AI-Assisted recommendations in gamedev: lessons from a research project
summary
Research project builds a Unity editor that calls a Python REST API and trains on Hard West 2 level data to give element and tag suggestions for designers, yielding about 30% productivity increase for level work. Level recommendations (grid + walking-frame proximity) work best, narrative suggestions fail for small-dialogue games, and tag-driven game-object design has too much overhead. Tested algorithms: a priori, logistic/Naive Bayes, word2vec (failed), with random forest performing best after months of tuning; stacking of models improves accuracy.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 3:36Modern game dev challenges
- 7:55Non-generative AI for design
- 10:45Recommendation system concept
- 14:20AI algorithm testing
- 20:32Stacking algorithms
- 23:46User testing and data
- 27:41AI tool applications
- 33:30Key takeaways
- 36:21Future of AI assistants
- 42:25Q&A
related talks
01h 01mIntroduction to Practical AI
20mStudio Speaks: The Reality of AI Adoption in Games
45mThe AI Revolution in Software Development: From Code Writing to Quality Assurance
42mMaximising Impact with AI: Ethical Strategies for Developing a Top-Rated Game on a Modest Budget
39mAI-assisted development: Common pitfalls and smarter practices
37mAI-assisted development: Common pitfalls and smarter practices
27mEnhancing Personalization Using Recommendation Systems & Multi-Armed Bandits
26mHow to develop useful and unbiased AI