Results for “agentic AI” — Page 7

DOE OSTI2025

Privacy-Aware RAG-Enabled LLMs for Collaborative AI in Organizations

Recent advancements in Large Language Models (LLMs) based on Transformer architectures have significantly improved capabilities in natural language processing and generation. However, deploying LLMs for inter-organizational communication poses challenges, in ensuring privacy and facilitating effective collaboration. This paper introduces a novel decentralized inference meta-agent chatbot that leverages privacy-aware Retrieval-Augmented Generation (RAG)-enabled LLMs for collaborative AI communication across organizations. Built on Microsoft’s Autogen, the platform enables LLMs to autonomously refine responses, enhancing accuracy and relevance. It incorporates advanced hallucination mitigation techniques using Uptrain and a privacy-focused RAG framework that employs synthetic document generation to protect sensitive information. Comprehensive evaluations demonstrate the platform’s effectiveness in maintaining contextual relevance and stringent privacy standards, effectively addressing critical challenges in LLM-enhanced collaborative AI communication. This work represents a significant step toward secure and efficient inter-organizational collaboration using advanced generative AI technologies.

97 - MATHEMATICS AND COMPUTING↗

NASA NTRSDate not supplied

Data Augmentation for Intelligent Contingency Management Using Generative Adversarial Neural Networks

Artificial intelligence (AI)-based techniques for intelligent contingency management (ICM) require that intelligent agents learn various aspects of system dynamics to create and execute contingencies. For high assurance contingency management, agents achieve the most compelling results through supervised or semi-supervised machine learning, for which agents require large datasets to learn the dynamics of the system. Unfortunately, data collection in aerospace applications can be costly, due to both time and resources. Presented work describes a framework for data augmentation of ICM databases containing training data for machine learning models. This framework populates the database with the outputs of generative adversarial network (GAN) models that were trained on flight data. Methods for evaluating the suitability of these models based on the equations of motion, as well as other physical constraints, are discussed. The paper demonstrates the utility of this database for training intelligent agents on the NASA T2 generic transport aircraft model and experimental vertical takeoff and landing (VTOL) simulation model.

Generative Machine Learning↗

NASA NTRSDate not supplied

Data Augmentation for Intelligent Contingency Management Using Generative Adversarial Neural Networks

Artificial intelligence (AI)-based techniques for intelligent contingency management (ICM) require that intelligent agents learn various aspects of system dynamics to create and execute contingencies. For high assurance contingency management, agents achieve the most compelling results through supervised or semi-supervised machine learning, for which agents require large datasets to learn the dynamics of the system. Unfortunately, data collection in aerospace applications can be costly, due to both time and resources. Presented work describes a framework for data augmentation of ICM databases containing training data for machine learning models. This framework populates the database with the outputs of generative adversarial network (GAN) models that were trained on flight data. Methods for evaluating the suitability of these models based on the equations of motion, as well as other physical constraints, are discussed. The paper demonstrates the utility of this database for training intelligent agents on the NASA T2 generic transport aircraft model and experimental vertical takeoff and landing (VTOL) simulation model.

Generative Machine Learning↗

DOE OSTI2022

Genetic programming for understanding cognitive biases that generate polarization in social networks

Recent studies have applied agent-based models to infer human-interpretable explanations of individual-scale behaviors that generate macro-scale patterns in complex social systems. Genetic programming has proven to be an ideal explainable AI tool for this purpose, where primitives may be expressed in an interpretable fashion and assembled into agent rules. Evolutionary model discovery (EMD) is a tool that combines genetic programming and random forest feature importance analysis, to infer individual-scale, human-interpretable explanations from agent-based models. We deploy EMD to investigate the cognitive biases behind the emergence of ideological polarization within a population. An agent-based model is developed to simulate a social network, where agents are able to create or sever links with one another, and update an internal ideological stance based on their neighbors' stances. Agent rules govern these actions and constitute of cognitive biases. A set of 7 cognitive biases are included as genetic program primitives in the search for rules that generate hyper-polarization among the population of agents. We find that heterogeneity in cognitive biases is more likely to generate polarized social networks. Highly polarized social networks are likely to emerge when individuals with confirmation bias are exposed to those with either attentional bias, egocentric bias, or cognitive dissonance.

