Session Overview

Small Language Models, Autonomous Weapons, and the Human in the Loop.

Ashish Choudhary framed Generative AI in the military not as a silver bullet, but as a way to automate highly structured, repetitive tasks — with armed forces globally exploring it for national security strategy, training simulation, logistics, and cybersecurity. He walked through concrete applications: a pilot project for the Indian Air Force that digitized complex aircraft maintenance knowledge from veterans into an interactive training platform for new recruits, efforts to push decision-making data from central "War Rooms" out to operational task forces on the ground, and predictive logistics systems like the French software "Masa" that model troop movements and shortest-path deployments. On the intelligence side, generative AI is now used to synthesize years of satellite data to spot enemy movement — work that used to require tedious manual analysis.

Because military data is highly classified, he explained why defense organizations can't risk cloud-based models like ChatGPT, and are instead investing heavily in Small Language Models that run locally — which also reduces hallucination risk when trained on tightly scoped, classified data. This set up the session's central tension: autonomous weapons like the kamikaze and swarm drones already in use in Ukraine and Israel raise serious accountability questions, since many operate on simple GPS programming without a "human in the loop." A US-led coalition of 16 countries is now working to build an international legal framework around this.

He closed on two sobering notes: the extreme danger of algorithmic bias on the battlefield — where a poorly trained facial recognition targeting system could be misused against people based on skin color or religion — and the broader geopolitical picture, where the world is now in an autonomous weapons race between the US and China. For a neighboring country like India, this poses a direct national security concern, and the Indian government is responding by opening collaborations with the tech startup ecosystem to rapidly build indigenous autonomous AI capabilities.

Key Takeaways & Concepts

Not a Silver Bullet: Militaries use Generative AI to automate structured, repetitive tasks — national security strategy, training, logistics, and cybersecurity — not as a wholesale replacement for human decision-making.
Knowledge Retrieval & Training: A pilot project digitized veteran aircraft-maintenance expertise for the Indian Air Force into an interactive training platform for new recruits.
Logistics & War Gaming: Systems like France's "Masa" have modeled troop movements and predicted deployment logistics for over a decade.
Cybersecurity & Intelligence: AI is used to predict and counter cyber threats, and to synthesize years of satellite data to spot enemy movement automatically.
Why SLMs Matter for Defense: Classified data can't go to cloud models like ChatGPT — locally run Small Language Models reduce both data-leak risk and hallucinations.
Human in the Loop: Autonomous drones used in Ukraine and Israel often operate on simple GPS programming without human oversight, raising unresolved accountability questions — a 16-country coalition is working on a legal framework.
Bias on the Battlefield: A biased facial recognition targeting system could be misused to target people based on skin color or religion — an extreme ethical risk.
The Autonomous Arms Race: A US-China divide is driving a new global arms race, pushing India to build indigenous autonomous AI capability through startup collaboration.