In the last unit, we explored what defences need to be build in order to defend against AI harms.

Very few people in the world are trying to steer the development of AGI towards beneficial outcomes for humanity. We estimate it’s less than 2,000 full-time people!

You could be one of the few people making a different, but doing so requires stepping up, taking responsibility, and trying to solve new problems that don’t have an instruction manual.

In this unit, you’ll prioritise a single intervention, you’ll do a shallow investigation into it, then you'll scope it down to something achievable in the near-future. 

During your group discussion, you'll present your research to your peers.


Interventions

Based on everything you've learnt about so far during this course, prioritise an intervention from the long list that you believe could work against the kill chain you developed in Unit 3.

Note: If your Unit 3 threat doesn't clearly map to any intervention in the list, we encourage you to generate your own intervention idea.


Prioritise a single intervention

Your own research

Focus on understanding the technical mechanism and one real-world implementation attempt.

Research both the general intervention and any attempts at your specific scoped version.

Spend ~1 hour on this.


Research your chosen intervention 

How could you make this intervention smaller?

What version of this could be achieved in 6 months, or in a specific context or location?

If the intervention is already quite specific (e.g., pandemic response teams), scope it to a specific geography, organisation, or pilot programme. If it's very broad (e.g., AI alignment), narrow it to a specific technique, context, or implementation.

E.g. reducing the scope from "establish a global compute governance regime" to "the US government passes legislation saying any training runs above 10^26 FLOPs must be pre-registered with NIST”

Spend ~20 mins on this.


Scope down your intervention

Your own analysis

Write 2-3 paragraphs describing:
- The theory of change - how it's meant to reduce harms from AI
- Who would implement it and what resources they'd need (e.g. AI companies, governments, civil society groups, members of the public, tech entrepreneurs)
- Any existing attempts or pilots of this approach


What is your scoped-down intervention?

Why isn't this intervention already deployed at sufficient scale and effectiveness to prevent your Unit 3 threat? 

Write 2-3 paragraphs explaining why this intervention isn't already successful.
- What specific actors or forces oppose it (and why)
- What technical/logistical challenges exist
- What makes this harder than similar successful interventions
- The deeper reason this hasn't happened yet (not something vague like "lack of awareness")


Diagnose the failure

E.g. it’s deployed at all major AI companies, or most governments have taken a specific action, or this technology is widespread (if it’s good) or halted (if it’s bad).

Write 2-3 paragraphs describing what the world looks like if this intervention succeeds wildly in 2-3 years:
- What specifically exists that doesn't exist today
- What measurable changes have occurred
- How the AI risk landscape has shifted
- How does this reduce risks from threats like your Unit 3 scenario?


AGI Strategy

Interventions

Resources (15 mins)

Exercises