The Most Effective Way to Regulate AI Is to Govern Compute

A controversial but highly effective approach to controlling the use of AI + Conversation with Claus Aasholm

YouTube

Here is my recent conversation with the knowledgeable

Claus Aasholm

who is an experienced analyst of the semiconductor industry.

We discuss the global semiconductor supply chain, Nvidia's market dominance, TSMC's origins, the geopolitics of chips manufacturing, and how it affects the AI industry.

I learned a lot from this conversation and you will too.

---

The Most Effective Way to Regulate AI Is to Govern Compute

Introduction 

Regulating AI through governing compute would be the most direct and effective approach to contain and control AI. The approach is also borderline Orwellian as it ultimately grants the government monopoly powers over the technology. 

This week’s post is based on a thought-provoking paper submitted to arXiv on February 13, titled “Computing Power and the Governance of Artificial Intelligence”. It was written by nineteen authors including deep learning pioneer, Yoshua Bengio. The lead authors are Girish Sastry from OpenAI, Lennart Heim from Centre for the Governance of AI, and Haydn Belfield from University of Cambridge. 

In this post, I highlight what I found to be the paper’s most important parts and points. My post here is a bit more technical than my usual writings but hopefully, at least some of my subscribers will take something away from it.

If you are not a paying subscriber and want to learn more, you can check out the official blog post here.