How Resilient Is Your AI Model?

Recently, I had the chance to catch up with my biggest mentor—an oil and gas veteran from the golden era of the '70s. Looking through his lens again made me rethink my beliefs in AI and automation.

Having worked across different phases of oil and gas fields—from exploration to the bitter end (a.k.a. end of field life)—one thing stands out: each phase has its own priorities. But rarely do we stop and ask:

?? How will what we do today impact the next phase?

Now, let’s talk AI. With today’s tech and mountains of data, we can automate a lot of things—like well optimisation. Great, right? But have we ever asked: how resilient is our AI model?

?? What critical data does it rely on?

?? Are these data sources (meters, gauges, sensors) going to be around forever?

Let’s say your AI model depends on wellhead meters. Cool. Now imagine one day your meters die. Not just die—they go extinct. Replacements of discontinued equipment are very expensive and take forever to source. Moreover, are they essential because, hey, “the wells are still flowing, aren’t they?”

So now you have an AI model that’s as useful as a drill bit with no rig. And to make things even more fun, over-reliance on automation means no one remembers how to do things manually anymore. Oops. The classic inventor’s dilemma—your own success has made you vulnerable.

So, what’s the plan?

??? Option 1: Keep those meters alive at all costs—literally. (Good luck convincing finance.)

?? Option 2: Build AI models that rely only on the most essential field equipment—the kind that will get replaced no matter what, or the most abundant and easily replaced ones. These models might not be perfect, but at least they won’t crumble the moment a sensor retires. But then are these "not-perfect" models worth the effort to build?

But before diving into either, ask yourself two critical questions:

1. Are you willing to let AI replace human skills, knowing that perfect models will be expensive to maintain toward the end of field life?
2. If the perfect model has a finite window of usefulness, can you build it fast and cost-effectively enough to make it worthwhile?

Moral of the story:

If your AI model only works in an ideal world, it’s not a resilient AI model—it’s a very fancy, very expensive house of cards.

#AI #OilAndGas #MachineLearning #ResilientAI #DigitalTransformation #FieldOptimisation