Freedom sounds like a good idea but there's a reason why you'd want a fence near a ravine and a speed limit on a car. One might even call it common sense. The problem is that machine learning models don't typically have that and the results can be disastrous. This talk is about celebrating constraints on models in order to improve their applications.
There is a fundamental danger in blindly calling model.fit(). Maybe person.think() is a better starting point when designing an ML system. An ML model on the loose can have side-effects that you didn't anticipate and these side-effects can be more damaging than any ROC/AUC curve ever will be. I will demonstrate some tricks and share some stories of horrible mistakes and how to prevent them.
The general theme will be to remain critical. We shouldn't blindly trust models, especially if we intend to roll them out at scale.