Next

This page was last updated on 11 December 2024.

This space is reserved for questions and ideas I want to explore (or at least I want them explored even by other people). I am treating this space as a living document. The things written here reflects what I’m currently thinking about and how I might be thinking about them. I could be wrong about certain assumptions, as I’ve been countless of times before.

Note that I have not done a comprehensive literature for some of these so it would be good to check the existing literature before starting work on any of them or believing it as is.

If you are interested in working on any of these ideas, notify me at mail@lenz.wiki.

An AI safety agenda for APAC/SEA

I am now part of the set of people who believes that long-term AI risks are becoming more and more of a near-term concern. While frontier AI safety within AIS hubs are obviously very relevant, I do believe there’s more of an urgency to amplify the concerns of nations in the east given current developments in the field. More specifically, I find that concerns for low-middle income countries (LMICs) when it comes to emerging technologies, or AI specifically, are different from the concerns of the rest of the world. Some interesting projects can be:

• Mapping how innovation systems in APAC are interacting with AI advancement and adoption in the region. As someone who is working within my local startup ecosystem, I do believe there’s a big role that funders like VCs and government agencies play in both the innovation agenda, as well as the actual regulation of a technology (at least within SEA). I’m curious about modes of regulation from a business perspective.
• Understanding our role in the AI supply chain from chips to tokens. A lot of compute power is very concentrated on specific countries (or more accurately, specific companies). East Asia, along with Southeast Asia, has been increasingly vital in the supply chain. I would imagine figuring out technical governance from this perspective is now also increasingly vital.
• Making progress in multilingual unlearning. Context varies per language, and apparently, so does safety training. Guardrails may not be as effective in another language than it is in English which is incredibly concerning when you think about how there’s around 1800 languages in SEA alone, and way more in the rest of APAC.

Field-building in AI safety to have a better talent pipeline for LMICs

I also want to (very specifically) have a section allotted for field-building initiatives that I think are promising. I think we are entering a space where local priorities AI governance research may be relevant in the next 5 years or so. We also have lots of technical talent in LMICs that could have promising contributions given their industry expertise and skills. Some things I think people can do to help others transition towards AIS are:

• Creating digestible AIS educational materials. Content like the ones by Rob Miles seem to be very promising in keeping people informed about the risks of AI development. Policy briefs or informal educational documents can also keep policymakers updated on the technological scenarios that lie ahead and their economic implications.
• Organizing camps and hackathons about AI safety. Unlike the US or UK, there’s way less opportunities to build career capital in AI (safety) in LMICs. These sorts of project-based program can help people explore their fit as soon as they finish an AISF course. It seems to also be effective in getting people to work on AI safety without getting stuck in “tutorial hell.”

General AIS projects that I think are promising

Many people have written about the following. Perhaps I’ll fix my thoughts on this when I have more bandwidth.

• Figuring out alternatives to evidence-based policies
• Autostructures and explorations of live theory (in governance)
• Scenario planning for multipolar failures
• Exploring cost-effective interventions for robust agent-agnostic processes
• A compilation of strategies for making AI go well (a comprehensive taxonomy maybe)

Non-AIS projects that would be really cool if it were real

• An IP broker in SEA
• A digital notebook that helps you run simulations of your diagrams
• Any project that promotes better access to plant-based food without breaking the bank