I wouldn't say simple. Someone programs it. That'll be fodder for fighting, probably forever, if introduced.
And even if the programmer really is completely neutral, there's a very real risk the AI will end up developing issues of its own. I recall reading about how one AI basically turned racist. But I suppose its goals were different than redistricting.
"AI" can have issues when the underlying data it is working with contains bias.
Amazon tried to use AI to streamline the hiring process. They had it use historical hiring data to try and figure out what characteristics on resumes resulted in a higher likelihood of getting a job offer. The idea is that they'd only interview the candidates that the AI determines possess the characteristics historically that get offered jobs, improving efficiency.
Even though this AI was not fed data on the race of the candidates,
the AI discriminated against black people. But how could an AI show bias against black candidates when it doesn't even know which ones are black? Well, a literal checkbox for race isn't the only indicator of race. Memberships or volunteer work with certain black organizations, education history, addresses, even
names work as indirect indicators of race, and the machine picked up on the fact that
these candidates got hired less often. The AI wasn't told which particular characteristics to favor, it was just looking at everything that was on resumes. Because Amazon had racial bias in its hiring history, the dataset will be reflective of that bias. So, the AI was seeing "hmm, people in these zip codes rarely get hired," and those were minority-heavy zip codes. So it would toss out future candidates from those zip codes.
The point is, you have to be careful what data gets fed to a machine learning algorithm. If that dataset has underlying issues, the AI may very well amplify them for you. With redistricting, however, I expect this shouldn't be too much of a problem. If the program only knows population numbers and the geography it has to work with, there should be a way to get an algorithm to draw the mathematically most-uniform districts possible.