In the race to create and contain carbon-free nuclear fusion energy, Alphabet’s artificial intelligence lab DeepMind is the latest contributor.
In collaboration with the Swiss Plasma Center at EPFL — a university in Lausanne, Switzerland — the DeepMind AI has applied its algorithms to control the plasma inside the nuclear fusion reactor, that’s hotter than the sun’s surface and maintain its temperature long enough to take energy out of it. All of which has been near impossible so far.
Before going any further with the scientific lingo, let us break it down for you:
What is nuclear fusion
The best and easiest understandable example of nuclear fusion energy is the sun. The process of nuclear fusion generates heat in the sun.
Creating nuclear fusion energy in laboratories has proven difficult as it consumes far more energy than it produces, making it useless as an energy source at a large scale.
Existing nuclear power stations work on nuclear fission reactions that create energy by splitting atoms; a nuclear fusion reactor works exactly the opposite; it releases energy by combining atoms.
Recently the Joint European Torus (JET), a fusion reactor based in the UK’s Oxfordshire, produced 59 megajoules of energy, equivalent to 11 megawatts of power, over a five-second period.
The scientists built a process that allowed for the self-heating of the matter when it is in a plasma state, using nuclear fusion, which could represent a major step towards nuclear fusion.
According to the Independent, the scientists took the hydrogen isotopes deuterium – which can be found in seawater – and tritium, made in a reactor. They used the hydrogen isotopes to create a burning plasma.
In short, the researchers were able to compress and heat a plasma, which will then be heated by the reactions themselves, allowing the energy to sustain itself.
Due to huge gravitational pressure in the core of the Sun, nuclear fusion is possible at around 10 million Celsius temperature. Since creating such pressures on earth is impossible, temperatures need to be much higher – above 100 million Celsius.
Since no material can withstand such temperature, fusion is achieved in a super-heated gas, or plasma, held inside a doughnut-shaped magnetic field.
The problems with harvesting nuclear fusion energy
Even though scientists have created nuclear fusion energy, it still faces an engineering challenge – to heat the plasma and hold it together to take energy out of it.
The process to confine and control the plasma can take up more energy than what is produced from it, thus making it a counterproductive process.
To put it into perspective, the recent experiment at the JET lab produced enough energy to boil 60 kettles of water for five seconds. Yet it was considered a breakthrough.
Researchers have tried to confine nuclear fusion reactions and nudge them into different shapes that may yield a maximum output with the help of powerful magnetic coils.
However, while doing so, they have to prevent the plasma from touching the walls, which would damage the walls and waste the heat. Thus, slowing down the nuclear fusion process.
What is DeepMind?
A division of Google’s parent company, Alphabet, DeepMind, is responsible for developing general-purpose artificial intelligence technology. The technology takes in input and learns about it from experience.
DeepMind claims that its system is not pre-programmed: it learns from experience, using only raw pixels as data input.
To put it simply, earlier, it was used to learn and play games on its own. When tasked to beat the library of Atari games, it learned to understand the games, and with time, the AI could play the games better and with more efficiency than humans.
The AI made headlines in 2016 when its AlphaGo program beat Lee Sedol, the world champion of the game “Go,” in a five-game match.
After proving its might in playing video and board games, the AI has been used in healthcare, identifying eye diseases and kidney injuries, creating computer programs, and making strides in nuclear fusion research.
DeepMind’s contribution to nuclear fusion research
As per a report by Business Insider, the artificial intelligence lab and its co-researchers trained an algorithm on the Swiss Center’s simulator to hypothesize by itself how best to control the magnetic coils through the use of reinforcement learning.
This is where algorithms are effectively “rewarded” for achieving strong outcomes.
It developed an architecture to maintain the plasma, sculpt it into different shapes and maintain separate plasma simultaneously.
The algorithm was then applied to the real-life tokamak – a doughnut-shaped vacuum chamber with metal coils and a huge magnet to create sun-like conditions on earth.
The report said that the AI manipulated the magnetic field for over two seconds, just as it had done in the simulator.
“This work is a significant step in our understanding of how we might design new tokamaks that incorporate AI and going forward, we expect to see increasing sophistication in the use of reinforcement learning in the field,” said Ambrogio Fasoli, director of the Swiss Plasma Center, as quoted by Business Insider.