AI in Defence by Dennis Garcia

Key Concepts:

• AI in Defense
• Operational Domains (Space, Land, Sea, Air, Cyber)
• C5ISR (Command, Control, Communications, Computer, Cyber, Intelligence, Surveillance, and Reconnaissance)
• Digital Spine
• Data-Driven Decision Making
• Generative AI (GenAI)
• Large Language Models (LLMs)
• Data Classification and Governance
• Authority to Operate (ATO)
• Tactical Edge
• Mobility, Size, Weight, and Power (SWaP)
• Disrupted, Intermittent, and Latent (DIL) Environments

1. Introduction

• Dennis Garcia, a system engineer at General Dynamics Mission Systems (GDMS) Canada, discusses the emerging role of AI in defense.
• GDMS is a trusted partner to the Canadian Armed Forces for over 70 years.
• GDMS provides capabilities to the Canadian Armed Forces in four out of five operational domains.

2. Operational Domains and Cyber's Role

• Defense Forces divide operations into five domains: space, land, sea, air, and cyber.
• The Cyber domain, recognized by NATO in 2016, is the newest.
• Cyber enables data generation, movement, analysis, and presentation, informing decisions across all domains.

3. C5ISR: The Integrated Digital System

• C5ISR is a key system within the Cyber domain.
• It integrates digital systems for data gathering and processing, enhancing situational awareness and decision-making.
• The operational view of C5ISR interconnects systems, assets, and personnel.

4. Enterprise vs. Tactical Edge

• Enterprise Systems: Cloud platforms, big data centers, high-bandwidth communication links (fiber).
• Deployed Headquarters: Smaller data centers, Edge Cloud systems, wireless connectivity.
• Tactical Edge: Land communication support systems, vehicle/man-portable systems, low-bandwidth wireless (UHF/VHF radios), limited infrastructure.
• Mobility increases from Enterprise to Tactical Edge, impacting size, weight, power, and operation in DIL environments.
• System capability (compute, storage) decreases from Enterprise to Tactical Edge.

5. GDMS and the Digital Spine

• GDMS enables C5ISR with its "digital spine," a set of hardware and software services.
• The digital spine optimizes data needs from the mission network to the dismounted soldier.
• The goal is to become more data-driven for better-informed decisions.

6. The Need for AI: Overcoming Data Overload

• Modern sensors (uncrewed vehicles, F-35 aircraft) generate vast amounts of data.
• Human analysts are overwhelmed by the volume, diversity, and complexity of this data.
• AI, particularly generative AI, can unlock the value of this data.
• The goal is to complement, not replace, human intelligence with AI.
• Quote from the DND and Canadian Armed Forces AI strategy: "The sheer volume diversity complex it of data produced by modern sensors has already overwhelmed the capacity of human analyst to process and interpreted."
• Quote from the DND and Canadian Armed Forces AI strategy: "In many domains human intelligence supported by AI can deliver results that are superior to those achieved separately."

7. AI Problem Spaces (UK Defense AI Centre)

• Recognize: Detect subjects of interest (object recognition).
• Predict: Anticipate likely outcomes (equipment parts failure prediction).
• Simulate: Explore scenarios for planning (autonomous vehicle operation).
• Comprehend: Derive insights from data.
• Generate: Create new and original content.
• LLMs naturally fit into the "Comprehend" and "Generate" categories.
• AI application is currently narrowly focused due to resource limitations.
• LLMs and generative AI may broaden the application of AI.

8. Challenges of Applying Generative AI in Defense

• GenAI Iceberg: The visible portion is small compared to the underlying data storage and processing requirements (Nvidia cards). Deploying this technology outside of the Enterprise space is challenging.
• Data Classification and Governance:
  • Data classifications restrict information access based on security clearance and need-to-know.
  • Export regulations (ITAR, CGP) control information sharing.
  • Enforcing data governance on systems that compute, summarize, and disseminate vast amounts of information is difficult.
• Authority to Operate (ATO):
  • ATO is the process of authorizing a system to operate on the CAF network.
  • It assesses risk and ensures operation within defined boundaries.
  • Rapid innovation in AI makes it difficult to determine proper regulations for security and risk mitigation.

9. GDMS's Role in Addressing Challenges

• GDMS provides scale, speed, and resources for applying innovative technologies in the defense sector.
• This includes use case development, addressing governance and ATO concerns, and implementing technologies in challenging tactical environments.
• GDMS builds enabling solutions and architectures like the digital spine.
• GDMS creates physical hardware and software products like meshnet capabilities.
• GDMS incorporates enabling technologies like public cloud, Edge cloud, and classified cloud.
• GDMS partners with end-users through programs like the LC4ISR support programs.

10. GDMS's Internal AI Initiatives and Partnerships

• GDMS leverages generative AI through internal digital transformation initiatives (Noma, Gaia).
• These programs cautiously leverage LLMs to extract value from internal data while navigating risks.
• GDMS partners with SMBs, academia (UNB), and big industry (AWS, Google).
• GDMS has created a battle lab as an incubator for industry to identify use cases and collaborate on solution development.

11. Conclusion

• AI, especially generative AI, holds significant potential for the defense sector.
• Challenges related to deployment, data governance, and security must be addressed.
• GDMS is actively working to overcome these challenges through internal initiatives, partnerships, and the development of enabling technologies.