Gunaratne, Chathika↗

DOE OSTI2026

REFSafE: A RAG-Enabled Framework for Predictive Risk Analysis and Automated Safety Report Generation in Mission-Critical Environments

Operational safety in mission-critical environments requires AI systems that are accurate, interpretable, and resistant to hallucination. We present an agentic Retrieval-Augmented Generation (RAG) framework, REFSafe, for grounded hazard analysis and automated safety report generation. The system integrates Large Language Models (LLMs) with structured operational data, historical incident repositories, policy documents, and external authoritative sources. Through iterative agentic reasoning, the framework retrieves, verifies, and synthesizes evidence prior to generation, enforcing citation-backed outputs with explicit source attribution (documents, links, and prior events) to ensure traceability and trust. To mitigate hallucinations and unsupported claims, all risk assessments and forecasts are constrained to retrieved evidence, with confidence signals derived from retrieval relevance and source consistency. A transparent pipeline enables subject matter experts (SMEs) to validate predictions, and provide structured feedback, forming a continuous performance calibration loop. Preliminary deployment demonstrates improved reliability in hazard detection and safety/vulnerability report generation. This work advances trustworthy, evidence-grounded AI for predictive safety intelligence in mission-critical operations.

Das, Sanjay [ORNL] (ORCID:0009000542591915)↗

DOE OSTI2024

Employing artificial intelligence to steer exascale workflows with colmena

Computational workflows are a common class of application on supercomputers, yet the loosely coupled and heterogeneous nature of workflows often fails to take full advantage of their capabilities. We created Colmena to leverage the massive parallelism of a supercomputer by using Artificial Intelligence (AI) to learn from and adapt a workflow as it executes. Colmena allows scientists to define how their application should respond to events (e.g., task completion) as a series of cooperative agents. In this paper, we describe the design of Colmena, the challenges we overcame while deploying applications on exascale systems, and the science workflows we have enhanced through interweaving AI. The scaling challenges we discuss include developing steering strategies that maximize node utilization, introducing data fabrics that reduce communication overhead of data-intensive tasks, and implementing workflow tasks that cache costly operations between invocations. These innovations coupled with a variety of application patterns accessible through our agent-based steering model have enabled science advances in chemistry, biophysics, and materials science using different types of AI. In conclusion, our vision is that Colmena will spur creative solutions that harness AI across many domains of scientific computing.

Workflows↗

DOE OSTI2024

Towards a High Fidelity Training Environment for Autonomous Cyber Defense Agents

Cyber defenders are overwhelmed by the frequency and scale of attacks against their networks. This problem will only be exacerbated as attackers leverage AI to automate their workflows. Autonomous cyber defense capabilities could aid defenders by automating operations and adapting dynamically to novel threats. However, existing training environments fall short in areas such as generalization, explainability, scalability, and transferability, making it intractable to train agents that will be effective in real networks. In this paper we take an important step towards creating autonomous cyber defense agents — we present a high fidelity training environment called Cyberwheel that includes both simulation and emulation capabilities. Cyberwheel simplifies customization of the training network and easily allows redefining the agent’s reward function, observation space, and action space to support rapid experimentation of novel approaches to agent design. It also provides visibility into agent behaviors necessary for agent evaluation and sufficient documentation / examples to lower the barrier to entry. As an example use case of Cyberwheel, we present initial results training an autonomous agent to deploy cyber deception strategies in simulation.

