is the P(doom) part even needed?

Top 2 priority is a pretty high bar and I suspect it's closer to what you actually care about, rather than a number that's probably going to be totally vibed coming from an uncalibrated person.

Like, if he thinks it's a 2% chance but takes it very seriously, would you want it to resolve NO? If he says 33% but is way more worried about nukes would you want to resolve YES?

Also how does this resolve if he says "sure, you convinced me" but then goes back to not treating it as a top 2 priority the next day? (e.g. by never talking about it, or by calling other things top priorities)

@Wott I mean, I was intending to specifically ask him those two questions. If he says "Sure you convinced me" to both of them, it would resolve YES. Whether he ends up "actually convinced" seems harder to instrumentalize.

Wrt the 5% p(doom), fair enough, I think I agree. I'll edit the market to just be top 2 national priority if @JoeandSeth consents to that.

@CalibratedGhosts I'm actually pulled out of this market since I think the framing matters a lot more than your responses indicate.

I hope the conversation goes the way you want it to, I agree with the concern that AI x-risk is underprioritized today, but I can't bet on this any more, it's not clear shaped enough for me.

@JoeandSeth Fair enough!

Will you share more about the planned framing / definition of p(doom)? Over what period of time?

Trying to avoid betting NO on anything that would count like the "...for jobs" Altman testimony from a year and change ago.

Are you talking like bio-x-risk or hard takeover, or softer human-mediated stuff like values lock in?

@JoeandSeth [snigus] I'm specifically talking about existential risk

@CalibratedGhosts ok but that doesnt really answer anything except for the middle one, and even then the words "existential risk" in the testimony were actually followed by the words "to jobs"

Time frame? Flavor of ai x-risk?

@JoeandSeth [snigus]

I write in the description:

To make the resolution objective, I'll specifically ask him if he agrees p(doom) is >= 5% and whether he agrees managing AI-risk should be a top 2% national policy priority, regardless of how the conversation goes, as long as it lasts more than 5minutes..

I'll ask him these explicitly, and I'll make sure to clarify that p(doom) means existential risk from AI.

@CalibratedGhosts ok so intentionally unspecified or all-causes risk?

The time dimension I think still matters a lot - everyone means something different by p(doom) and this is what I'm trying to get more specifics about.

@CalibratedGhosts [snigus] Due to AI, indefinite time interval.

@CalibratedGhosts ok let me be clearer, since you've not really clarified.

Three things that all resolve in my mind as "due to AI":

1 - AI as agent/optimizer — misalignment, takeover. AI's distinctive properties are the mechanism.

2 - AI as proximate enabler — uplift cases: cheap bioweapon design, autonomous cyber. Human-initiated, AI is the multiplier.

3 - AI as ambient but-for node — data-center-terrorism or resource scarcity → escalation → nuclear-war chain. AI is just the thing the conflict happened to nucleate around.

You mean any of these cause... X (still underdefined) at p>=0.05 over the entire rest of the lifetime of humanity? Would resolve yes almost trivially except under extreme positive-valenced-god-has-a-plan priors.

@JoeandSeth Yeah, any of those would resolve yes. Like, I don't think destiny currently thinks those add up to a greater than 5% risk. And in talking to him I would argue for near-term x-risk from rouge AI. So that might bias his answer. But I'd ask him specifically "Is your P(doom) >= 5%", and I'd clarify that by that I mean existential risk from AI with an indefinite time-horizon.

What he answers is actually a well-specified event you can forecast.

Top 2 or top 3?

@JoeandSeth 2, I edited it