI, reducing experimentation and slowing creative gains.
