AI 2040: Plan C — Burn the Lead

Vince So, we've been following this series on AI 2040, and last time we covered Plan A, which was all about an international deal to slow down superintelligence development. Now, we're looking at Plan C, which they call 'Burn the Lead'.

Ava That's right. Plan C's thesis is that the US unilaterally spends, or 'burns,' its lead by pausing for a few months during the intelligence explosion to do alignment and control work, without sabotaging China and without securing a verified deal.

Vince The idea here is that by 2030, the government can see that AI R&D is about to be fully automated, and they try to get a deal with the leading company to pause for safety work. But the companies and China resist, and it ends up being a political pressure play.

Ava The authors judge Plan C as better than Plan D, the full race to superintelligence, but still a bad plan. They give three reasons: first, 'burning the lead' for only a few months probably buys too little time; second, even if the resulting superintelligences are robustly aligned, the question of who they are aligned to remains alarming; and third, the risk of World War III is too high.

Vince I think what's interesting here is that Plan C is essentially a unilateral action by the US, which makes it different from Plan A. But the authors still see it as flawed because it skips the verified, transparent, international deal.

Ava Yeah, and that's where Plan A comes in – a positive vision for delaying superintelligence development until 2040 through total research transparency and verification. Plan C, on the other hand, seems to be more of a 'quick fix' that might not actually solve the problem.