With the recent launch of Malawi (available as a native Android app and as a Human vs. Human WebAssembly demo right on the games website, I’ve been closely watching how players interact with the game. While the rules of this 1986 classic are straightforward, making the computer opponent feel challenging, fair, and strategically sharp is an ongoing effort.

In this post, I want to share a quick update on the latest AI improvements and what's happening under the hood.

What’s New in the Latest Update

The main AI opponent in the Android app has received a major upgrade powered by a brand-new version of the OptimalBaseBoardEvaluator. Here are the key technical enhancements included in the latest build:

Alpha-Beta Engine: Faster and deeper AI calculations with optimized move ordering, allowing the computer to look further ahead without slowing down game performance.

Tapered Board Evaluation: Dynamically adjusts positional weights as the game transitions from the opening into the endgame, leading to significantly improved tactical accuracy and better endgame survivability.

Refined Tactical Safety: Enhanced threat-detection logic for 1-step baseline threats and immediate move opportunities, eliminating previous blind spots.

Everything continues to run 100% locally on your device without relying on any cloud servers, keeping the app lightweight and fast.

Behind the Scenes: Benchmarking Other Evaluators

In addition to the main engine, I have been testing specialized evaluators behind the scenes to benchmark and measure AI performance:

CenterDominanceEvaluator: Focuses heavily on controlling central board space to force more dynamic mid-game scenarios.

SniperTrapEvaluator: Specialized in recognizing long-range control patterns and setting tactical traps.

Note: These additional evaluators are currently strictly used for internal evaluation and benchmarking purposes. They are not yet selectable in the public release, but the insights gained from them directly feed into refining the core AI.

Give it a Try!

If you want to challenge the upgraded AI, make sure to grab the latest version of Malawi on Google Play. If you'd rather analyze moves or play a quick local match with a friend, you can also check out the WebAssembly version (Human vs. Human) directly in your browser on the project page.

As always, feedback on the AI's playstyle and difficulty balance is very welcome!