|
> ML is the one I'm most excited about at the moment. I don't deny that ML is useful (I work in the field of statistical inference), but I am bothered by the characterization of "ML" as a great invention. First off, "ML" is not really new per se; it has gone through iterations over a long time frame. Pattern recognition (via discriminant analysis, a method that is still popular in a variety of circles) was introduced in 1936. One could argue that there were statistical works addressing this at even earlier time frames - see e.g "The History of Statistics" by Stigler that focuses on linear regression, a method with at least 250 years of history. Popular methods like nearest neighbor date to at least the 1970's. Even the "revolutionary" deep learning paradigm had most of its key architectural ingredients mapped out by the 1980's or early 1990's - the main changes have been the scale at which one can operate these algorithms, tweaks of algorithms, etc due to improvements in hardware. This in turn has fed a bunch of further developments, the stage we are at right now. Essentially, what has happened is that "ML" has appropriated a variety of disciplines and techniques under its umbrella, many of which have existed for a long time. As for why concretely (i.e from an economic perspective) I lack the same level of enthusiasm as many others here, I have the following test: Amazon (or for that matter Google or your favorite search engine) has rarely been able to recommend me a nontrivial product (nontrivial in the sense of transaction magnitude) that I liked within the first few links. The best nontrivial products that I have acquired have always needed meticulous searching through a variety of diverse sources on the web (reddit, youtube, random blog posts/discussion forums including hn, etc), conversations with people, etc.
Same goes for movie recommendations, and miscellaneous other things.
This was true 7 years back, it is still true today for me. This suggests either that these problems are genuinely hard (e.g at an NP-complete fundamental level), or that we still don't know how to come up with good recommendation systems, or that companies have failed to incorporate/package techniques into a good product, or that I am just an outlier with respect to recommendation engines. |
Additionally on a theoretical level there are new things being discovered every day it seems. I mean, batch normalization wasn't even discovered until 2015.
There's a long road ahead, but I believe it will be filled with better and better inventions.