Oesch, T↗

NASA NTRS1991

Design and Analysis Tools for Concurrent Blackboard Systems

A blackboard system consists of a set of knowledge sources, a blackboard data structure, and a control strategy used to activate the knowledge sources. The blackboard model of problem solving is best described by Dr. H. Penny Nii of the Stanford University AI Laboratory: "A Blackboard System can be viewed as a collection of intelligent agents who are gathered around a blackboard, looking at pieces of information written on it, thinking about the current state of the solution, and writing their conclusions on the blackboard as they generate them. " The blackboard is a centralized global data structure, often partitioned in a hierarchical manner, used to represent the problem domain. The blackboard is also used to allow inter-knowledge source communication and acts as a shared memory visible to all of the knowledge sources. A knowledge source is a highly specialized, highly independent process that takes inputs from the blackboard data structure, performs a computation, and places the results of the computation in the blackboard data structure. This design allows for an opportunistic control strategy. The opportunistic problem-solving technique allows a knowledge source to contribute towards the solution of the current problem without knowing which of the other knowledge sources will use the information. The use of opportunistic problem-solving allows the data transfers on the blackboard to determine which processes are active at a given time. Designing and developing blackboard systems is a difficult process. The designer is trying to balance several conflicting goals and achieve a high degree of concurrent knowledge source execution while maintaining both knowledge and semantic consistency on the blackboard. Blackboard systems have not attained their apparent potential because there are no established tools or methods to guide in their construction or analyze their performance.

McManus, John W.↗

NASA NTRSDate not supplied

A 3D Simulation Platform for Decentralized Decision-Making in Advanced Air Mobility

This paper presents a general purpose, plug-and-play simulation platform for the use of future aviation stakeholders, such as urban airspace planners, air vehicle operators, ground operation managers, air traffic controllers and aviation researchers. The presented simulator platform is envisioned to serve as a toolkit to visualize, evaluate, and configure future advanced air mobility (AAM) operations. Highlighting features of this toolkit include a modular architecture that allows multiple smart unmanned aerial systems (UASs) to remotely connect to the simulation server and participate in decentralized decision-making scenario simulations. As an example of the decentralized decision-making scenario, an inter-agent negotiation-based conflict resolution use case is considered in this paper, where the UASs leverage the on-board/on-the-edge artificial intelligence (AI) capability to continually build situational awareness, and use this information to predict future conflicts and resolve them through machine-to-machine negotiation. As such operations are non-existent at scale currently, the presented simulation platform offers a viable and cost-effective alternative for assessing the efficacy of AAM research outcomes and challenges in future shared airspace usage. The simulation platform allows plug-n-play connectivity with AI and non-AI compute modules representing individual UAS’s flight control. Each module can interact with the simulation platform independently to communicate current and desired future states, situational awareness, and conflict resolution utilization costs for inter-agent negotiation. The simulation environment orchestrates realistic operational scenarios with spatiotemporal details, dynamic events, tactical conflict-resolution methods, interfaces for customizing air traffic control parameters, and information exchange uncertainties. In the future, this can serve as a community focused cloud simulation platform, incorporating multi-stakeholder airspace constraints from regulatory, government, city, and local agencies.

Aditya N Das↗

NASA NTRS2022

A 3D Simulation Platform for Decentralized Decision-Making in Advanced Air Mobility

This paper presents a general purpose, plug-and-play simulation platform for the use of future aviation stakeholders, such as urban airspace planners, air vehicle operators, ground operation managers, air traffic controllers and aviation researchers. The presented simulator platform is envisioned to serve as a toolkit to visualize, evaluate, and configure future advanced air mobility (AAM) operations. Highlighting features of this toolkit include a modular architecture that allows multiple smart unmanned aerial systems (UASs) to remotely connect to the simulation server and participate in decentralized decision-making scenario simulations. As an example of the decentralized decision-making scenario, an inter-agent negotiation-based conflict resolution use case is considered in this paper, where the UASs leverage the on-board/on-the-edge artificial intelligence (AI) capability to continually build situational awareness, and use this information to predict future conflicts and resolve them through machine-to-machine negotiation. As such operations are non-existent at scale currently, the presented simulation platform offers a viable and cost-effective alternative for assessing the efficacy of AAM research outcomes and challenges in future shared airspace usage. The simulation platform allows plug-n-play connectivity with AI and non-AI compute modules representing individual UAS’s flight control. Each module can interact with the simulation platform independently to communicate current and desired future states, situational awareness, and conflict resolution utilization costs for inter-agent negotiation. The simulation environment orchestrates realistic operational scenarios with spatiotemporal details, dynamic events, tactical conflict-resolution methods, interfaces for customizing air traffic control parameters, and information exchange uncertainties. In the future, this can serve as a community focused cloud simulation platform, incorporating multi-stakeholder airspace constraints from regulatory, government, city, and local agencies.

