I have been working on a project that might benefit from AI, but I am a complete novice. Please advise. We are interested in looking at voter data at the level of congressional districts (number of voters, party affiliation, etc.) and census data from the boundaries of congressional districts (type of employment, family size, education level, income, etc.).
We would like to create visual representations of hypothetical conditions wherein certain groups of people could be moved from one area of the country to another. We would like to show a map of the U.S. with Electoral College votes by state, and be able to see how it would change if subsets from various Congressional districts moved to districts in other states.
Is this an appropriate project for AI? What advice would you have for an AI learner who wants to focus on this project?
I think the SOTA for gerrymandering is already quite advanced, but I don’t know if any of the algorithms used for this have ever been published. That might be worth a bit of googling. My first reaction is that this is more of an optimization problem than a pattern recognition problem, so it’s not clear AI is the right approach. There are a number of well known techniques for different types of optimizations in areas like Operations Research. A quick google turns up this textbook. It might be worth scanning the ToC of that book and then googling up the Wikipedia pages for all the algorithms that they cover to see if any of them sound “on target” for the type of problem you are trying to solve.
How much effort have you put into understanding the nature, structure and completeness of the data that you are proposing to use for this project? Seems like that’s pretty critical to understand before you launch into this for real.
I’m struggling to see the AI angle in this task.
The “… and see how it would change” goal seems like a straightforward data analysis task.
Generally for AI to be applicable, you need some sort of condition you are trying to optimize, and a dataset of examples that provides a basis for learning relationships within the data set.
Some type of Generative Adversarial Networks (GANs) might be useful (the projects that create images or text based a large collection of examples). But you still need to have that large set of examples. I don’t think a big table of where people live will do the trick. And that isn’t an introductory level project.
Thankyou for the input. There is a lot for me to consider here as I begin to move forward.