How King Uses AI in Testing
Talkat CAST 2016 · on Tuesday, 2016-08-09· at 16:00· Fletcher Challenge Theatre (1900), Simon Fraser University Vancouver 515 West Hastings Street Vancouver, BC V6B 5K3·
summary
King uses bots that combine Monte Carlo tree search and NEAT to validate and tune levels at scale rather than restating topics. The system targets Candy Crush Saga with over 2000 levels, trains bots overnight (4–6 hours) and runs designer feedback loops in about ten minutes per level with a roadmap to under five minutes on commit using GPU and headless mode. Outcomes: faster difficulty tuning, automated crash and regression hunting, measurable perf signals. Tradeoffs: training time, engine support, ownership and maintenance, and bespoke infra and bot API.
AI-written summary · about
chapters (24)
- 0:00Intro
- 1:14QA at King
- 5:26Manual Testing
- 6:27AI
- 7:05Bots
- 7:23Why Bots
- 8:48Hardcore Bots
- 10:16Monte Carlo Research
- 12:20How It Works
- 13:58Results
- 16:02Takeaways
- 22:19Learnings
- 23:28Bot Support
- 26:33Level Design Process
- 29:12The Takeaways
- 30:41Challenges
- 32:43Questions
- 33:23Parallelization
- 36:41Criteria
- 39:29Code Coverage
- 40:13Headless Mode
- 41:17QA Questions
- 43:07Open Source Tools
- 44:03QA Teams
related talks
37mUsing Artificial Intelligence to Test the Candy Crush Saga Game
29mCandy Crush QA AI Saga
41mNot Only Cars: “AI, Please Test My App”
32mQuality engineering in the AI age: From testing AI to leveraging GenAI
29mDragonCrawl: The AI Testing Framework That Adapts Instead of Breaks
46mPositronic Brain, Your Tester
35mBeyond the Hype and Hyperbole
35mApp Testing with AI Tools – Make it Easy