Aditya Das↗

DOE OSTI2023

Unraveling Vulnerabilities in Endocrine Therapy-Resistant HER2+/ER+ Breast Cancer

Abstract Breast tumors overexpressing human epidermal growth factor receptor (HER2) confer intrinsic resistance to endocrine therapy (ET), and patients with HER2/estrogen receptor–positive (HER2+/ER+) breast cancer (BCa) are less responsive to ET than HER2–/ER+. However, real-world evidence reveals that a large subset of patients with HER2+/ER+ receive ET as monotherapy, positioning this treatment pattern as a clinical challenge. In the present study, we developed and characterized 2 in vitro models of ET-resistant (ETR) HER2+/ER+ BCa to identify possible therapeutic vulnerabilities. To mimic ETR to aromatase inhibitors (AIs), we developed 2 long-term estrogen deprivation (LTED) cell lines from BT-474 (BT474) and MDA-MB-361 (MM361). Growth assays, PAM50 subtyping, and genomic and transcriptomic analyses, followed by validation and functional studies, were used to identify targetable differences between ET-responsive parental and ETR-LTED HER2+/ER+ cells. Compared to their parental cells, MM361 LTEDs grew faster, lost ER, and increased HER2 expression, whereas BT474 LTEDs grew slower and maintained ER and HER2 expression. Both LTED variants had reduced responsiveness to fulvestrant. Whole-genome sequencing of aggressive MM361 LTEDs identified mutations in genes encoding transcription factors and chromatin modifiers. Single-cell RNA sequencing demonstrated a shift towards non-luminal phenotypes, and revealed metabolic remodeling of MM361 LTEDs, with upregulated lipid metabolism and ferroptosis-associated antioxidant genes, including GPX4. Combining a GPX4 inhibitor with anti-HER2 agents induced significant cell death in both MM361 and BT474 LTEDs. The BT474 and MM361 AI-resistant models capture distinct phenotypes of HER2+/ER+ BCa and identify altered lipid metabolism and ferroptosis remodeling as vulnerabilities of this type of ETR BCa.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2026

Wolf

The Workflow Orchestration Language Framework (WOLF) is an agentic framework grounded in natural language with an architecture inspired by reinforcement learning (RL)—designed to orchestrate, scale, and accelerate complex workflows. The concept of WOLF was born out of the very successful ASC Tri-lab Multi-Agent Design Assistant (MADA) project, but extends beyond its domain-specific design agents to provide a more general and extensible architecture. WOLF capitalizes on the lessons learned from MADA and is fully aligned with Sutton’s The Bitter Lesson—that the most enduring progress in AI comes from general-purpose methods that scale with computation, rather than narrow techniques built on domain-specific human knowledge. In this spirit, WOLF enables agents to autonomously learn workflows, capture strategies as reusable playbooks, and build a growing corpus of interpretable, auditable “wisdom artifacts.” These artifacts, expressed in natural language, bridge human and machine understanding while preserving adaptability and scalability as computational power continues to expand.

