I work in machine learning these days. I'm not biased against it -- it's literally my profession.
I'm biased against a specific category of applications that are being heavily pushed by people who don't actually understand the problem they're purporting to solve.
Put another way, the automated tools produce verifiably non-physical results nearly 100% of the time. The video there is a great example -- none of those faults could actually exist. They're close, but are all require violations of conservation of mass when compared to the horizons also picked by the model. Until "automated interpretation" tools start incorporating basic validation and physical constraints, they're just drawing lines. An interpretation is a _4D_ model. You _have_ to show how it developed through time -- it's part of the definition of "interpretation" and what distinguishes it from picking reflectors.
I have strong opinions because I've spent decades working in this field on both sides. I've been an exploration geologist _and_ I've developed automated interpretation tools. I've also worked outside of the oil industry in the broader tech industry.
I happen to think that structural geology is rather relevant to this problem. The law of conservation of mass still applies. You don't get to ignore it. All of these tools completely ignore it and product results that are physically impossible.
Incidentally, I don't even mean to pick on that video specifically. I actually quite deeply respect the folks at Enthought. It's just that the equivalent functionality has been around and been being pushed for about 15 years now (albeit it enabled via different algorithms over time). The deeper problem is that it usually solves the wrong problem.
