If the experts are wrong, and bandwidth growth (or CPU growth or memory growth or whatever) slows or stops in ten years, then fine: change the largest-block-I'll-accept formula. Lowering the maximum is easier than raising it (lowering is a soft-forking change that would only affect stubborn miners who insisted on creating larger-than-what-the-majority-wants blocks).
Lowering the limit afterward wouldn't be a soft-forking change if the majority of mining power was creating too-large blocks, which seems possible.
I think that a really conservative automatic increase would be OK, but 50% yearly sounds too high to me. If this happens to exceed some residential ISP's actual bandwidth growth, then eventually that ISP's customers will be unable to be full nodes unless they pay for a much more expensive Internet connection. The idea of this sort of situation really concerns me, especially since the loss of full nodes would likely be gradual and easy to ignore until after it becomes very difficult to correct.
As I mentioned on Reddit, I'm also not 100% sure that I agree with your proposed starting point of 50% of a hobbyist-level Internet connection. This seems somewhat burdensome for individuals. It's entirely possible that Bitcoin can be secure without a lot of individuals running full nodes, but I'm not sure about this.
Determining the best/safest way to choose the max block size isn't really a technical problem; it has more to do with economics and game theory. I'd really like to see some research/opinions on this issue from economists and other people who specialize in this sort of problem.