## Building Blocks for AI Governance

## Source details

**Canonical URL:** [Building Blocks for AI Governance](https://www.imf.org/en/publications/fandd/issues/2023/12/pov-building-blocks-for-ai-governance-bremmer-suleyman)

## Other formats

- [Markdown version](/en/publications/fandd/issues/2023/12/pov-building-blocks-for-ai-governance-bremmer-suleyman/index.md)
- [Structured JSON version](/en/publications/fandd/issues/2023/12/pov-building-blocks-for-ai-governance-bremmer-suleyman/index.json)
- [Bundle manifest](/en/publications/fandd/issues/2023/12/pov-building-blocks-for-ai-governance-bremmer-suleyman/bundle-manifest.json)

## Bibliographic details
- Authors: Ian Bremmer, MUSTAFA SULEYMAN
- Published: December 1, 2023

---

### Executive summary
- Artificial intelligence will open people’s lives and societies to groundbreaking scientific advances, unprecedented access to technology, toxic misinformation that disrupts democracies, and economic upheaval.
- AI will trigger a fundamental shift in the structure and balance of global power, creating an unparalleled challenge for political institutions around the world.
- Policymakers must adhere to five guiding principles to govern AI effectively.

### The AI power paradox and technological dynamics
- AI is hyper-evolutionary: its rate of improvement will far surpass Moore’s Law (doubling of computing power every two years).
- The amount of computation used to train the most powerful AI models has increased by a factor of 10 every year for the past 10 years.
- Processing that once took weeks now happens in seconds; foundation technologies will become smaller, cheaper, and more accessible.
- AI systems may soon be capable of quasi-autonomy and may become the first technology with the means to improve on itself.
- Proliferation characteristics:
  - AI algorithms are far easier and cheaper to copy and share (or steal) than physical assets.
  - Models will soon run on smartphones as computing gets cheaper.
  - Marginal cost of models and marginal cost of delivery is zero, meaning once released, AI models can and will be everywhere.
  - Most models will be safe; but a single malign or “breakout” model could wreak havoc.

### Geopolitical and commercial incentives that undermine governance
- Dual-use nature: AI’s civil and military uses are blurred, but unlike nuclear enrichment, AI’s low cost enables endless deployment for civil or military use.
- AI supremacy is a strategic objective of every government and company with resources to compete; the US and China both see AI supremacy as a strategic objective that must be achieved—and denied to the other.
- This zero-sum dynamic incentivizes acceleration of AI development rather than slowing it down.
- Control and enforcement challenges:
  - Every aspect of AI is currently controlled by the private sector.
  - A handful of large tech firms control AI today, but gradual proliferation may bring many small players into the space, complicating governance.
  - Private businesses and individual technologists have limited incentive to self-regulate.
- Conclusion: The combination of technological features and incentives renders traditional governance models inadequate.

### Five guiding principles for AI governance
- Precautionary
  - The risk-reward profile of AI is asymmetric; policymakers must guard against potentially catastrophic downsides.
  - The precautionary principle needs to be adapted to AI and enshrined in any governance regime.
- Agile
  - AI governance must be as agile, adaptive, and self-correcting as AI is fast-moving, hyper-evolutionary, and self-improving.
  - Static policymaking structures that prize stability and predictability will not work.
- Inclusive
  - Regulators must collaborate with the commercial sector; the only way to properly oversee AI is to collaborate with private technology companies.
  - Governments should make companies parties to international agreements to reflect the borderless nature of AI.
  - Including private companies in high diplomacy may be unprecedented but excluding them would doom governance structures.
- Impermeable
  - Given AI’s ease of proliferation, compliance mechanisms should be watertight, with easy entry to compel participation and costly exit to deter noncompliance.
  - One defection from a regime could allow a dangerous model to escape.
- Targeted
  - AI governance must be targeted, risk-based, and modular rather than one-size-fits-all.
  - Develop a live, working taxonomy of discrete potential AI impacts to determine which tools target which risks.

### Institutional implications and recommended approach
- AI governance must account for unpredictability: rules passed today may not be effective or relevant in months or years.
- Avoid boxing regulators with inflexible regimes; instead build institutions and conceptual frameworks that enable adaptive policy.
- Policymaking should avoid a false binary between progress and doom; the goal is to stimulate innovation while avoiding dangerous proliferation and geopolitical arms-race dynamics.
- The international community needs to agree on a conceptual framework for how to think about AI; the five principles are offered as a start.

*Ian Bremmer and MUSTAFA SULEYMAN, F&D Magazine, December 2023*

---

## Content in this bundle

- **Point of View: Building Blocks for AI Governance**
  - [Point of View: Building Blocks for AI Governance (Markdown version)](/-/media/files/publications/fandd/article/2023/december/pov-bremmer-suleyman.pdf.md){rel="alternate" type="text/markdown"}
  - [Point of View: Building Blocks for AI Governance (PDF)](/-/media/files/publications/fandd/article/2023/december/pov-bremmer-suleyman.pdf){rel="external" type="application/pdf"}

---

_Source: https://www.imf.org/en/publications/fandd/issues/2023/12/pov-building-blocks-for-ai-governance-bremmer-suleyman_