I recently decided to try to learn more about the AI regulations of China, the EU, and the US. Instead of taking private notes, I decided to create a webpage to organise and summarise the main things I found, with the idea that it might be useful to others. I’ve read and written about tech regulation before, and I’ve generally struggled to find accessible overviews of the topics I’m curious about. With the exception of The Future of Life Institute’s wonderful website on the EU AI Act, the same goes for AI regulation, in my experience.
In this post, I outline how I’ve structured the website and some changes I’m considering making to it. I’d love help, feedback, and suggestions.
The goal was to make structured and concise overviews of the AI regulation regimes of the main world powers. So far, I’ve only covered the EU and China. To figure out what to focus on, I wrote a brief “intro to tech regulation”, where I identified the following as key elements of a regulation regime:
The goals and priorities. High-level goals and priorities, such as safety and transparency, guide the design of the regime and its evaluation after implementation. Because so many competing interests go into the design of regulations, these overall goals and priorities cannot be specified with great precision, but must be inferred based on an understanding of the political context.
The target area(s) and group(s). A particular tech sector is comprised of an interconnected web of processes and groups that span from materials extraction to final product assembly and from companies to users. A well-crafted regulation targets those areas and groups in the sector that enable the regulation to achieve its goals.
The rules and standards. These are the specific sets of rules, standards, rights, and obligations applied to the relevant areas and groups.
Oversight & enforcement. This concerns who has jurisdiction and the authority to monitor compliance and impose sanctions when there are violations.
Sanctions. The teeth of the regulation.
The overviews are structured roughly in accordance with these elements. Rather than try to list the many rules and standards of the different regimes, I have a separate “Topics” page for how each regulation regime applies to important topics, like deepfakes and systemic risks. So far, these topic overviews only mention the rules that China or the EU apply to the issue, but I think it would be a good idea to include info about the practical implementation of the rules, such as ongoing or completed enforcement cases.
I’m considering adding a section on unenforceable or extremely-difficult-to-enforce rules, such as rules on deepfake labelling in a world of open source models. I’m also considering creating an overview page for international treaties and agreements on AI.
The AI Regulation Regimes webpage is in my view just a tiny version of what should exist, namely a well-researched, comprehensive, and accessible webpage where people can go to get up to date information about how different countries regulate AI and the rules that apply to various important issues. I think there is a path from widespread public dissatisfaction about the AI race to public pressure on policymakers to make sane and useful laws, although it’s currently not clear what that path looks like. In any case, it seems worthwhile to lower the friction that people face when trying to learn about AI regulation.