A friend asked me to review the JavaScript for a Minecraft add-on, and I suspect parts of it may have been generated with AI, even though several AI detectors said otherwise. My friend insists they wrote it themselves. The code is a large script for signal flares, including flare guns, projectiles, glow sticks, configuration menus, particles, sounds, and entity event handling. Are there any reliable clues in the style or structure that indicate whether AI was involved?
4 Answers
The best test is to ask the supposed author about specific design decisions. For example, ask why the code tracks entities in maps, how the projectile piercing works, why certain event handlers use delayed execution, or what a particular configuration value does. Someone who wrote or substantially understood the code should be able to explain it and modify it without guessing. That is more useful than scanning for stylistic tells.
At a glance, it looks like amateur or school-level code that probably evolved over time: it may work, but it is repetitive, inconsistently formatted, and not especially clean. AI can produce code like this too, especially after several edits, so the appearance alone cannot settle the question. It could also be hand-written code mixed with copied snippets.
The bigger issue is that the full source was shared without the author’s permission. If you are concerned about whether your friend wrote it, discuss it with them directly and review a small excerpt together instead of publishing the entire project. Also, whether AI was used is not automatically a problem; understanding, testing, and taking responsibility for the code matter more.
There is no dependable way to determine authorship just by looking at the finished code, and AI detectors are especially unreliable for source code. The typos, inconsistent naming, uneven indentation, short one-line helper functions, and generally cobbled-together style all look more like a person’s project—or code assembled from several sources—than a clean AI-generated result. None of those clues proves anything, though.
Functions such as the short random-number helpers and the inconsistent formatting are common human coding habits, so they are weak evidence either way.