Skip to main content

Hannah Mezei

The recently published National Security Commission on Artificial Intelligence (NSCAI) Report outlines the United States’ strategy to lead the Artificial Intelligence (AI) era. The 16 chapters explore the steps that the national security community must take to responsibly leverage AI in defense settings, defend against AI threats, and bolster AI innovation. Notably, the NSCAI Report underscores the imperative for US Department of Defense (DoD) leadership to quantify confidence when deploying AI, stating that those who use AI “need an informed understanding of risks, opportunities, and tradeoffs.”

During the NSCAI Final Report Small Group Briefing on February 24th, much of the discussion centered around chapter seven of the Report: Establishing Justified Confidence in AI Systems. When asked where the U.S. was headed in the AI space, again, the chapter was referenced. This emphasis on justified confidence in AI systems highlights an important need, an important focus, and furthermore, an important gap, the DoD and broader United States needs filled: a gap CalypsoAI is working to fill. 

At the core of the issue with AI confidence is the fact that AI/ML products are being developed with minimal standard tooling, leading to algorithms of uncertain quality, subjective trustworthiness, and potential vulnerability to attack. Across all sectors, AI solutions are not being deployed due to the lack of tooling to conduct such assessments and provide ongoing monitoring.  

The NSCAI proposes recommendations for five issue areas in establishing justified confidence in AI systems, including robust and reliable AI, human-AI interaction and teaming, Testing and Evaluation, Verification and Validation (TEVV), leadership, and accountability and governance. CalypsoAI concurs with the NSCAI’s findings that confidence and trust are integral to effective AI/ML deployment in the national security space and supports it’s recommendations for establishing justified confidence. Our Secure Machine Learning Life Cycle Platform, VESPR, provides the necessary tools to address these issue areas, enabling confidence, trust, and transparency in AI systems. 

The report suggests that without justified confidence in AI systems there may not be a future for AI in the United States. Therefore, it is crucial for industry and government to recognize that, in order to take full advantage of all AI can offer and increase trusted AI deployment, we must first prioritize the testing and evaluation of AI/ML models, ensuring their robustness, validity, and adherence to standards. 

Hannah Mezei is an Engagement Associate at CalypsoAI. Mezei is a current Master’s candidate in Asian studies with a focus on China, security, and technology at George Washington University. Hannah has experience overseas and is proficient in Mandarin Chinese. Prior to joining CalypsoAI, Hannah was a Research Assistant at the Nuclear Security Working Group, and a Fellow with Pallas Advisors, in Washington, DC.