Britain's Ten Year Rule - A Warning on AI Adoption in Defence

The following report was submitted to the Defence Select Committee of the House of Commons in January 2024, in order to educate policy makers on the adoption of AI in Britain’s defence sectors.

Recommendations:

• Move to a proactive approach to AI capability development and deploy capabilities today without perpetual delay waiting for ideal conditions

  • Do not wait for the MOD’s data strategy to progress, or for massive hand labelled data sources. Training on available data (including open source data and internal classified sources) and using self-supervised and low-shot learning AI techniques can enable capability gains today.

  • Do not wait for or overly rely upon future secure cloud solutions, edge computing and on-premise hosting should be used now.

• Support Small and Medium Enterprises (SMEs) to overcome bureaucratic challenges, and the “chicken and egg” problem of not being able to do classified work (except typically after years of delay), because they have not managed classified work before, including:

  • Secure Compute - Offer at-cost secure hosting/cloud services or blueprints for self-provision to handle classified data.

  • Facility Security Clearance - Provide or sell at-cost secure facilities to provide SMEs locations to handle classified data.

  • Personnel Clearances - Hold clearances on behalf of SMEs at-cost (similar to its own employees), to facilitate their participation in classified defence work.

  • Commercial Frameworks - Make existing frameworks like G-Cloud accessible to SMEs, without needing to wait for infrequent entry windows which introduce years of delay.

• Separate software and hardware contracts, avoiding “Prime” bias, or at minimum create distinct “lots” within procurements

  • Stop bundling software and hardware by default, and make this the exception, to genuinely get “best athlete” support. Bundling is the default because MOD fears integration challenges.

  • No matter how astute the MOD considers itself as a customer, the Primes exploit bundling to their advantage and dominate the integrated contract market, especially given cultural MOD biases towards hardware over software.

  • Seek to purchase COTs (commercial off-the-shelf) products, and hardware-agnostic software solutions, rather than developing new software from scratch for each hardware platform. Hardware is the commodity, software is the differentiator.

  • Unbundling and COTs also opens up opportunities for a broader range of suppliers, including smaller, more agile firms that specialise in AI and software

• Comprehensively reform acquisition and commercial processes to enabling a rapid end-to-end process

  • Streamline business case and approval processes, especially for lower value projects and cutting-edge technology.

  • Pull capabilities through after experimentation programmes - initiatives like DASA should provide a realistic pathway for SMEs to scale up, rather than offering false hope, consuming scarce resources and condemning them to the “valley of death”.

  • Embrace genuine “agile” development, giving capability owners the opportunity to get end use feedback, “fail fast” and “learn by doing” to quickly test new opportunities. MOD is still culturally dominated by lethargic “waterfall” development, where only years or decades later programmes are declared a failure, with no accountability, as incentives block early intervention.

  • Abolish social value criteria in procurements and return to scoring tenders based on the quality and price of the capability - as explored in more detail in the ASI’s dedicated report on the Social Value Act,1 it adds waste to the procurement system and reduces value for money for taxpayers.