Pokemon G/S/C – Mean Look / Toxic AI oversight

Crystal_

In Pokemon Gold, Silver and Crystal, the AI (sub)layer that handles the use of the move Mean Look, will favour this move if the user (enemy) is badly poisoned, rather than if the target (player) is badly poisoned. *
This happens because this AI sublayer checks the “EnemySubStatus5_Toxic” byte instead of the “PlayerSubStatus5_Toxic” byte. This is either a programming oversight, or just really weird.

* 80% chance to favour it to be more precise, and only as long as the target is above 50% health.

Continue reading for details:

Move-specific sublayers such as this one (“AI_MeanLook”) are sublayers of the “AI_Smart” layer, which is only shared by gym leaders, elite four, and a few other trainer classes, meaning this only applies to these classes. Most generic trainer classes only account for other more basic and general AI layers.

For this video, I went with a simple set of Mean Look and a neutral damaging move (Psychic), making things much simpler to predict. Because note that the moves choosen by the AI in the video are not a 100% thing. The Pokemon games use an AI scoring system where each move is given “points” from different AI layers depending on the current battle scenario, and there is also some randomness involved. So when I say that sublayer X will favour move Y, another move Z could be outfavoured by (an)other layer(s), resulting on move Z being used instead.
For example, Mean Look may have been picked before I Toxic’d Misdreavus (between 0% and 50% chance each turn depending on the type effectiveness of the already used moves of my current Pokemon against Misdreavus or, failing that, on my current Pokemon’s type).

Source