April 12, 2025
The GIST Editors' notes
This text has been reviewed in line with Science X's editorial course of and insurance policies. Editors have highlighted the next attributes whereas guaranteeing the content material's credibility:
fact-checked
preprint
trusted supply
proofread
Utilizing text-generating AI to energy nuclear analysis
Textual content-generating AI applications like ChatGPT are recognized for on a regular basis duties like answering questions, however that's not all they're good for. These AI brokers can use their text-parsing powers to assist additional nuclear science and help well being physicists, as proven by nuclear engineering graduate scholar Zavier Ndum's analysis.
Giant-language fashions (LLM) like ChatGPT are a type of synthetic intelligence that makes use of written knowledge to generate textual content primarily based on predictions. Though LLMs are helpful for a lot of domains, reminiscent of software program engineering and analysis help, their use in nuclear science is much less frequent. That is partly as a result of nuclear engineering analysis usually makes use of proprietary knowledge that may't be securely fed right into a common LLM.
"In terms of nuclear science, there may be quite a lot of proprietary knowledge, and there are safety points with sharing knowledge," Ndum mentioned. "You may't simply give this information to ChatGPT or Copilot. If in case you have one thing that can be utilized inside a company to automate workflows, it will increase productiveness and effectivity."
Ndum lately wrote a paper a couple of mission utilizing an LLM for nuclear science. The AI agent described in Ndum's paper, AutoFLUKA, may help automate duties in nuclear science analysis, reminiscent of operating pc simulations in a software program known as FLUKA. The applying can take and edit enter recordsdata, run simulations, after which analyze the outcomes, full with plotting graphs.
Different researchers can use Ndum's software with their very own knowledge, securely constructing a database utilizing proprietary knowledge that doesn't go wherever aside from the originating pc.
"You're capable of type by this info very simply and really quick, as a substitute of studying by to extract a easy reply," Ndum mentioned. "If in case you have these paperwork, and you’ve got the applying acquire entry to these paperwork, it will probably simply do Q&A, and you may immediate the mannequin to present you options to reinforce your particular software."
To construct AutoFLUKA as a proof of idea, Ndum needed to work with what he had. He didn’t have entry to work with Monte Carlo N-Particle (MCNP), a pc simulation code that’s generally utilized in nuclear science however can also be regulated. Nevertheless, FLUKA is comparable sufficient to MCNP that Ndum's mannequin may be simply replicated for it and different simulation applications.
One other problem for Ndum was transitioning his analysis focus to finding out AI brokers. Earlier than arriving at Texas A&M, he labored on well being physics and dosimetry, or the examine of radiation absorbed by dwelling tissue. Nevertheless, after working with Dr. John Ford, a nuclear engineering professor who researches well being physics, Ndum used that work to offer case research for the AutoFLUKA paper. He additionally cites the steering of nuclear engineering professor Dr. Yang Liu, his Ph.D. advisor, and Dr. Jian Tao, a Faculty of Efficiency professor affiliated with the Division of Nuclear Engineering, whereas constructing the applying.
"While you step into new territory, it's at all times difficult, however you retain on working and also you consider that there’s something helpful you may make out of this," he mentioned. "While you put in that effort, it seems good."
Ndum, who serves because the president of the State of Texas Chapter of the Well being Physics Society (STC-HPS), additionally hopes to deliver this method again to his authentic area of well being physics. On the chapter's annual convention in October on the College of Texas at Arlington, he introduced a speak about the usage of LLMs in well being physics to function digital assistants for professionals within the area. He examined the applying from his paper on retrieving info from well being physics-related documentation, slicing down an hours-long search to seconds. This may help well being physicists like radiation security officers (RSOs), who should learn about annual radiation dose laws and the operation of assorted machines.
"Sorting by these regulatory guides for RSOs may very well be time-consuming and strenuous," Ndum mentioned.
Ndum additionally delivered a seminar as a part of a semester-long sequence final fall hosted by Texas A&M's Analysis in Synthetic Intelligence for Science and Engineering Initiative. He plans to current one other speak on this matter on the STC-HPS Pupil Convention at Texas A&M College in April, the place nuclear science, well being physics, and engineering college students from universities aside from Texas A&M will attend.
"They will see how one can harness this expertise to their very own particular analysis areas," he mentioned.
Presently, Ndum is advancing his analysis by growing a classy LLM software designed to reply complicated, domain-specific questions in nuclear science analysis. This software kinds by and analyzes paperwork, and it will probably additionally combine info from on-line sources in actual time. It could actually learn and course of a number of file sorts, together with PDFs, photographs and spreadsheets, providing complete help for analysis duties. These talents permit the applying to offer nuanced solutions and insights to make nuclear science analysis extra environment friendly.
"That is new territory, and exploring it inside nuclear science is admittedly necessary," Ndum mentioned. "I'm going to maintain working to see what we're capable of construct with this information."
Liu, who serves because the director of the Scientific Machine studying for Superior Reactor Applied sciences (SMART) group, believes Zavier's work will make an influence on the broader nuclear engineering group.
"LLMs are a analysis focus of our group, and we’re lucky to have Zavier pushing the boundaries of its software in nuclear science," Liu mentioned. "His revolutionary method to integrating AI into nuclear analysis is precisely the type of forward-thinking mindset we’d like on this area. The power to leverage AI for safe, domain-specific automation is a game-changer, and Zavier's contributions are paving the way in which for extra environment friendly, data-driven developments in reactor modeling, well being physics, and nuclear security."
Extra info: Zavier Ndum Ndum et al, AutoFLUKA: A Giant Language Mannequin Primarily based Framework for Automating Monte Carlo Simulations in FLUKA, arXiv (2024). DOI: 10.48550/arxiv.2410.15222
Journal info: arXiv Offered by Texas A&M College Quotation: Utilizing text-generating AI to energy nuclear analysis (2025, April 12) retrieved 12 April 2025 from https://techxplore.com/information/2025-04-text-generating-ai-power-nuclear.html This doc is topic to copyright. Other than any truthful dealing for the aim of personal examine or analysis, no half could also be reproduced with out the written permission. The content material is offered for info functions solely.
Discover additional
A brand new device for quicker, extra in-depth evaluation of nuclear properties and mass knowledge shares
Feedback to editors
