Basically, an AI that can create, rig, and texture 3d models and game environments (by extrapolating from collections of reference models, according to user input), and that can set up physics and mechanics (assuming that the AI has access to a 3d modeling studio and a game engine, both designed for compatibility with the AI, or as a component of the AI), all according to user commands (and allowing for tweaking and optimizations of models, rigging, mechanics, etc, by the user).

An example of user commands would be something like: "Gaming AI, create a casual style* male model, European build, 6'5", fit and slightly skinny, with red scaly skin, green eyes, a reptilian tail, demonic wings, claws, sharp teeth", etc. The user probably wouldn't add all of these characteristics at once, but rather one at a time, tweaking each feature via AI commands or manually.

*"casual style" is a fictional "style class". Style classes would refer to the visual style of the models. Possible example styles include "cartoon", "abstract", "gothic", "steampunk", "serious" and "realistic".

Here's another example of user commands, for a environmental model: "Gaming AI, create a serious style house, Victorian, two story, white with beige trim, with porches and shutters. Give it a creepy aesthetic." Again, models could be created and modified or have features added in a step by step process, in order to tweak and refine them.

I believe that such an AI would significantly reduce the amount of time, labor, and difficulty involved in designing games; making games cheaper and easier to produce, and making game design available to everyone. A variation of such an AI could also be used to create 2d artwork and animations.

But is such an AI even remotely possible? And would it take a supercomputer to run the thing? (I'm under the impression that such an AI would need to be capable of learning and adapting, and would require a massive and expansile "association library"*—including 2d and 3d models, and verbal and textual speech—as well as near human intelligence)

*if the term "association library" doesn't exist, or doesn't currently relate to AI, Then I just made it up. According to my made up definition, an association library is the library of programmed or learned associations that an AI uses to generate responses, and to, in this context, generate 3d models; and probably to write or select code as well, in order to set up physics and mechanics and the like.


2 Answers 2


In the history of programming the productivity of the programmer has increased. Early MS-DOS base games were programmed in Pascal and Assembly language direct for the CGA graphics card adapter. With the upraising of the C/C++ language and standardized operating systems, it become possible to use software libraries. (1) In early MS-DOS period, a team of programmers was needed for a simple jump'n'run game, while nowadays a single programmer can create such a game with existing game-engines in one weekend.

It is obvious to think this development forward and imagine a much advanced technique which support automatic programming. Creating games in a point&click fashion is possible with so called game construction sets. That are games, which comes with a level editor and a Lua scripting engine for modification of existing content. This is not directly related to Artificial Intelligence but goes into the same direction. The idea is, to use computer programs for increase the productivity. In the context of programming such AI-related support is hard to implement. Because programming and game-design contains lots of domain specific knowledge which has to be formalized. If an AI system should support the programming itself, the AI has to parse existing Stackoverflow threads and give hints to the programmer how to implements this in his software. So called “agent-based software engineering” is researched since the 2000s.

The most advanced example in semi-autonomous game design is perhaps the RPG adventure generator described in Barros, Gabriella Alves Bulhoes, et al. "Who killed albert einstein? From open data to murder mystery games." IEEE Transactions on Games (2018) It can parse Wikipedia article to generate a playable game from scratch.


I realise this question was asked a couple of years ago, but I think it's worth reflecting that the OP asked an insightful question, as indeed the gaming industry is now actively working on the questions and ideas raised by the OP.

Indeed, huge sums of money are now being invested in the AI field within the gaming industry, given the increase to productivity it would produce.

You might find this YouTube video interesting, which discusses the topic of producing game elements from description, and ray-tracing techniques. It's not produced from a AI research/developer-perspective, but certainly a useful overview of where things stand mid-2019



You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .