JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Systems postmortem

A guide to testing an AI that plays a video game

You will learn how a project uses an AI tool to play a video game through a simulated spacecraft computer. It shows why developers must carefully separate strict rules from AI choices to find hidden errors.

Original by DevonanceEvaluation

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits

“DoomSat: playing Doom through a real mission stack” by Devonance. Read the original source.

This expanded guide is an AI-narrated adaptation prepared by JevMade. It expands the source’s essential ideas, examples and caveats in JevMade’s own words and is not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

Our summary

This guide reviews a project where an AI tool plays the classic video game Doom. The game runs on a simulated spacecraft computer. Developers use this setup to test how well an AI can make quick decisions while dealing with the strict communication rules of a space mission.

The system splits the work between strict code and an AI tool named Jev, which chooses from a list of options. The code handles exact rules like walking and shooting. Jev only scores which direction looks best. Later, a second AI assistant reviews the attempt and suggests better instructions.

This guide is useful for developers testing complex software, but it has strict limits. The AI still struggles to finish the first level. The author found that testing small pieces of the software hid freezing issues that only appeared during live gameplay. You must test complete sessions to find real errors.

Key takeaways

  1. Keep strict rules in your regular code instead of asking the AI to guess them.
  2. Always test complete gameplay sessions, because looking at short clips can hide major failures.
  3. Record exactly what the AI sees when it makes a choice so you can fix mistakes later.

The author provided all the numbers, test results, and corrected comparisons. No one else has independently tested this project to prove these claims are true.

Repository README · Source reviewed

Read the original guide Opens the author’s site in a new tab.