Slowing down AI progress is an underexplored alignment strategy

Link post

LessWrong user “Norman Borlaug” with an interesting post on reducing existential risk through deliberate overregulation of AI development:

In the latest Metaculus forecasts, we have 13 years left until some lab somewhere creates AGI, and perhaps far less than that until the blueprints to create it are published and nothing short of a full-scale nuclear war will stop someone somewhere from doing so. The community strategy (insofar as there even is one) is to bet everything on getting a couple of technical alignment folks onto the team at top research labs in the hopes that they will miraculously solve alignment before the mad scientists in the office next door turn on the doomsday machine.

While I admit there is at least a chance this might work, and it IS worth doing technical alignment research, the indications we have so far from the most respected people in the field are that this is an extremely hard problem and there is at least a non-zero chance it is fundamentally unsolvable.

There are a dozen other strategies we could potentially deploy to achieve alignment, but they all depend on someone not turning on the doomsday machine. But thus far we have almost completely ignored the class of strategies that might buy more time. The cutting edge of thought on this front seems to come from one grumpy former EA founder on Twitter who isn’t even trying that hard.

From Kerry Vaughan’s Twitter thread:

I’ve recently learned that this is a *spicy* take on AI Safety: AGI labs (eg OpenAI, DeepMind, and others) are THE CAUSE of the fundamental problem the AI Safety field faces.

I thought this was obvious until very recently. Since it’s not, I should explain my position.

(I’ll note that while I single out OpenAI and DeepMind here, that’s only because they appear to be advancing the cutting edge the most. This critique applies to any company or academic researcher that spends their time working to solve the bottlenecks to building AGI.)

To vastly oversimply the situation, you can think of AI Safety as a race. In one corner you have the AGI builders who are trying to create AGI as fast as possible. In the other corner, you have people trying to make sure AGI will be aligned with human goals once we build it. If AGI gets built before we know how to align it, it *might* be CATASTROPHIC.

Fortunately, aligning an AGI is unlikely to be impossible. So, given enough time and effort into the problem, we will eventually solve it.

This means the actual enemy is time. If we have enough time to both find capable people and have them work productively on the problem, we will eventually win. If not, we lose. I think the fundamental dynamic is really just that simple.

AGI labs like OAI and DeepMind have it as their MISSION to decrease the time we have. Their FOUNDING OBJECTIVE is to build AGI and they are very clearly and obviously trying *as hard as they can* to do just that. They raise money, hire talent, etc. all premised on this goal.

Every day an AGI engineer at OpenAI or DeepMind shows up to work and tries to solve the current bottlenecks in creating AGI, we lose just a little bit of time. Every day they show up to work, the odds of victory get a little bit lower. My very bold take is that THIS IS BAD

Now you might be thinking: “Demis Hassabis and Sam Altman are not psychopaths or morons. If they get close to AGI without solving alignment they can just not deploy the AGI.” There are a number of problems with this, but the most obvious is: they’re still robbing us of time.

Every. Single. Day. the AGI labs are steadily advancing the state of the art on building AGI. With every new study they publish, researcher they train, and technology they commercialize, they also make it easier for every other AGI lab to build and deploy an AGI.

So unless they can somehow refrain from deploying an unaligned AGI and stop EVERYONE ELSE from doing the same, they continue to be in the business of robbing humanity of valuable time.

They are the cause of the fundamental problem faced by the AI Safety community.

In conclusion: Stop building AGI you fucks.

Notably, a number of people in the AI Safety community basically agree with all of this but think I shouldn’t be saying it. (Or, at least, that EA Bigwigs shouldn’t say it.)

I obviously disagree. But it’s a more complex question which I’ll reserve for a future thread.

Elon Musk has been vocal on the need to regulate AI development, even if it includes regulating (and slowing down) his own companies:

All orgs developing advanced AI should be regulated, including Tesla