UK and US Take Strides in AI Governance: Supercomputing Initiatives and Executive Orders

By Samudrapom DamReviewed by Susha Cheriyedath, M.Sc.Nov 9 2023

An article published in the journal Nature discussed the steps taken by the United Kingdom (UK) and the United States (US) toward advancing governmental oversight of artificial intelligence (AI). This article discusses several steps, including the robust computing efforts for democratizing/broadening access to supercomputers that will facilitate research on AI systems to bolster their capabilities in AI research.

New directives for AI technology

The first AI executive order of the US, with several directives for federal agencies to guide the utilization of AI, was signed by the US President on 30 October, 2023. The executive order of the US President has considered AI as an all-purpose tech and covered different aspects related to the technology. It emphasized on creating definitions and standards in AI and mandated the US National Institute of Standards and Technology to develop such standards, tools such as watermarks, and red-team testing to ensure that AI systems are trustworthy, secure, and safe.

In red-team testing, good actors test the security of an AI system by misusing it. Additionally, the executive order directed agencies funding life-sciences research to establish standards to prevent the use of AI to develop dangerous biological materials. Agencies were encouraged to assist skilled immigrants with AI expertise to work, stay, and study in the US. The order also directed the National Science Foundation (NSF) to launch and fund at least one regional innovation engine program prioritizing AI-related work and to establish at least four AI research institutes in the next 18 months.

The UK hosted the AI Safety Summit convened by the UK PM on 1–2 November, which was attended by representatives from several countries and tech companies, including Meta and Microsoft. The AI Safety Summit produced the Bletchley Declaration, which agrees to manage better and assess the risks of frontier AI/advanced AI systems that can be utilized for the development of risky technologies, such as biological weapons.

The UK and the US have committed to developing a national AI ‘research resource’ to provide AI researchers with cloud access to powerful supercomputers. Specifically, the UK has made a significant investment to develop such research resources.

Importance of AI research resources

Supercomputers are crucial to efficiently train AI systems as these systems heavily rely on expensive computing infrastructure. Although a better performance can be realized from AI systems by scaling them up, the approach is very expensive as frontier AI system training can take several months and cost hundreds of millions of dollars. Thus, implementing this approach in academia is currently not feasible, necessitating the democratization of all capabilities required to work with these AI systems through research-resource initiatives.

Currently, these capabilities are controlled by companies that intend to generate revenue from them, which increases the importance of providing these capabilities to government-funded organizations and academics working to understand these frontier AI systems better.

The executive order of the US President commits the NSF to launch a pilot of the National AI Research Resource (NAIRR), a proposed system to provide access to AI-capable, powerful computing power through the cloud/a distributed set of servers within 90 days.

A White House NAIRR task force report has estimated an amount of US$2.6 billion as the budget for NAIRR over an initial six-year period.  Similarly, the UK has planned to establish a national AI Research Resource (AIRR) to offer supercomputer-level computing power to different researchers interested in frontier AI research.

The UK government announced £300 million AIRR funding at the AI Safety Summit as part of a £900-million investment to transform the computing capacity of the UK. The investment by the UK is more substantial compared to the US proposal based on the gross domestic product and population of both countries.

Additionally, the UK plan is based on two new supercomputers, including Dawn in Cambridge and Isambard-AI cluster in Bristol. Dawn will start operation in the next two months, while the Isambard-AI cluster will come online in next summer. Isambard-AI will become one of the leading AI-capable supercomputers in the world. Both supercomputers will enable UK researchers to train even the largest frontier models within a reasonable duration.

These steps allow countries such as the UK to develop the expertise required to guide AI for public welfare. However, legislation must be enacted to prevent the development of AI systems in the future that are smart and cannot be controlled easily.