IEEE CoG 2019 - Industry Day - Csaba Toth and Duygu Çakmak (Creative Assembly)
summary
Total War: Three Kingdoms teams build an automated test harness that runs thousands of autotest matches in parallel overnight, logs per-turn facts like original ownership and army positions into a central database, and surfaces regressions within a day. A minimal combinatoric example shows roughly 550 billion states for four factions with one army each, so the system focuses on broad scenario coverage, saved-game repros from QA and forums, visualizations, and cross-difficulty runs to find surprising AI failures.
AI-written summary · about
chapters (4)
related talks
01h 18mTotal War: how AI can deliver entertainment.- Duygu Cakmak & Guy Davidson [ACCU 2018]
35mBeyond the Hype and Hyperbole
18mAutomated testing
01h 01mIntroduction to Practical AI
59mWriting, Rewriting and Re-Rewriting Armello's AI System
15mCI in the Era of AI: From Unit Tests to Stochastic Evals
39mEmbarrassing Mistakes: a short story about AI
29mBuilding a Game Engine for 300 DEFCON Hackers to Smash