Artificial intelligence (AI), machine learning (ML) and high-performance computing can help the energy industry make better decisions through simulations, but AI and ML are only as good as the data and models driving them.

Simulations of “everything” are on the horizon, and quantum computing will have more of an impact on how science is done, Bill Vass, vice president of engineering at Amazon Web Services (AWS), said on March 10 during a Voices of Innovation session at CERAWeek by S&P Global.

“Computers are really stupid. People think of them as smart, but they're actually quite stupid. And they take a lot of data,” he said.

Training a robot to pick up a glass, he said, would require the robot to take 107,000 pictures of the glass to build up a very dense mathematical model of the object, while a 2-year-old child could be easily trained to pick up that glass.

“Remember, machine learning is just math, right? It’s not thinking,” he said. “It’s very good at understanding very complex systems and seeing trends.”

Humans, on the other hand, can build complex things like road systems, yet not understand it in the aggregate, such as how traffic occurs.

Harnessing machine learning

To harness the power of ML, Vass said it’s important to prevent drift, or dissonance, in the modeling.

“It's very easy for machine learning models to drift,” Vass said. “You can get drift in your model as you feed data and do model evolution.”

Drift can also occur due to a lack of density in the model.

“That's why at Amazon we also do a lot of synthetic data generation. Because it's not possible for me to take 107,000 pictures of that [glass]. So I'll simulate those pictures and feed them into a machine to build a dense model,” he said.

Simulations aren’t new.

“We’ve done reservoir simulation for ages in the industry,” said Vass, who started his career in reservoir simulations and once worked as a doodlebugger in oil and gas exploration. Now, he added, everything can be simulated. “At that time, it was a very cool thing.”

Simulations are critical because some things aren’t practical to test in real life. For instance, he said, autonomous systems companies might simulate 15 million miles of driving in the cloud.

“You could not physically do 15 million miles of test driving” because it would require thousands of cars and thousands of people driving to hit the 15-million-mile milestone, he said. “It's just not economic, it's not practical. And it really slows your development cycle down.”

AWS SimSpace Weaver shares memory across multiple machines to make it possible to create complex 3D spatial simulations, he said.

Vass said AI and ML are “just like any tool. It can be misused, and it can be used properly, and you just need to understand what you’re using.”

Otherwise, the result could be “artificial stupidity.”

Over time, Vass said, quantum computing will “change the way we do imperial sciences, and that’s really the future.”

In fact, Amazon has invested in quantum computing because it can help the company “save a few percent on logistics” which “is billions of dollars for us.”