The U.S. Navy deploys ensembles of deep learning models on unmanned underwater vehicles (UUVs) for real-time, automated mine detection. This on-device AI allows the UUVs to operate autonomously in communication-denied environments and dynamically adjust their missions based on what they find.
The Department of Defense employs a dual-pronged approach to technology acquisition. For common needs like large language models, it's more effective to adapt requirements to fit existing, at-scale commercial products. For highly specialized military applications with no civilian equivalent, like underwater mine detection, bespoke development is necessary.
The speaker argues that technology, including advanced AI, is a critical enabler but not the decisive factor in winning wars. He emphasizes that human elements like fighting spirit, training, decision-making under pressure (the "fog of war"), and logistics are ultimately more important than marginal technological superiority.
The conversation highlights the vast gap between a working technology prototype and a field-ready, sustainable military system. Professionals focus on logistics, cybersecurity, policy compliance, and sustainment—the critical elements that constitute the majority of the development effort.
The speaker identifies multi-agent collaborative autonomy as a critical and underserved area for future military AI development. The goal is to create systems where different autonomous agents can share information, coordinate tasks, and feed into each other's decision-making in real-time to overcome the "fog of war."
Keep pulling the thread on Jon Haase.