Getting through

Suppose the worry about AI is right. What would it take to get through? This path reads the strategic debate in order: why no single fix is enough, the three broad ways out, and what each one depends on. Most pages along it are cruxes, which give the strongest case on each side rather than an answer.

1. The problem

Start with The case for AI risk, step by step if you have not read it. Then AI catastrophe is overdetermined argues that getting through means clearing several separate hurdles, and lays out three routes: build it safely, don’t build it, or build up to the edge and then stop. Survival needs one route to work, and each depends on many things going right.

2. Build it safely

The first route turns on the technical problem. How hard is alignment is the central question; Will alignment happen by default asks whether ordinary training already lands near human values. Two questions decide whether there is time to fix mistakes: Can we learn from AI failures before they are fatal, and Can AI automate alignment research.

Alignment is only the first hurdle on this route. Someone else’s careless system, misuse, and unaccountable power remain; see Pivotal act.

3. Don’t build it

The second route is a pause. Can AI progress be paused sets out what a pause could mean and whether any form could hold, and Can the US and China agree to limit AI takes the hardest case. A pause only buys time argues that even a pause that holds can only delay the problem, so it should be judged by what the time is used for.

4. Build up to the edge, then stop

The third route sits between the other two: keep building, then stop just before AI becomes dangerous and use the AI already built to solve alignment. Plans for the AI transition names the options, with Plan C, “burning the lead”, as this route. When should AI development pause asks whether stopping late can work: whether anyone will know where the edge is, and whether there will be a lead to burn. The 2026 Against SURA debate is this question applied to the people inside the labs; The SURA debate turns on how doomed Plan C is says what it comes down to, and Should safety-focused labs keep building collects both sides.

5. If it works

Success on any route leaves the question of power: Who should control aligned AI. And then the question of what to aim for, which the A good future path takes up.

For how these questions are playing out in legislatures and campaigns, follow The politics of AI risk.