National AI governance ecosystems

Today, I spoke at Aaron Maniam’s class at Blavatnik School of Government, University of Oxford about national AI governance ecosystems.

Institutional approaches

I outlined three focus areas of institutional approaches I’ve noticed countries using:

Research-focused: first-wave AI Safety Institutes (UK, US, Japan, Singapore) who have focused on evaluations, testing, and “advancing the science of AI safety”

Regulators: EU AI Office, China’s ecosystem (led by the Cyberspace Administration of China), and 🇰🇷South Korea’s AISI given its newly-established responsibilities under the Korean AI law. 🇧🇷Brazil is likely to join this club once it passes its AI bill.

Industry-focused: light-touch government involvement focusing on boost national AI business. 🇫🇷France is the primary example. Another one might be 🇮🇳 India, depending on which approach it chooses.

Key institutional elements

Stepping back, I described some of the key institutional elements that underpin these approaches:

Inputs available for governments (at varying degrees):
• Public sector investment
• Economic stimulus
• Legitimacy and power
• Expertise
• Pressure

Outputs to aim for given these tools:
• Research
• Innovation
• Best practices/standards
• Enforcement

Institutional trade-offs

Inevitably, governments can't cover everything. They need to choose which inputs to prioritize, and which outputs to aim for. I described some of these institutional trade-offs that governments have been facing:

Regulatory powers: regulate vs let industry self-govern
Scope: frontier AI (natsec, critical infrastructure) vs broad coverage of applications across sectors
Statutory footing: legislate or not
Interoperability: prioritize domestic needs vs international harmonization
Originality: create a new agency vs adapt a new one

I simplified many aspects in the slides for educational purposes (the curse of breadth over depth), but hopefully they are helpful as a primer. I wished I had covered more approaches, such as the UNESCO-inspired national strategies of countries like Saudi Arabia – but luckily I was followed by a great presentation by Abdulrahman Habib that provided much more depth into a particular ecosystem.

I felt really energized by the students’ interest in thinking about the government’s role in ensuring AI is developed safely. Very grateful to Aaron Maniam for the invite!

Previous
Previous

Open roles in AI policy #9

Next
Next

Open roles in AI policy #8