Anna Koop

I’m used to thinking of computing science theses being primarily: pose an algorithm, test the algorithm, analyze results. With some leeway, of course, but I’ve been trying to twist my research question into this framework: posing a way of using predictions for conceptual knowledge, testing the use . . .

And there we run into the first difficulty. I have a learning algorithm, that’s fine. But testing how well something does for conceptual knowledge? Turns out that’s an ill-defined question. How the tests are run is heavily influenced by assumptions about the nature of knowledge, and there is very little cross-communication between different approaches.

Thus a shift in my thesis topic.

I’m still interested in the use of predictions for conceptual knowledge, what I’ve been calling my empirical knowledge representation. But for my thesis, I need to do some thorough groundwork first.

Rob (from ASC, super helpful) used the phrase “comparing, analyzing, and synthesizing different approaches”. That’s what I need to do. I need to lay out some of the old stand-bys and new upstarts in front of me and say “look, you agree on this! And you disagree on this, because of this different emphasis or that assumption. But we can still be friends, because we both care about this similar result!”

In other words, my computing science thesis has turned into a particularly tricky kind of humanities thesis. But would it ever be helpful to succeed at this task, to bring some clarity to the murky waters of knowledge representation ACROSS ideologies.

And I will still have math formulas and experimental implementation. I’m not that far gone. But I’m tired of brushing over this lack of clarity. Let’s do some science, right at the root of the problem.