Uncommon Descent Serving The Intelligent Design Community

Machine Learning or Curve Fitting?

arroba Email

The term “machine learning” has started to be thrown around everywhere. However, I just don’t get excited about it, myself.

Machine learning is literally just another name for curve-fitting. That’s all it is. Curve fitting has always been very useful for establishing patterns where the underlying mechanism is unknown, and I’m glad that we have automated the curve-fitting process, and developed a number of techniques for it. But let’s be real. The hype is ridiculous. 99% or greater of what is passed off as “machine learning” is really just glorified curve fitting.

If you see someone say something about “machine learning” on the news, pretty much every time they just mean that someone ran a curve-fitting system on data for which we don’t have underlying physical means of projecting.

Now, the funny thing is, the people doing machine learning have generally convinced themselves that this is not what they are doing. But if you look at the components, it is just a stochastic curve-fitting technique which can make use of non-linear components.

Again, it is useful in its context, but we’ve been doing curve-fitting for hundreds of years, and it wasn’t until the last decade that people have been writing articles about it like it was a brand-new idea.

The question is, is the mind doing something more than curve fitting? If so, then the mind is not a neural network. More broadly, all of AI is objective function minimization. Is that what human intelligence is doing? Human intelligence defines the objective function in the first place, and there is no objective function defining objective function. So, human intelligence is intrinsically different from AI. Finally, if AI is the best materialistic definition of the human mind, yet human intelligence is intrinsically different from AI, then the human mind is immaterial and the supernatural exists. EricMH
Mung @ 5 - Neural nets are one form of machine learning, so there is at least a connection. Bob O'H
Timely clarification of a general misunderstanding. Thanks Dionisio
Neil - As I said - I'm all for curve fitting. But the hype behind what we are doing today is ridiculous. If I told someone - "hey, I've got a curve-fitting system here," no one would care. If I tell someone - "hey, I've got a machine learning system here," then I can sell newspapers ;) People think that this is some brand new feat of intelligence and that the machines are soon to take over. But really we are just doing basic curve fitting. It's cool that it is becoming more usable by more people, I just don't think we should get carried away with ourselves. johnnyb
It's not as if machine learning has anything at all to do with artificial intelligence. Mung
I'm not quite sure of the point here. Personally, I don't see machine learning as much like human learning. However, let's not forget that Kepler's laws of planetary motion came from "glorified curve fitting." Neil Rickert
Yep, I'm with you 100% on this. What happened was computer scientists tried to do statistics, without first talking to statisticians. To be fair, I don't think machine learning is 100% useless, but I think it is being over-hyped. There was a joke going round twitter a couple of months ago to the effect that you don't need machine learning, just linear regression and a sensible observation model. Bob O'H

Leave a Reply