Re: Leslie's "The End of the World"

Nick Bostrom (
Thu, 13 Aug 1998 22:02:22 +0000

Robin Hanson wrote:

> I think Mark has a point. Say you knew that all intelligence was either bio or
> postbio, with bio intel living for days to decades, and postbio intel living
> from thousands to billions of years. Then you expect that conditional on
> being a bio intel, you are typical, so bio intel shouldn't last much more than
> a few decades more (assuming continued exponential growth). And conditional
> on being a postbio intel, you know you must be one of the very first

Why? If being a postbio intel is all you conditionalize on, this doesn't follow. If I knew nothing except that I was a postbio intel, I wouldn't expect to be one of the very first; I would expect to be one of the middle 95%.

>, so you 

> couldn't say much about how many more postbio intel observer moments there will
> be. And without introducing more info, you couldn't say much about your chances
> of being postbio vs. bio intel.

If postbio and bio intelligences all belong to the same reference class (just as tall people and short people, black and white etc.) then the fact that you find yourself being a bio intel. would indicate that the postbios don't vastly outnumber the bios.

Whether postbios and bios should be in the same reference class is not clear. That's a deep problem, and I suspect but am not yet certain that there might be some kind of loophole for the transhumanist here, if the postbios are fundamentally different from bios.

> In general the validity of the doomsday argument depends on how much detailed
> information one includes about the nature and history of life.

I don't see why that should be the case. Of course, the exact conclusion depends on how you estimate future population figures etc., but in general I see the DA as, so to speak, *superimposed* on whatever other evidence you might have.

> With the
> very stark models that are easy to construct, little info is reflected in
> the model, and so the doomsday conclusion bites. But I think that the
> conclusion bites much less as we put more info into the model.

That depends on what this info would show. It's easiest to explain what I mean if there are only two possibilities: Doom Early, or Doom Late. If your prior evidence led you to be extremely certain that there was no way that doom could happen soon (no meteories, no germ warefare, no black goo, no vacuum decay etc.), then after taking the DA into account you might still be fairly certain that Doom Early would not happen. (Exactly how certain would depend on how many observers Doom Late says there will have been.)

However, it doesn't matter how much information you have based your empirical prior on. If you have too little information, that should be reflected in your priors by setting them closer to fifty-fifty. That situation, as far as the doomsday argument is concerned, is identical to if you have fifty-fifty priors that are based on a lot of information.

Nick Bostrom
Department of Philosophy, Logic and Scientific Method London School of Economics