One indicator of this progression is the commoditization of many core ML algorithms. You can now pull a decision tree (or SVM, or GP) off the shelf and just us

One indicator of this progression is the commoditization of many core ML algorithms. You can now pull a decision tree (or SVM, or GP) off the shelf and just use it. GIGO still applies, as it always has in engineering problem-solving. But if you take care to do the basics, like clean your data and find good features, you can use some ML methods as components with somewhat known characteristics (performance, flexibility, runtime). It used to be that you needed a skilled person to guide you to a decent FFT or matrix library, and then core numerical software was commoditized. Same for data structures. It's fun to observe that some of this is happening to ML as well.