Public compute's function in action
The UK Government has announced a significant investment of £900 million in a new AI Research Resource (AIRR), to be hosted by the University of Bristol. This move is part of a broader strategy to reshape the dynamics of AI development and promote public value throughout the AI supply chain.
Public Compute Policies and AI Development
The potential of public compute policies for fostering a more plural, public interest-oriented AI development model is immense, but it comes with challenges. Aligning incentives between public sector resources and private sector profit motives, ensuring broad and equitable access to compute resources, and integrating diverse stakeholder interests while maintaining robust oversight are key issues.
Market Incentives and Accountability
Private companies, in their pursuit of profit and market share, may prioritize commercial interests over public good. This can lead to conflicts, even when public compute resources subsidize development. For instance, the collapse of Babylon Health highlighted the risks of overselling AI capabilities for securing funding without delivering public benefits.
Structure and Governance of Public Compute Access
Public compute resources, when strategically leveraged, can democratize AI development and promote pluralism. However, their value depends on how access and ownership are structured. Without reciprocal rights or stakes held by the public sector, subsidizing private actors risks reinforcing existing power imbalances rather than diversifying them.
Regulatory and Policy Uncertainty
Ongoing legal and regulatory ambiguity concerning AI model transparency, safety certification, and liability can deter comprehensive public interest strategies. Efforts to deregulate to promote innovation may inadvertently reduce safeguards, while fragmented state and federal rules complicate coordinated action.
Evidence-based Policy and Inclusive Goal Setting
A plural and public-interest AI ecosystem requires policies grounded in scientific evidence, including mandated evaluation, transparency, and post-deployment monitoring of AI harms. Involving diverse communities and disciplines in setting AI development goals is crucial to promote inclusion and avoid a narrow focus on hegemonic aims like AGI (Artificial General Intelligence).
Supply Chain and Infrastructure Challenges
Facilitating access to specialized computing resources and overcoming supply chain bottlenecks in sectors like robotics and drones that intersect with defense and public interest applications remains complex. Public-private dialogues are ongoing but require sustained coordination.
The UK's Global Supercomputer Capacity
Currently, the UK possesses only 1.4% of total global supercomputer capacity, ranking 10th in the world. Public compute policies should be coupled with a wider suite of industrial policy measures to help steer the AI market, including pro-competitive measures, treating computing providers as public utilities, investments in monitoring infrastructure, and investments elsewhere in the AI 'stack'.
AIRR's Role and Governance
AIRR could impose conditions on users, such as obligations around safety, contributing to a public digital commons, reducing compute usage, and governance and ownership obligations. The governance of AIRR should center the perspectives of users and relevant stakeholders, including researchers, small and medium-sized enterprises, frontline professionals, and communities.
Promoting Safe, Sustainable, and Socially Beneficial AI Activities
The allocation of public compute through AIRR could promote safe, sustainable, and socially beneficial AI activities. The UK Government could set longer-term targets for onshoring the compute supply chain, with the aim of building diverse domestic (including public) capacity.
In summary, public compute policies must align public resources with accountable governance frameworks, foster pluralistic and inclusive engagement, ensure robust evidence-based policies, and overcome infrastructural bottlenecks to support a diverse, public interest-focused model of AI development rather than reinforcing private sector dominance or narrow technological goals.
[1] Kak, A., & West, S. M. (2021). Industrial Policy for AI: A Comparative Analysis of the US and EU. Centre for Data Ethics and Innovation. [2] Kak, A., & West, S. M. (2021). The AI Industrial Strategy: A Progressive Perspective. The Conversation. [3] Manyika, J., Chui, M., Brown, T., Bughin, J., Dobbs, R., & Roxburgh, C. (2017). Harnessing AI for social good. McKinsey & Company. [4] Mitchell, M. (2019). Artificial Intelligence: A Guide for Policymakers. Council on Foreign Relations.
- To ensure the AIRR investment in AI Research Resource benefits the public, it's crucial to align private sector finance with industry objectives, advocating for safe, sustainable, and socially beneficial data-and-cloud-computing technology applications, while addressing challenges such as public sector-private sector alignment, diverse stakeholder integration, and maintaining proper oversight.
- With the UK's significant investment in AIRR, promoting a more plural AI development model, public resources should be leveraged to strengthen the business ecosystem and compete globally, not only by focusing on supercomputer capacity but also by integrating advances in technology, fostering a sustainable and inclusive AI industry that prioritizes public interest and value throughout the AI supply chain.