Boureima, Ismaeal↗

NASA NTRS2004

Preliminary results of the autonomous sciencecraft experiment

The Autonomous Sciencecraft Experiment (ASE) will operate onboard the Earth Orbiter 1 mission in 2003. The ASE software uses onboard continuous planning, robust task and goal-based execution, and onboard machine learning and pattern recognition to radically increase science return by enabling intelligent downlink selection and autonomous retargeting. In this paper we discuss how these AI technologies are synergistically integrated in multilayer control architecture to enable a virtual spacecraft science agent.

Rabideau, Gregg↗

NASA NTRSDate not supplied

Inertial Transfer: Concept and Multi-Agent Approach

Research and development of cost saving technologies is vital to the success of NASA’s strategic goal to extend human presence deeper into space and to the moon for sustainable long-term exploration and utilization. Reduction of mass has long been a method of reducing space mission costs. In this submission, Inertial Transfer, a new approach to mass transfer in space, is presented. A Multi-Agent System solution to the problem space is discussed. A scaled base-line simulator is constructed and demonstrated which will enable the development of AI capabilities necessary to perform Inertial Transfer. Finally, future work and key challenges are discussed.

Multi-Agent↗

NASA NTRS2004

Applications for Mission Operations Using Multi-agent Model-based Instructional Systems with Virtual Environments

This viewgraph presentation provides an overview of past and possible future applications for artifical intelligence (AI) in astronaut instruction and training. AI systems have been used in training simulation for the Hubble Space Telescope repair, the International Space Station, and operations simulation for the Mars Exploration Rovers. In the future, robots such as may work as partners with astronauts on missions such as planetary exploration and extravehicular activities.

Clancey, William J.↗

NASA NTRS1991

A framework for building real-time expert systems

The Space Station Freedom is an example of complex systems that require both traditional and artificial intelligence (AI) real-time methodologies. It was mandated that Ada should be used for all new software development projects. The station also requires distributed processing. Catastrophic failures on the station can cause the transmission system to malfunction for a long period of time, during which ground-based expert systems cannot provide any assistance to the crisis situation on the station. This is even more critical for other NASA projects that would have longer transmission delays (e.g., the lunar base, Mars missions, etc.). To address these issues, a distributed agent architecture (DAA) is proposed that can support a variety of paradigms based on both traditional real-time computing and AI. The proposed testbed for DAA is an autonomous power expert (APEX) which is a real-time monitoring and diagnosis expert system for the electrical power distribution system of the space station.

Lee, S. Daniel↗

NASA NTRS2004

Preliminary results of the autonomous sciencecraft experiment

The Autonomous Sciencecraft Experiment (ASE) will operate onboard the Earth Orbiter 1 mission 2004. The ASE software uses onboard continuous planning, robust task and goal-based execution, and onboard machine learning and patter recognition to radically increase science return by enabling intelligent downlink selection and autnomous retargeting. In this paper we will discuss how these AI technologies are synergistically integrated in multi-layer control architecture to enable a virtual spacecraft science agent.

Rabideau, Gregg↗

DOE OSTI2021

Gaming the beamlines—employing reinforcement learning to maximize scientific outcomes at large-scale user facilities

Abstract Beamline experiments at central facilities are increasingly demanding of remote, high-throughput, and adaptive operation conditions. To accommodate such needs, new approaches must be developed that enable on-the-fly decision making for data intensive challenges. Reinforcement learning (RL) is a domain of AI that holds the potential to enable autonomous operations in a feedback loop between beamline experiments and trained agents. Here, we outline the advanced data acquisition and control software of the Bluesky suite, and demonstrate its functionality with a canonical RL problem: cartpole. We then extend these methods to efficient use of beamline resources by using RL to develop an optimal measurement strategy for samples with different scattering characteristics. The RL agents converge on the empirically optimal policy when under-constrained with time. When resource limited, the agents outperform a naive or sequential measurement strategy, often by a factor of 100%. We interface these methods directly with the data storage and provenance technologies at the National Synchrotron Light Source II, thus demonstrating the potential for RL to increase the scientific output of beamlines, and layout the framework for how to achieve this impact.

36 MATERIALS SCIENCE↗