Solving The AI Black Box: ZK-Proofs in Defence Tech

Sebastian Cuchillo hosts Ismail, CEO of Lagrange Labs, to discuss the integration of zero-knowledge proofs (ZK-proofs) in national security and defense technology. Lagrange aims to enhance decision-making processes and validate model outputs through ZK machine learning. Their core technology, DeepProve, is a zero-knowledge machine learning library that ensures the correctness of model executions while maintaining privacy for inputs and data. Ismail critiques existing privacy solutions that rely on air-gapped systems or hardware-based security, advocating for a first principles approach to integrate privacy into AI applications.

The conversation highlights the commercial viability of AI models, noting that many crypto companies produce models lacking market relevance. ZK machine learning is presented as a solution that allows for privacy and verifiability, emphasizing that models must be trained as ZK from the outset to avoid performance issues. Ismail stresses the importance of protecting both intellectual property and consumer privacy in AI applications, pointing out that many private solutions fail to enhance privacy for commercially relevant models.

The discussion shifts to Lagrange's focus on military and industrial applications, underscoring the significance of AI in sensitive contexts. Ismail reflects on the historical neglect of national security by the tech industry, advocating for innovation that prioritizes defense and global hegemony. He calls for a redirection of advancements in frontier cryptography towards dual-use applications for defense, highlighting Lagrange's expertise in repurposing technology for critical applications.

The podcast also addresses the evolving landscape of capital funding in defense technology, noting that recent reforms in procurement have enabled defense companies to grow rapidly. Increased competition among prime defense contractors is expected to foster innovation and align private sector advancements with military needs. The conversation touches on the geopolitical implications of software superiority for U.S. leadership, emphasizing the critical nature of access to chips and energy independence, as well as concerns about potential disruptions in chip access from Taiwan.

Advancements in U.S. reindustrialization, particularly in domestic manufacturing and AI, are highlighted, along with notable progress in drone technology since the war in Ukraine. The U.S. is shifting its drone doctrine to integrate unmanned UAVs with existing fighter jets. However, challenges remain in small drone manufacturing and swarm technology compared to China.

The conversation transitions to verifiable AI for defense, focusing on Lagrange's work to address gaps in cryptography within national security. Current systems may not be effective in decentralized combat environments, and Lagrange aims to enhance cryptographic security for autonomous weapons and drones. The concept of accountable autonomy is emphasized as vital in contested environments, where AI systems may be compromised. Ensuring the correctness of AI models, particularly in drones, is essential for reliable outputs.

Zero-knowledge machine learning is proposed as a solution to verify the correctness of AI models used in decision-making, particularly in scenarios involving UAVs. Accountability is crucial for ensuring that actions taken are based on accurate models and data. The U.S. military doctrine mandates human involvement in kill decision-making, reflecting the contentious nature of autonomous weapons. Historical examples illustrate that while the technology is not new, advancements in AI could enhance precision and reduce collateral damage.

The conversation also addresses the legal implications of AI in military applications, emphasizing the importance of provability and accountable autonomy to ensure compliance with laws like the Geneva Convention. Concerns about the auditability of AI decisions in military contexts and accountability for failures are raised. The potential shift in military strategies towards targeting adversary AI systems rather than traditional military assets is discussed.

Lagrange's technology involves a decentralized prover network for generating proofs, with a token staked in this network. The deployment of Lagrange's technology can be either decentralized or centralized, depending on user preference. The conflict in Ukraine has spurred innovation in military doctrine, particularly in drone technology, where AI is employed to operate in jammed environments. The U.S. is currently seen as the leader in drone innovation, with onboard AI playing a crucial role in successful military operations.

The conversation highlights the transformative impact of AI on military technologies, emphasizing the risk of a "runaway phenomenon" that could shift military superiority among nations. As countries adapt their military strategies to counter each other's advancements, the U.S. military's reliance on air power is increasingly challenged by China's development of hypersonic missile technology. Concerns are raised about the vulnerability of U.S. aircraft carriers to large-scale hypersonic missile attacks.

The discussion also addresses the evolving role of tanks in modern warfare, particularly in the context of the Ukraine conflict, where they have become more vulnerable to drone attacks. The importance of a robust U.S. market and collaboration between public and private sectors in sustaining military strength is underscored, drawing parallels to historical military manufacturing during World War II. The ongoing nature of global conflicts is acknowledged, with a focus on the necessity of projecting Western values through technology.