A PERSONAL ARTIFICIAL INTELLIGENCE (AI) RADIATION HEALTH PHYSICIST COMPANION FOR RADIOLOGICAL ENVIRONMENTS

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Advanced Computing and Software; Human-Machine Interfaces; Trusted AI and Autonomy OBJECTIVE: DTRA seeks to develop an AI radiation health physicist companion leveraging state-of-the-art machine learning and AI technologies, including large language models, to provide real-time, layperson-friendly guidance on radiation exposure during military operations, with inference running efficiently on a small local machine. DESCRIPTION: The concept involves harnessing the power of machine learning and AI to create an intuitive and accessible AI companion akin to Baymax from the movie Big Hero 6, but specializing in radiation physics and health effects. By utilizing cutting-edge technologies such as large language models (LLMs), the AI companion will be capable of understanding and interpreting complex technical information about radiation dose rates, exposure limits, and biological effects. The primary function of the companion will be to translate this information into easily understandable language for laypersons, enabling individuals to make informed decisions regarding radiation exposure in various scenarios. Within a tactical unit, the companion may be used to recommend stay time based on prescribed dose limits or potential acute effects. At the strategic level, along with information on military formations along with real-time dosimetry measurements, the companion may be used to predict degradation of combat power over time due to acute and delayed effects of radiation exposure. This AI health physicist companion has the potential to revolutionize how we approach radiation safety and risk management across industries such as healthcare, nuclear energy, aerospace, and emergency response. By providing real-time guidance and personalized recommendations, it can enhance safety protocols, optimize radiation exposure limits, and mitigate health risks for workers, patients, and the general public. Furthermore, its user-friendly interface and ability to deliver information in plain language will democratize access to expertise in radiation physics. PHASE I: In phase 1, the goal is to develop a proof of concept for an AI health physicist companion utilizing state-of-the-art machine learning and AI technologies. This involves creating a small-scale prototype capable of running real-time inference on a local machine, leveraging large language models (LLMs) to interpret and translate complex technical information about radiation into easily understandable language for laypersons. Our primary focus during this phase will be on testing the accuracy, usability, and effectiveness of the prototype through initial user feedback and validation. PHASE II: Building upon the insights gained from phase 1, phase 2 will focus on the development of a fully functional prototype. Here, we will scale up the prototype to handle a broader range of radiation-related scenarios and questions, while also refining the user interface and experience based on iterative feedback from users. Additional features, such as personalized risk assessment and emergency response guidance, will be implemented to enhance the prototype's utility and practicality in real-world settings. Extensive user testing and validation will be conducted to ensure the prototype's reliability and effectiveness across various industries and scenarios. PHASE III DUAL USE APPLICATIONS: In phase 3, the focus will shift towards maturation and early operational assessments. The AI health physicist companion will be matured for commercial release, developing marketing strategies to target industries and organizations where radiation safety is a concern. Establishing partnerships and collaborations with relevant stakeholders, including regulatory bodies and industry associations, will be crucial for market penetration and ensuring regulatory compliance. REFERENCES: 1. https://solve.mit.edu/challenges/healthy-cities/solutions/10133 KEYWORDS: Large language models; LLM; artificial intelligence; radiological; nuclear; machine learning; health care; physics