(no subject)

[sigmaleph]

I was contemplating the expression "steep learning curve" and thinking I don't actually know what variables a learning curve is supposed to be graphing, and the obvious idea, something like "amount you learn per effort expended", seems like should have the exact opposite meaning. A steep curve in that graph means you get a lot of benefit from a small amount of effort. So what are they graphing?

So I looked it up and the answer is that in point of fact to the extent a learning curve is a thing it's exactly that thing and the expression "steep learning curve" in common usage means the opposite of an actual steep learning curve.

This is interesting because if I had to guess where the "steep = difficult" association comes from is from the intuitive idea that climbing a steep hill is harder than a shallow one. Compare that with potential energy curves, where thinking in terms of an actual landscape with gravity where going up costs effort is a very useful intuitive shortcut that gives you the right answers.

Comments

[sigmaleph]

It goes against general convention to put the dependent variable on the X axis (which is admittedly sometimes pretty arbitrary, in monotonous curves).

You absolutely can do it, or redefine variables otherwise for the same effect, but I do think there's a sense in which this is the most natural way to define learning curve, it's not a coincidence my guess without looking it up matched, and presumably whoever first defined the concept of a learning curve did not think "ah, people will want to talk about steep curves as difficult by analogy to hill-climbing" and thus it was not a desiderata of the definition. And once you have a literature with the function defined one way it's usually a bad idea to use the same term for a different-yet-related concept (not that it doesn't happen)