Blog: Rube Goldberg Machines and AI
The original Rube Goldberg machine comics showed a series of machines that were designed to be impractical, wildly improbable but if you sort of squint you could think “this might work”. The comics themselves were split into two parts, the description of the machine and the cartoon visualizing the machine. In our paper The Narrative Logic of Rube Goldberg Machines, Mark Nelson and I do a more thorough analysis of the comics and we ultimately create a system from some of the gleaned rules that can generate the path that appears in the description albeit in not necessarily in the generated language sort of way.
This system was a good start and a good base but it is far from being a fully realized system that can create a virtual version of the comic. It is missing some pretty obvious points: there is no comic part from which much of the humor is derived, each machine has to be manually written and connected, and there is no fine control over what kind of comic is produced.
So given the problems with the system I can see three current projects that can be done somewhat independently of each other. The first is building a dictionary of objects with their built in conditions, the current system uses hand coded rules, they can be time consuming to write each time. The second is generating the placement of objects in space, this is a precursor to actually generating a comic (or a game for that matter), the system has to be able to be able to place things in a reasonable manner, to not break the internally consistent rules. The last piece is a little more vague, but there should be more interesting controls on what is generated; both giving the author more of fine control on what is output and deriving interesting rules internally to find the most relevant or interesting results. There is some interconnection between these three pieces but they largely can done independently from each other so there can be interleaving of work between the pieces.
I plan to have a series of articles that make a diary of sorts on the process of building the new components of the system. This will involve putting some ideas into practice and seeing if they work, it will also involve a fair amount of learning about new things and making a bunch of mistakes.
 You should read Rube Goldberg: Inventions by Maynard Frank Wolfe