Can a game be generated by AI?
Recently, we have been witnessing some of the most groundbreaking demonstrations in the field of AI arts. Some of the contemporary AI applications are capable of producing stunningly convincing pieces of art (both visual and auditory), which are clear signs of job insecurity for those who are working as professional artists.
This may be a welcoming news for indie game developers who are seeking every opportunity to develop their in-game arts without having to hire artists. They too, however, do not seem sufficiently safe from the upcoming technological Armageddon; it seems just a matter of time until the entire pipeline of the videogame industry becomes fully automated.
Conceptually speaking, it is not hard to imagine an AI system that is designed to auto-generate a game. A game is essentially a combination between a set of situations and a set of alternative choices which lead to each of those situations. Each "situation" can be represented as a vertex and each "choice" as an edge of a graph (in mathematical terms).
The simplest possible game would be a graph which comprises 3 vertices (representing the game's "start", "win", and "loss", respectively) and 2 edges (representing the game's "winning choice" and "losing choice", respectively), where one of these 2 edges connects the "start" vertex to the "win" vertex and the other one connects the "start" vertex to the "loss" vertex. The availability of 2 alternative choices (one that wins the game and the other one that loses the game) technically makes this graph a game, as opposed to a movie or novel in which everything happens in a linear fashion.
Obviously, this is just an abstract conception of what a "game" is, but one can easily expand upon this idea by letting the graph grow itself like an organism by splitting its vertices recursively, grouping them into semantic clusters based upon their positions in a feature space, and so on.