my book is now an actual book! *Architectures of Global AI Governance* offers a history, a framework, and a roadmap for AI: Q. does this book answer all my #AIgovernance questions, everywhere, all at once? A. yes Q. Will this book stop my door? A. Yes, barring some vault doors

"The Origins of Efficiency" is out today!

Or get the whole book (Open-Access at: https://t.co/8aqdxALX7F / print at https://t.co/q2W0nQPzmO )

By reviewing the advantages, drawbacks, and open questions for each avenue, I aim to provide a definitive account of where global AI governance has come from; the challenges to overcome, and the choices that face us next. Read Chapter 3 at:

the chapter then charts in exhaustive detail the tools and pathways that have been proposed for the way forward. Industry self-governance? Soft-law standards and codes of practice? Unilateral state action? Should we apply existing international law, or advance new treaties? ...

it covers it all, from the OECD AI Principles to the Partnership on AI; from the ITU AI For Good Summits to the EU AI Act; from the G7 Hiroshima Process to the AI Safety Summits; from the Council of Europe's Framework Convention to the GC-REAIM...--and more...

It maps the emergence, growth and fortunes of three different tracks in the global AI governance architecture, tracing efforts focused on governing military AI systems (since 2009), conventional narrow AI systems (since 2015), and advanced general-purpose AI systems (since 2022)

What is the trajectory of global AI governance so far? What institutions, initiatives & instruments have been advanced this past decade and a half? Which actually was the *best* AI Safety Summit? Where do we go from here? My book's 📖 Chapter 3 answers this and more across >100pg

sunrise

Or the whole book (Open-Access PDF at: https://t.co/8aqdxALX7F / print at https://t.co/q2W0nQPzmO )

in sum, it's a one-stop shop for all the hot challenges facing the world in AI. Use it to position your AI policies; use it in class; give it to your toddler (mine loves it I swear)! Get it at: Chapter 2: https://t.co/JiRITEhDAF ...

...and I argue that for many AI issue clusters, global governance regimes are challenging but remarkably feasible to enforce ...

...finally, I explore the *feasibility* of enforcing any global regimes for AI (once established), by mapping different actors, industries, resources and artefacts across the AI supply chain--and identifying what transnational regulatory levers are available for each...

In response, I explores potential *rationales* for (specifically) global governance (rather than local responses), by considering which of these issues qualify as collective action problems, and what are their resulting features as global public goods...

... I cover old and new debates, and topics including: military use of LLMs, (social) AI assistants' systemic impacts on the world's epistemic, democratic & social commons, threat models for AI-enhanced biorisk, hacking, loss-of-control; and what is distinct about AI agents...

I provide a detailed analysis of how these systems and their applications differ technically and politically, and discuss recent (and live) debates over the impacts and risks of these types of AI systems, and how they can be rigorously evaluated...

📖 Chapter 2 of my book, 'Scaling Law for AI', charts the space of global #AIgovernance issues raised by conventional (i.e. narrow), military, and advanced (i.e. general-purpose/frontier) AI systems; and examines rationales for- and enforceability of specifically global responses

amazing post, totally overlooked. Mark's sheer patience probably got him an early look at Gemini 3. One of the only evals I buy is "world expert (here, in palaeography, reading cursed old handwriting) does a stress test and is very impressed"

Read chapter 1 at: https://t.co/p42oJYK21r Or get the whole book (full PDF Open-Access at: https://t.co/8aqdxAMuXd / print at https://t.co/q2W0nQQ7cm )

This is a history in the making, so rather than preach, I hope to enable greater nuance and shared understanding about the assumptions, intuitions, comparisons, and evidence bases that people depart from in debates over whether AI is imminently transformative or a complete bubble

... and evaluates these for two near-term future scenarios: (1) further capability progress at the AI frontier, or (2) further proliferation of today’s algorithmic state of the art. Either scenario would have quite transformative impact; both remain plausible.