A research group has taught AI to magnetically wrangle a high-powered stream of plasma used for fusion research — but wait! Put away your EMPs and screwdrivers, this is definitely a good thing, not a terrifying weapon for use against humanity in the coming robocalypse.

The project is a collaboration between Google’s DeepMind and l’École Polytechnique Fédérale de Lausanne (EPFL) started years ago when AI researchers from the former and fusion researchers from the latter met at a London hackathon. EPFL’s Federico Felici explained the problem his lab was having with plasma maintenance in his tokamak.

Such an everyday complaint! Yet it struck a chord with DeepMind and the two got to work.

Fusion research is conducted in many ways, but all of them involve plasmas formed at incredibly high temperatures — hundreds of millions of degrees. Sounds dangerous, and it is, but a tokamak is one way to keep it under control and allow close observation of the fusion activity happening within. It’s basically a torus or donut through which the superheated plasma travels in a circle, its path carefully constricted by magnetic fields.

To be clear, this isn’t a fusion reactor of the type you hear about giving nearly unlimited clean energy; it doesn’t produce energy, and if it suddenly started, you wouldn’t want to be anywhere nearby. It’s a research tool for testing and observing how these volatile but promising processes can be controlled and used for good.

In particular, the “variable-configuration” tokamak at the Swiss Plasma Center allows not just the containment of a ring of plasma, but for researchers to control its shape and path. By adjusting the magnetic parameters thousands of times per second, the ring can be made wider, thinner, more dense or diffuse, all kinds of factors that might affect its qualities.

The precise settings for the machine’s magnetic fields must be determined ahead of time, naturally, as the cost of improvising them badly is potentially serious damage. The settings are configured using a powerful simulator of the tokamak and plasma, which the team has been updating for years. But as Felici explained in an EPFL news release: “Lengthy calculations are still needed to determine the right value for each variable in the control system. That’s where our joint research project with DeepMind comes in.”

The teams trained a machine learning system first to predict what plasma pattern a given set of settings would produce, then to work backwards from a desired plasma pattern and identify the settings that would produce it. (Simply stated, not so simply achieved, as is often the case with AI applications like this.)

According to a paper published today in the journal Nature, the approach was a resounding success:

This architecture meets control objectives specified at a high level, at the same time satisfying physical and operational constraints. This approach has unprecedented flexibility and generality in problem specification and yields a notable reduction in design effort to produce new plasma configurations. We successfully produce and control a diverse set of plasma configurations on the Tokamak à Configuration Variable including elongated, conventional shapes, as well as advanced configurations, such as negative triangularity and ‘snowflake’ configurations.

This is important work because experimenting with plasma like this — let alone using it for power — involves lots and lots (think millions) of tiny tweaks and those can’t all be manually configured. If a theory calls for two streams, one 22% larger than the other, it might take weeks or months of work to come up with the theoretical settings to produce that using “traditional” methods (which, to be clear, are already fantastically complex digital simulations). But an AI could come up with a good match in a tiny fraction of that time, either creating the solution right there or giving human auditors a strong starting point to work from.

It also could be important for safety, since no human can improvise settings over a second or two that could contain an anomaly in time. But an AI might be able to change the settings in real time to prevent damage.

DeepMind researcher Martin Riedmiller admitted that it’s “early days,” but of course that can be said for nearly every AI application in science. Machine learning is proving to be a powerful and versatile tool for innumerable disciplines — but like good scientists they are taking every success with a grain of salt and looking forward to the next, more confident result.

Seafloor mapping today requires deploying a large, manned ship and using sonar-equipped devices and autonomous surface vessels to collect data about water depth and temperature, as well as the landscape and sedimentary makeup of the ocean floor. For offshore wind developers, this means a 200- to 300-foot, highly specialized ship that can operate 24 hours a day.

“We’ll start with a broad oceanographic survey that will include what we refer to as marine geophysical and geotechnical information,” Jesse Baldwin, head of US site investigation at offshore wind developer Ørsted, told Emerging Tech Brew.

“We’re collecting a lot of that remote sensing data.”

This initial survey usually takes about six months and the process costs tens of millions of dollars, Baldwin said. DiMare believes Bedrock’s tech can address several of the bottlenecks in the current mapping process.

Rather than waiting weeks or months for a massive ship to sail to the area you need to survey, Bedrock’s AUVs are small enough to be shipped or even checked as luggage on a flight and can be launched directly from the shore without the need for a boat.

“I’m trying to help enable that at a 10 times faster rate than what they’re doing now,” DiMare said of Bedrock’s technology. In other words, if the current process takes about six months, Bedrock aims to shrink that to about three weeks.

Medley, who co-chairs one of the interagency working groups collaborating on an effort to map the outer boundary of the US seabed by 2030, says AUVs can play a role in collecting seafloor data. NOAA already uses autonomous technology for some tasks, but Medley doesn’t necessarily see them as a replacement for ships.

"All autonomous systems complement ship activities,” she said. “And usually, you can use them as a force multiplier, to be able to achieve a broader goal, rather than just doing it with one particular platform."

And as more wind farms are built, companies will also need a way to check on the integrity of their turbines during construction and over their lifespan afterward.

“I do think that there is certainly going to be an avenue for companies like ours to use autonomous solutions in the future,” Baldwin said, with the caveat that these technologies will have to be able to meet Ørsted’s data quality standards. For now, the company still uses traditional ocean survey equipment, which is either mounted directly to the ship or towed behind it, he said.

Bedrock says it’s currently working with one customer in the offshore wind industry as well as another renewable energy customer and ocean and earth science companies, but declined to name any clients.

Although the company is focused on offshore wind, AUVs could also be used to build the maps needed for ocean carbon sequestration, laying telecom cables, underwater data centers, and decommissioning oil and gas pipelines, Bedrock says. And by getting its AUVs closer to the seabed, the company says it can use a higher-frequency sonar for these missions that is less likely to disturb marine life than the signals sent out by vessels on the surface.

Deep-sea data

Under the traditional methods, managing the enormous datasets collected by ships that don’t always have an internet connection can be slow and complicated. Even onshore, sharing terabytes of seafloor mapping data among stakeholders can be cumbersome.

NOAA makes its information publicly available, but DiMare said his request for NOAA’s data on the seafloor off the coast of New York took three weeks and involved shipping physical hard drives. NOAA told us hard drives are only used for large data requests, and even then only if a user explicitly requests the physical process rather than a piecemeal download.

Along with the AUVs, Bedrock says it is building a platform for its customers to more easily process and share the data collected about the seafloor, though it’s not technically possible to transmit the full datasets from their point of collection via today’s wireless methods.

“This will be one of the largest geospatial datasets in existence and we are collecting it in the least connected part of the world,” DiMare said. “So the challenges will exist around figuring out how to manage fleets—thousands of [AUVs]—get them recharged, and then get data back to land, where we can use cloud systems to make intelligent decisions.”

If Bedrock is able to deliver on that vision—to build a system that can efficiently provide a detailed understanding of the bottom of the ocean—it could be a boon not only to business, but also to science more broadly. The company is also involved in an international effort called Seabed 2030, that aims to map the entire ocean floor by the end of the decade.

DiMare hopes someday the world will have a fresh seafloor map each year.

“What is most exciting to me is we are going to find things and discover things and be able to do things that we’ve never been able to do before,” he said. “There’s not too many other environments, spaces, data gaps of this magnitude, where you just don’t know what we’re going to find.”