A US company claims it is speeding up the path to fusion power by using machine learning. TAE Technologies has trimmed computing tasks that once took months to just a few hours using AI. It’s one of many companies using AI to help with research. “What we still don’t know about fusion—e.g., how to reach and maintain stable fusion conditions—is hiding in the data,” Diogo Ferreira, a professor of Information Systems at the University of Lisbon in Portugal, who studies the application of AI in fusion research told Lifewire in an email interview. “Remember that a fusion machine is a complex scientific experiment, but one thing is for sure—all these machines have dozens, if not hundreds of diagnostic systems attached to it,” he added. “This means that a single experiment, which lasts for only a few seconds, can generate an amount of data on the order of 10 to 100 gigabytes.”
Star Power
Practical fusion is a form of power generation that creates electricity using heat from nuclear fusion reactions. It’s the same type of reaction that powers stars. After decades of slow progress, fusion research is heating up. Scientists recently announced they had generated the highest sustained energy pulse ever created by fusing atoms, more than doubling their own record from experiments performed in 1997. TAE Systems hopes that AI could help break through technical barriers. The company uses a 100 foot-long fusion cylinder, called Norman, for experiments. Google’s AI is being used to sift through the huge amounts of data generated during the research. “With our assistance using machine optimization and data science, TAE achieved their major goals for Norman, which brings us a step closer to the goal of breakeven fusion,” Ted Baltz, Senior Staff Software Engineer, Google Research, wrote on the company’s website. “The machine maintains a stable plasma at 30 million Kelvin for 30 milliseconds, which is the extent of available power to its systems. They have completed a design for an even more powerful machine, which they hope will demonstrate the conditions necessary for breakeven fusion before the end of the decade.” Machine learning is necessary to analyze experiments to discover the trends that govern the behavior of fusion plasmas, Ferreira said. And, researchers need sophisticated approaches to experiment control beyond the hard-coded alarms and triggers they currently employ. “Currently, we use primitive control systems that hit the brakes at the first sign of trouble,” Ferreira said. “We need AI techniques to drive us safely through the intricacies of operating a fusion machine reliably in order to generate a net energy output.”
AI to the Rescue
Medical research is another area where AI is being put to use. AI is a useful complement to the work of human scientists because machines and humans are good at different tasks necessary in research, Sungwon Lim, the CEO of Imprimed Inc., an AI-based predictive cancer detection tool, told Lifewire via email. “Where humans are able to come up with creative solutions and innovations, machines can analyze massive amounts of data quickly and accurately,” he said. “AI can also do the kinds of tedious, repetitive tasks that can cause human researchers to tire and make mistakes. This makes AI an ideal tool for research in which patterns must quickly be found in very large datasets.” A recent study by researchers at the University of Illinois published in the Journal of Critical Reviews in Oncology showed that machine learning currently rivals, and in some cases surpass, trained clinicians in diagnosis and outcome prediction in bladder cancer. “The critical role of AI in early diagnosis of cancer cannot be overstated because every year millions of cases of cancer go undiagnosed until the late stages of the disease where therapeutic options become extremely limited or nonexistent,” Soheila Borhani, one of the paper’s author’s told Lifewire in an email.
