AI

Powering the world with fusion

In 1980, Mt. St. Helens famously erupted in one of the largest volcanic events in US history. The eruption shook Washington state with over 170 earthquakes and covered over 22,000 square miles with ash. This single explosion released over one quintillion joules - or one exajoule (1018 J) - of energy.

To put worldwide energy consumption into perspective, people use the equivalent of this historic eruption of energy every 93 minutes. That energy powers our homes, cars, appliances, farms, factories, and much more. At this rate, as the world’s population grows and our energy demands increase, we’ll need the energy of that same explosion every 15 minutes in the year 2100. Over the next 100 years, that means we'll need a total of 0.2 yottajoules - 200,000,000,000,000,000,000,000 joules (0.2 x 1024 J) - or about five times as much as humans have consumed over the course of our history as a species.

So the big question is: where will that energy come from?

We know that current methods won’t suffice. Carbon emissions from fossil fuels would cause global temperatures to rise above tenable levels in just 30 years. And while zero-carbon sources like solar, wind and carbon capture might offer reasonable alternatives, these cleaner options become prohibitively expensive when they displace the last 20-30% of fossil fuel use. For some experts, the solution is rethinking how we power our lives today, through radical research into zero-carbon, 24/7 technology that can produce electricity cheaper than the cost of burning fossil fuels.

John Platt is on a mission to do that radical research. John is the head of the Applied Science team at Google, a group that works to combine computer science with physics and biology to drive breakthroughs that help the world. In this case, that means using machine learning to help find a sustainable energy source that can power the next several hundred years on Earth. To do this, John says, “We’re going to need to change the game. We’re going to need to do something kind of crazy.”

Radical research into zero-carbon energy sources

John and his team are turning to fusion—the process that powers the sun—as that game-changing solution. Scientists have been attracted to fusion as a potential energy source for decades. It generates a massive amount of energy in each reaction and, with the right elements in play, has the potential to be an inexhaustible source of power. But decades of research have also shown us that fusion comes with challenges. Unsurprisingly, replicating the conditions of the sun is incredibly difficult. The sun’s core is so hot and under so much pressure that it’s made up of plasma, not atoms. Under these intense conditions, nuclei actually fuse together, crash into one another, then re-split into different nuclei. Plasma is notoriously unstable and difficult to keep at the right temperature. In fact, that no one in the last 70 years has managed to achieve “breakeven fusion”—the point where physicists gain more energy from the plasma than they’ve put into keeping it stable.

John and his team think AI could be the technology that helps change that. They’ve teamed up with physicists at a fusion company called TAE Technologies, who are rapidly experimenting with a state-of-the-art plasma generator they call Norman. John explains,“What we do in Applied Science is work with domain experts in universities and industry to find people who have novel ideas. Then we take our expertise in large scale computation and machine learning and help them solve their problem.” With Norman, the team at TAE can run an experiment about every ten minutes, which allows them to gather a lot of data in a short amount of time. By applying machine learning to this problem, John and his team hope to help TAE accelerate their rate of learning and, consequently, their rate of progress towards breakeven fusion.

Video courtesy of TAE Technologies, Inc.

Rapid experimentation meets machine learning

Each experiment that TAE runs is incredibly complex. To start, physicists select which of the thousands of parameters on the machine they plan to test. Then they inject gas at the two ends of a cylindrical machine and ionize it to create plasmas. Physicists accelerate the two plasmas toward each other using powerful magnets that surround the walls of the machine. When the plasmas collide, they use magnetic and electric fields to try to keep the resulting plasma stable.

Each of these steps is, of course, dependent on the physicists’ selections, but the whole experiment is further complicated by the general instability of plasma. Even in a highly specialized machine like Norman, the plasma can tilt, wobble, or bounce into the walls of the cylinder. And in these cases, the physicists need to know what happened inside the machine in order to interpret the results of the experiment. As a result, they spend a good deal of time on an exercise they call “plasma debugging”—using sparse data from the machine’s sensors to try to infer the internal state of the plasma.

This is one of the processes where John and his team believe machine learning can make a big difference. Using TensorFlow, they’re building a plasma debugger that will more efficiently and more accurately provide physicists with information about the plasma state, including inferences about its density, shape, temperature, magnetic field, and more. To do this, the debugger takes the information from the machine’s sensors and maps it against a 3D simulated plasma model. It then introduces another layer of priors (things we know to be true) based on the laws of physics and the constraints of the experiment. Using all of this information, the debugger can produce a distribution of possible plasma states for TAE to review. In under five minutes, it allows physicists to explore different scenarios, make inferences about the experiment, and move on to their next trial. This keeps the pace of experimentation high and allows physicists to learn as much as possible from each test they run.

Moving toward the future faster

While this work is still ongoing, the early results of the partnership have been encouraging: so far, the team has been able to both reconstruct the magnetic field of the plasma and estimate its shape. “Results like this might take years to solve without the power of advanced computation,” says Michl Binderbauer, TAE’s president and CTO.

TAE and the Applied Science group have developed an ambitious roadmap to build on this progress. Over the course of the next year, the teams will work to maximize the electron temperature inside Norman and verify the validity of their scaling formula up to those higher temperatures. After that, TAE plans to build a new plasma generator, inside which they hope to demonstrate breakeven fusion. There’s still a lot of work to do, but this milestone would represent significant progress in solving one of the world’s most challenging scientific problems — and serve as a signal of what could be a fusion-powered future for all.

Next story

Under the hood of the Pixel 2: How AI is supercharging hardware