Skip to main content
Exploring what’s next in tech – Insights, information, and ideas for today’s IT and business leaders

How supercomputing democratizes AI

Organizations of all sizes are benefiting from machine learning apps running on high-performance computers

Digital twin technology is the practice of creating a computer model of an object such as a machine or a human organ, or a process like the weather. By studying the behavior of the twin, it is possible to predict the behavior of the real-world counterpart, using that knowledge to address problems before they occur. Digital twin technology is undergoing tremendous change, thanks to increases in the processing capabilities of high-performance computing (HPC) technologies and the use of artificial intelligence (AI) software. Enterprise.nxt spoke with Dr. Eng Lim Goh, vice president and chief technology officer for SGI, Hewlett Packard Enterprise, about the latest developments in digital twin technology and what they mean to different industries. 

How is digital twin technology changing?

In a word, democratization. Digital twins were once limited to big government institutions on supercomputers. But because computers have gotten faster and software has become easier to obtain and use, implementations are multiplying rapidly. Here are some examples:

  • Formula One racing carmakers are creating digital twins of their cars and testing them in virtual wind tunnels. Once they have tested a digital twin in many different scenarios, they use the knowledge gained to build a prototype. Because of better software and faster processing, it’s possible to create more digital twins, faster. This is enabling Formula One carmakers to design more aerodynamic shapes right before and specific to each fortnightly race.
  • In medicine, companies are building digital twins of organs such as the heart to observe how they might respond to various interventions, thus minimizing the risk of early human trials and accelerating availability of the treatments to patients.
  • Weather organizations are building digital twins of storm systems. When weather watchers see a certain cloud formation on Doppler radar, they can feed the data into a supercomputer to see if it will result in a tornado. If so, they can build the digital twin of the tornado and observe its behavior—including where it might touch down—before the actual tornado takes shape. The goal is to simulate more quickly than real time, which will be a tremendous advance in weather prediction and public safety.
  • Climate scientists are using digital twins of the earth to predict future temperatures to understand the effects of global warming.
  • Stephen Hawking is building digital twins of our early universe. His theory that black holes evaporate is now well accepted in cosmology.

How is AI changing digital twin technology?

AI is having a profound impact. Formerly, you used established equations related to things like energy, mass, and heat transfer to build a digital twin. To predict the weather, you would feed in information about today’s weather. Then you’d apply supercomputer power with the goal to predict tomorrow’s weather faster than tomorrow is coming.

One AI approach does not use equations, but what we call deep learning. By feeding in information about the recorded past 10,000 different days, the machine can learn enough to make an accurate guess as to tomorrow’s weather. Although you would still use the equation method to check your results of the digital twin, the AI approach might help you get to a result faster and with less compute power. 

When it comes to science, this AI approach could rise, in some cases, to the same status as experiment in validating scientific theory; as HPC modeling and simulation did.

How much money can a digital twin save a business? 

In the case of jet engines, it’s many millions of dollars. A jet engine that costs, say, $20 million must go through many tests before its ready for production. One test is for a bird strike. In this test, one or more birds are put through an engine to see what damage might occur. Does a blade break off? Is it thrown out of the engine and into the fuselage of the plane where it could harm people? If you were to do this test on an actual engine, the cost would be $20 million every time you did it. By performing the test on a digital twin, you can do many tests, making changes to your engine design each time, until you are ready for a test on a physical engine. The money saved can really add up.

Here’s an example from the oil and gas industry. After making a digital twin of an underground oil deposit, you can digitally pump water at high pressure in different places to see how it affects the flow of the oil. Studying the digital twin ahead of time can tell you the right place to drill and where to pump the water. That saves millions of dollars compared with the trial-and-error approach.

Looking into the future, what new possibilities are opening up?

AI systems built using deep learning are becoming capable of unique responses to problems. Although a developer trains the machine, when the machine becomes intelligent enough, it thinks and acts in its own way. In AlphaGo for example, a computer beat the world’s top player with a move that humans had not seen before. The developer made copies of the machine and had them compete against each other, using reinforcement learning. The digital twins got stronger and stronger, so they could compete against a human AlphaGo player. The developer was asked how the computer came up with the move, and he said, “I don’t know.”

Typically one AI implementation will do one thing very well, but what happens if you combine several together? You might get a new behavior that could not be explained by the properties of each of the AI implementations. This is called “emergent properties.” Multiple AI machines put together in a certain way will emerge to be intelligent in a new way. It’s exciting and amazing.

This article/content was written by the individual writer identified and does not necessarily reflect the view of Hewlett Packard Enterprise Company.