Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

"Steep Learning Curve" is an Erroneous Metaphor

yann.lecun.com

Post reply on HN
Print view View original

11–20 of 26 posts

Re: "Steep Learning Curve" is an Erroneous Metaphor

#11

Post by notsosteep »

ckuehne wrote: post #5
pflats wrote: post #4

I feel like the graph is a poor choice in illustrating this idea. As it's central to the argument, the rest of the article falls apart, in my eyes. First - why do both graphs start the user with 0% performance? Humans come into a new program/context/etc with prior experience. Put a child who's used Word before in front of ed, and he'll probably do well. Put him in front of vim, and he'll probably struggle to write a…

"I feel like the graph is a poor choice in illustrating this idea. As it's central to the argument, the rest of the article falls apart, in my eyes." What, in your eyes, does the word "curve" refer to then?

I've always pictured a steep learning curve as this - http://imgur.com/8Wfih. One doesn't learn much at first until the "aha" moment. So I think the metaphor is still correct.

Re: "Steep Learning Curve" is an Erroneous Metaphor

#12

Post by rimantas »

Metaphor (if it is really a metaphor) is OK for me. I always understood it this way: time on X, stuff to learn (not some performance) on Y. Somewhere on that Y is the threshold point and by passing it you cross from "still learning" to "doing actual work" area.

I admit, time on X may look ambiguous but I kind of look at it as "some reasonable time interval".

That way something complex will require to learn more going from 0 to threshold point and hence steeper curve (line) than simpler stuff.

Re: "Steep Learning Curve" is an Erroneous Metaphor

#13

Post by richardw »

The original meaning of a steep learning curve was exactly as he says. It's been changed in popular use from "easy" to "hard".

(Rummage rummage rummage...):

"Early uses of the metaphor focused on the pattern's positive aspect, namely the potential for quick progress in learning ... Over time, however, the metaphor has become more commonly used to focus on the pattern's negative aspect, namely the difficulty of learning once one gets beyond the basics of a subject."

http://en.wikipedia.org/wiki/Learning_curve

"One natural interpretation of such a curve, which was the predominant early usage (according to Wikipedia) and still exists in some technical circles, is that the thing being learnt is easy — a great amount of learning happens in a small amount of time. This is the opposite of the popular usage."

http://english.stackexchange.com/questions/6209/steep-learni...

"The phrase everyone gets wrong. Outside experimental psychology, where the term originated, I have never seen a correct usage. Learning curves show performance (e.g., percent correct) as a function of amount of training (e.g., number of trials). A steep learning curve means the organization, person, or animal quickly went from low to high performance — in other words, learning was fast."

http://blog.sethroberts.net/2008/08/18/steep-learning-curve/

alt.english.usage:

"The learning curve for bridge is not steep - on the contrary. But I have seen other examples of this confusing of "steep curve" with something that is hard to learn. The association is probably that it is hard to get uphill, but the true meaning is that with a steep learning curve a given effort will take you high up the curve and thus corresponds to something that is easy."

https://groups.google.com/forum/?hl=en#!topic/alt.usage.engl...

Definition:

"a graphic representation of progress in learning measured against the time required to achieve mastery."

...

'The phrase "steep learning curve" is sometimes used incorrectly to mean "hard to learn" whereas of course it implies rapid learning.'

http://dictionary.reference.com/browse/learning+curve

Re: "Steep Learning Curve" is an Erroneous Metaphor

#14

Post by karamazov »

Steepness doesn't have to refer to a graph of time versus skill, and at any rate the metaphor certainly predates machine learning curves. As others have pointed out, it's meant to evoke the difficulty of climbing a steep hill versus one that's flatter. This person is mixing concepts and etymologies amidst an insufferable air of superiority.

Re: "Steep Learning Curve" is an Erroneous Metaphor

#15

Post by ckuehne »

richardw wrote: post #13

The original meaning of a steep learning curve was exactly as he says. It's been changed in popular use from "easy" to "hard". (Rummage rummage rummage...): "Early uses of the metaphor focused on the pattern's positive aspect, namely the potential for quick progress in learning ... Over time, however, the metaphor has become more commonly used to focus on the pattern's negative aspect, namely the difficulty of learni…

Thanks. I had seen the Wikipedia entry (but not the stackexchange and the Seth Roberts post) as well but did not post it because I found it interesting to see everyone come up with rationalizations for why their usage of the phrase is 'the correct one'. :)

Re: "Steep Learning Curve" is an Erroneous Metaphor

#16

Post by bradleyland »

sixtofour wrote: post #10

I understand the objection, but like most idioms the intent is understood, and like most idioms I've always accepted this one for its intent. However, you can make it work logically. Look at his graph and imagine the two curves as hills, both going to the same altitude. X is time, Y is altitude. You'd have to work much harder in a short time to get to the same altitude as the shallow hill.

And unlike a machine, where the y-axis learning value is assumed to be achievable by any machine of a specific class, humans have a much higher degree of variability. Some will not reach the top of the curve where things level off. Thus a steep learning curve means that a greater percentage of individuals will give up as a result of the more compressed learning requirements.

Re: "Steep Learning Curve" is an Erroneous Metaphor

#17

Post by billswift »

lilspikey wrote: post #2

Or you have your axis the wrong way round...

That's what I noticed. If you label the horizontal axis for the amount learned (more or less equivalent to his "Performance" that he has on the vertical axis) and the vertical axis for effort required, it would fit standard usage just fine.

Re: "Steep Learning Curve" is an Erroneous Metaphor

#18

Post by billswift »

simonhamp wrote: post #3

What happens when you break "learning" down into specific tasks? Is the compound learning of lots of simple tasks accurately represented by the generic "learning" graphs for complex tasks? Steep may not be the correct word. Complex may be better.

That is actually the complaint I have seen most often with command lines - you need to learn so many commands at the very beginning to do anything useful; thus producing a steep curve. Once you have the absolutely necessary basic commands down (cd, pwd, ls, mkdir, chmod, and so on) you can easily add more commands gradually, as you need them.

Re: "Steep Learning Curve" is an Erroneous Metaphor

#19

Post by richardw »

ckuehne wrote: post #15
richardw wrote: post #13

The original meaning of a steep learning curve was exactly as he says. It's been changed in popular use from "easy" to "hard". (Rummage rummage rummage...): "Early uses of the metaphor focused on the pattern's positive aspect, namely the potential for quick progress in learning ... Over time, however, the metaphor has become more commonly used to focus on the pattern's negative aspect, namely the difficulty of learni…

Thanks. I had seen the Wikipedia entry (but not the stackexchange and the Seth Roberts post) as well but did not post it because I found it interesting to see everyone come up with rationalizations for why their usage of the phrase is 'the correct one'. :)

We use it 'wrongly', and will continue to (unless you want to not be understood), but hey - it's interesting that somehow it got mixed up! I wonder who got it backwards first?

Similarly, Joel has an article on Hungarian Notation that says common usage is wrong. Would those who defend the (current, wrong) usage of "steep learning curve" not find it interesting that the proper H Notation was a lot more useful than current usage?

http://www.joelonsoftware.com/articles/Wrong.html

Post reply on HN