4 ms·
I can see both sides of the argument. On one side, using ML feels like huge overkill when a simple trick exists. Plus AI can freak out in some circumstances. On
by s_gourichon 5y ago
I can see both sides of the argument. On one side, using ML feels like huge overkill when a simple trick exists. Plus AI can freak out in some circumstances. On the other side, it may find other, less obvious cues giving something more robust.
Rich Sutton's "bitter lesson" says the weight will move in time in favor of ML. http://www.incompleteideas.net/IncIdeas/BitterLesson.html http://www.incompleteideas.net/IncIdeas/BitterLesson.html
- rpmisms 5y agoFrom my limited experience, ML is good at massively multi-factor problems. If a human can understand the input, normal code will usually suffice. This is why ML is pretty much the only option for autonomous driving, but not for calculating credit scores.
- GeneralMayhem 5y agoCredit scores aren't a great point of comparison because they have specific explainability requirements. If your goal is to predict defaults - for instance, if you work for a bank or a hedge fund dealing in bonds - then more sophisticated ML techniques might be appropriate. But credit scores are optimized for consistency, not accuracy. I know that was probably an offhand example, but it's illustrative of the kinds of non-functional requirements that can make ML solutions more or less viable as soon as the technology has contact with human society.
- thinkharderdev 5y agoWas going to say the same. In any decision where the outcome affects a human being, "because the algorithm said so" is usually not a satisfactory answer either to the human being affected or to any regulators who have an interest.