Trustworthy Generative Artificial Intelligence (GenAI) to Structure Data and Deliver Accurate Insights of Command, Control, Communication and Computer (C4) Systems

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Trusted AI and Autonomy The technology within this topic is restricted under the International Traffic in Arms Regulation (ITAR), 22 CFR Parts 120-130, which controls the export and import of defense-related material and services, including export of sensitive technical data, or the Export Administration Regulation (EAR), 15 CFR Parts 730-774, which controls dual use items. Offerors must disclose any proposed use of foreign nationals (FNs), their country(ies) of origin, the type of visa or work permit possessed, and the statement of work (SOW) tasks intended for accomplishment by the FN(s) in accordance with the Announcement. Offerors are advised foreign nationals proposed to perform on this topic may be restricted due to the technical data under US Export Control Laws. OBJECTIVE: The objective of this SBIR Phase II topic is to develop an efficient prototype based on a prior feasibility study to utilize/develop GenAI models (e.g., transformer-based models, variational auto-encoders, generative adversarial networks (GAN)) on C4 systems to structure data, and deliver accurate insights of these systems to explain the decisions made by these models to develop trust between the models and an operator. According to secretary of the USAF, this SBIR topic follows two of the seven operational imperatives as an urgent need to be developed as below: II - Achieving Operationally Optimized Advanced Battle Management Systems (ABMS) / Air Force Joint All-Domain Command & Control (AF JADC2) V - Defining optimized resilient basing, sustainment, and communications in a contested environment DESCRIPTION: Though novel technology like ChatGPT dominated headlines recently based on transformer-based models (i.e., a type of GenAI (a class of machine learning (ML) algorithms that can learn from content such as text, images, and audio to generate new content)), it has not yet gained enough credibility to be used in the DoD systems due to its inability to provide accurate explainable decisions by deciphering the inner-workings of the models [1]. To be straightforward, commanders are not going to trust a tool unless they understand how and what data their system was trained on, and how decisions are made to execute an operation [5]. There are still numerous unresolved inquiries surrounding the enhancement of GenAI's capabilities and operator-friendliness. One such open inquiry: how can we enable explainability, allowing operators to grasp and form a clearer mental model of GenAI? Recent research conducted by Goodfellow et al. [2] and Ross et al. [3] has delved into the development of more explainable GenAI models that align with human-understandable processes. However, a comprehensive perspective on explainability of GenAI model such as ChatGPT is still missing. That begs another question: what does an operator need to understand/trust about a GenAI model easily to achieve his/her goals during operational use? Because of these unanswered questions about the model to build trust and transparency within the warfighters' system usages, it is imperative for the DoD operations to develop a system accordingly [4]. Therefore, this SBIR topic seeks proposals to develop a DoD based trustworthy GenAI (described above) system that will not only provide ChatGPT like information but also perform to structure data effectively and deliver/explain accurate insights/decisions about the system to build trust between an operator and a GenAI model. Additionally, proposal should address which uncertainty they are trying to solve such as epistemic or aleatoric in developing/utilizing GenAI based large language models (LLM). PHASE I: As this is a Direct-to-Phase-II (D2P2) topic, no Phase I awards will be made as a result of this topic. To qualify for this D2P2 topic, the Government expects the Offeror to demonstrate feasibility by means of a prior Phase I-type effort that does not constitute work undertaken as part of a prior SBIR/STTR funding agreement. A Phase I-type feasibility study is needed as minimum threshold to satisfy requirement for this Direct-to-Phase-II (D2P2)solicitation. The candidate applying to this solicitation will provide proof of having at least two or more of having extended, explored, analyzed or used ChatGPT in an applicable/similar case scenario that is being explored in the objective of this topic to utilize/develop GenAI models (e.g., transformer-based models, variational auto-encoders, generative adversarial networks (GAN), deep reinforcement learning (DRL)) on C4 systems to structure data, and deliver accurate insights of these systems to explain the decisions made by these models to develop trust between the models and an operator. PHASE II: This direct to phase description will seek to directly implement the objective of this topic setforth as to develop an efficient prototype based on a prior feasibility study to utilize/develop GenAI models (e.g., transformer-based models, variational auto-encoders, generative adversarial networks (GAN), deep reinforcement learning (DRL)) on C4 systems to structure data, and deliver accurate insights of these systems to explain the decisions made by these models to develop trust between the models and an operator. Performers will develop design and specifications and implementation to demonstrate a suitable prototype to proof the explainability factor of GenAI models to create trust between the models and an operator. PHASE III DUAL USE APPLICATIONS: Phase III efforts will focus on transitioning operationally ready technology to a commercial sector or DoD environment. The offeror will identify transition partners. TRL should be at a minimum of a TRL 6. The ChatGPT solution will should have a well developed transition plan to deliver the realization of such technology to the war fighter or commericial sector. The transition plan should work on identifying a program of record where the technology will be reside. REFERENCES: Sun, Jiao, et al. "Investigating explainability of generative AI for code through scenario-based design." 27th International Conference on Intelligent User Interfaces. 2022; Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. 2020. Generative adversarial networks. Commun. ACM 63, 11 (2020), 139 144.; KEYWORDS: ChatGPT;GENAI;explainable GenAI;trustworthy GenAI