Military AI veteran: 'AI-ready data' is a myth — data is the real bottleneck

ChinaTalk · rss · 2026-10-06

ChinaTalk interviews Bharat Patel, Accenture's AI and data lead for defense (former Navy sailor, MITRE and DoD veteran), on why data — not models — is the hardest part of military AI.

• "AI-ready data" is a myth: data is never perfect; use cases should drive data quality requirements. Early Project Maven models failed because training data lacked the actual targets, forcing continuous battlefield data collection.
• Ukraine case study: today's semi-autonomous systems grew from 3+ years of quietly recording drone video and sensor data, labeling it, and feeding it back into platforms. Patel says the DoD has no continuous ML data collection strategy — data is often deleted or loses context.
• Autonomous tanks are further off than expected: the 2035 vision is a vehicle that understands its environment and saves crew lives, but human-in-the-lead remains essential due to civilian-missidentification risks and policy gaps.
• The dark side of data: adversaries can poison battlefield datasets; camouflage, deception, electronic warfare and data poisoning must be tested against.
• Pentagon 4D chess: navigating requirements, funding, acquisition rules and key people determines whether programs survive.

Original post →

More from Infra

• Disco a dangerous short after HBM despec news; H2 China demand may flip — zephyr_z9 · 2026-10-06
• TensorFold bonds dual Thunderbolt 5 links for 83% throughput boost on Apple Silicon — AIFlow_ML · 2026-10-06
• Dual RX 6800 running Qwen3 27B: Linux + ROCm beats Windows Vulkan, 45 tok/s TG — SaGa31500 · 2026-10-06
• DeepSeek open-sources DeepGEMM, a clean and efficient GPU BLAS kernel library, now 8.5k stars — deepseek-ai · 2026-10-06
• AMD's Lisa Su predicts "very high" chip demand for years despite AI safety concerns — AryHHAry · 2026-10-06
• Photonic AI chip can help calculate its own corrections via in-situ gradient descent — bravo_abad · 2026-10-06

Infra channel →