Stealing Baseball Signs with a Phone (Machine Learning)

Mark Rober has many great and some good science and engineering videos that my kids love: link He get's my vote for the current best communicator of science, on par at times with Carl Sagan. This video explains neural nets/machine learning and had my kids riveted: ▶ this video We've gone down some deep lego rabbit holes, such as the great ball contraption: link These are pretty light but entertaining: link link The latter is a guy that builds fresh water aquariums. The fact that he can do this full time for a living (via YouTube) is one of the most amazing and unlikely results of the internet (IMO).

A quick How To guide with links to source, for those of you with baseball teams who are behind the times: ▶ this video

The cheating aspect is using any outside technology. A runner on second isn't forbidden from trying to steal the sign and communicate it to the batter. However you aren't allowed to have someone watching live video, using binoculars to watch the catcher from the outfield, or any other technologically aided approach to stealing signs. Plus teams often do "encrypt" their signs. That is why you might see a third base coach give 10 different signals. Only one or two of them are needed to convey the instructions. The rest are just used to hide the true signal. However it is now trivially easy to decrypt these signs using machine learning. Here is a video on it.[1] [1] - ▶ this video

Here's a fun (semi) technical video on using machine learning for sign stealing: ▶ this video I wonder if the signs employed by Major League teams are as simple as the ones dissected above?

I'm not sure if the complexity of the codes is significantly higher in the pro teams, but I found this video on baseball code decoding pretty fun. ▶ this video Combined with this, it certainly seems like there's potential for a fully automated pipeline