Neural networks have seen some limited success in Super Smash Bros. Melee. Examples include Vlad Firoiu’s Philip and Bryan Chen's Project Nabla. To give the reader an idea of how successful these projects have been at replicating human play, when I first saw videos of Project Nabla, I was convinced that it was a human player intentionally playing in suboptimal ways (e.g. a mix of human-like movement with distinctly unhuman "forced" errors). I now accept that it was indeed an AI, and having had the opportunity to play against it, I'm thoroughly impressed by the project as a proof of concept.

My hope is that advances in state-of-the-art game-playing AI will eventually lead to the creation of general game-playing AI agents which can be applied to any video game and which are intelligent and capable enough to be "fine-tuned" on a relatively small sample of data (e.g. a few hundred/thousand replays of a given player playing a given matchup).

Resolves to YES if, before EOY 2029, I am able to trick my most common practice partner into believing that he is playing versus me when he is actually playing versus an AI. I will specifically use matchups that we play frequently and in which my playstyle is particularly distinct from the average player, for at least 5 rounds (referred to as "games" by Smash players).

Resolves to NO otherwise.