AI Researcher Spent 5,000 Hours on Tekken, Reached Top 0.5%
An AI/ML researcher with an MS in Computer Science spent 5,000 hours playing Tekken and reached the top 0.5% of the player base. He argues that fighting games deserve the same academic rigor as chess or Go.
Why it matters
This article provides a unique perspective on the depth and complexity of fighting games, which are often dismissed as simple 'button mashers'. It highlights how AI/ML techniques can be applied to understand and master these real-time, imperfect information systems.
Key Points
- 1Fighting games like Tekken have complex information structures and require predictive modeling, unlike simple 'button mashing'
- 2Tekken has a 'reaction gap' where human reaction time is much slower than the game's decision windows, so it's about 'reads' not reactions
- 3There is a shift from 'Legacy' players who learn intuitively to 'Newcomers' who use data, notation, and spreadsheets to improve
Details
The author, an AI/ML researcher, spent close to 5,000 hours playing the fighting game Tekken and reached the top 0.5% of the player base. He argues that fighting games like Tekken have a highly complex, real-time system that deserves the same academic rigor we give to games like Chess or Go. Unlike Chess which has perfect information, Tekken is closer to Poker where players must constantly model their opponent's hidden state and mental model. The game also has a 'reaction gap' where human reaction time is around 250ms, much slower than the 16.67ms decision windows, so it's more about predictive modeling and 'reads' than pure reactions. The author also notes a shift from 'Legacy' players who learn through intuition to 'Newcomers' who use data, notation, and spreadsheets to improve their play.
No comments yet
Be the first to comment