Artificial Intelligence: The 21st Century Horseless Carriage*
Data Science & Analytics

Artificial Intelligence: The 21st Century Horseless Carriage*

Rosemary Jackson  •  

"It is only a question of a short time when the carriages of every city will be run by motors… A great invention which facilitates commerce, enriches a country just as much as the discovery of vast hoards of gold." —Thomas Edison 1895 Uses an advanced computer technology that can be trained, referred to as Machine Learning, ML, or more broadly, Artificial Intelligence, AI. Uses a special version of domain-based AI that applies geomechanics, productivity prediction and drainage modeling, which is really a fancy calculator.

AI: This program can figure out what makes a group of things similar and like the horseless carriage will become the self-propulsion of a new technology, that in a short time will be as ordinary as a car.

Artificial Intelligence is not really intelligent. You are intelligent. AI is actually an incredibly complex calculator. And you can’t just lift the hood and look underneath to understand. For discerning scientists and engineers that’s just not acceptable. Succinct answers are tough, and AI is not a magic tool. It needs to be verified, to be accountable, to be in context, to be interrogated. But how do we do that and be sure?

1. First ask, is the AI domain-based?

Dr. Troy Ruths, CEO of weighs in, "Look at the kind of AI or machine learning being offered. Does it solve a specific problem? The only scalable AI is domain-based AI, artificial intelligence focused on a single subset of cognitive abilities.

“There’s a lot of science sales talk around the capabilities of generalized AI, the ability of a machine to apply knowledge and skills in different contexts. Generalized AI doesn’t work well yet. You cannot give it to someone who doesn’t understand AI or ML and actually make something with clear outputs. You have to have taken a real class on AI and ML to get something out of it.  

“The computer scientists that have taken those heavy AI and ML classes are now applying it, with oil and gas engineers, in the creation of You need the domain experts and the computer scientists working together to make AI functional in any industry.”

Screenshot of economic development for the North Midland Basin (oil and gas)

2. Then ask, what is the model trained on?

“How do we explain how the model works?” Ruths adds. “Typically, you explain what you’ve trained it on. Then the next question is usually about extrapolation. How far will this go? How far can I push the boundaries? Most of AI is sophisticated interpolation math in this highly dimensional feature space. What it’s doing is taking the input data and trying to extrapolate all the potential feature sets that you could sample from. The further you get away from your input data the worse your predictions get. The way we’ve gotten around that with the AI we’ve written, is that we’ve imbedded it with physical limitations. Those physical limitations make sure that our extrapolations are fair and good.

“Inside there are points where you can verify those assumptions graphically or through characterization. For our stress model, you can go see our grids in 3D. That’s how people would rapidly verify it. They would look at an actual result. Then you could look at something we didn’t show the system and see how accurate it is. Accuracy of the output is what it’s all about.”

3D model of subsurface unconventional oilfield

3. Make sure there’s testing, testing, testing

“In terms of the AI we’ve developed, the certainty is based on a lot of testing,” Ruths continues. “We did all the math. We did all the modeling. And we got the predictions. Our certainty as the creator of it, is the testing. Even the people that build these machines realize they’re so complicated and sophisticated that you need to test it for all the different inputs.

“Although, we can test fewer inputs because we know that it was designed in a consistent way to meet the constraints of geology, geomechanics and reservoir engineering. Because of that we’re not going to give you things that can’t exist. But you also don’t want to overfit to your data. You don’t want your model to be so tuned to your data that when you show it something new you can’t make a good prediction.”

Screenshot frac hit well summary in the Permian basin, Shmax, offset well sticks

4. Last question, is it reproducible?

Dr. Brendon Hall, VP of Geoscience explains, “This gets back to reproducibility. It would be hard to verify an AI if you can’t reproduce it. Imagine a prediction engine that doesn’t give you consistent results based on those input sets. One of the issues is, to be reproducible you need to assemble all the inputs that go into the configuration.

“That’s why the well spacing is so profound. The input set is clean and simple, and you can get reproducible results easily. When people say it’s intuitive or easy to use, that’s another way of saying that you can make the same input set as another person with the same confidence.

“ is doing what I’ve been wanting to do for a long time which is to bring advanced AI to the geoscientists and engineers that are making these decisions. It has been the domain of R&D. It has been the domain of ML data scientists and engineers who know how to code and know how to use GPUs and launch all these things. But it hasn’t been practical to use to make business decisions. And now we’ve connected geology to economics in this intuitive workflow that they can use to get these answers in a reproducible fashion.”

About the cover image:

Alexander Winton, the early inventor of the horseless carriage, writes in the April 15, 1911 issue of the Saturday Evening Post:

“In spite of my banker’s displeasure, I went ahead with my model and finally had it in such shape that I thought it would run. All I needed to finish the job was a set of tires. I went to the Goodrich Company, in Akron, and told them I wanted something bigger than their biggest bicycle tire, something that would fit the wheels of a horseless carriage.

“That’s a new one on us,” cried a man to whom I had been directed. “A horseless carriage, eh? Hmph! Will it run?”

“You bet it will.”

“Well, I guess we can make them, although we never have.”

“That’s fine.”

The man hesitated, rubbed his chin, and observed: “We will make them, but you will have to pay for the molds.”

“Do what?”

“Yes, sir. There won’t be enough call for tires for horseless carriages, and we can’t afford to pay for the molds. Also, you will have to pay for them in advance — and the tires too. We’ll have them on our hands if you don’t get them.”

I paid… And built the first commercially successful American automobile.”

114 Main Street Ste. 400
Houston, TX 77002
4-5609 Avenue du Parc
Montreal QC, H2V 4S8