Solve Alignment in 1 Hour

An algorithm for improving your model of AI X-risk

​

1. Maker Phase: Write down your best guess / proposal for how things go well.
2. Breaker Phase: Dig into the details until you understand how the proposal fails.
3. Iterate: if you can find a fix, go back to step 1.

Example

​

1. Maker Phase: Instead of building a system that acts in the world, build a system that just tries to make good predictions (an oracle). This should avoid existential risk from AGI, because an oracle has no agency and thus no reason to act against us.

Example

​

1. Maker Phase: Instead of building a system that acts in the world, build a system that just tries to make good predictions (an oracle). This should avoid existential risk from AGI, because an oracle has no agency and thus no reason to act against us.
2. Breaker Phase: (Dig into the details) What exactly are you planning to do with this oracle? What happens when the runner-up AI lab builds an agentic AGI 6 months later?

Example

​

• Maker Phase: Instead of building a system that acts in the world, build a system that just tries to make good predictions (an oracle). This should avoid existential risk from AGI, because an oracle has no agency and thus no reason to act against us.
• Breaker Phase: (Dig into the details) How will you build this oracle? Does that process actually result in a non-agentic oracle?

Getting Stuck

​

I can’t come up with an alignment solution. (Don’t worry, this is a common problem)

​

→ Write down what you expect to happen if we develop AGI.

​

→ If that inevitably ends in doom: start with step 3 (write scenario down in detail, then try to find a fix).

​

Getting Stuck

​

AFAICT, the solution works.

​

→ Consider different kinds of ways your proposal might fail (cf List of Lethalities)

​

→ Get others to critique your proposal. (Write it up)

Getting Stuck

​

The proposal is unfixable.

​

→ Talk to others about your idea.

​

→ If that fails: congratulations, you have completed the exercise! Please start again from scratch with a new idea :)

Let’s go!

​

Open the linked google doc and do the exercise:

• Maker Phase: Write down your best guess / proposal for how things go well.
• Breaker Phase: Dig into the details until you understand how your proposal fails.
• Iterate: if you can find a fix, go back to step 1.

If you get stuck:

​

• Write down what you expect in detail (if still stuck, start with step 3).
• Get other people to critique (or consult list of lethalities).
• Proposal unfixable? Yay! Start again.