Show HN: Searching for optimal strategies in a puzzle game with chance (Drop7)
This article explores the technical challenge of developing an autonomous AI strategy for the puzzle game Drop7. It details various computational approaches, including expectimax search, neural networks, and evolutionary optimization.
Why it matters
Applying advanced machine learning techniques to game theory provides a sandbox for testing AI decision-making capabilities in complex, high-stakes environments.
Drop7 is widely considered the greatest puzzle game of all time [1] . The goal of this research is to find an autonomous strategy for playing the game as well as a human can.
An experienced human player can score a million points in this game with the right long-term strategic thinking. The strongest research candidate so far ( learned lookup tables inside a depth-4 search ) averaged 516,155 points over 512 games it had never seen, where the same depth-4 expectimax with its hand-written leaf averaged 382,567 on those same games. Both are well under a million, so the research problem is very much still open.
New here? Learn the game , then the ideas behind the strategies , then see what has been tried and what is still open .
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in