[ExI] Fwd: [tt] The Habitable Epoch of the Early Universe

[Anders Sandberg]

On 2013-12-12 13:22, Eugen Leitl wrote:
> On Thu, Dec 12, 2013 at 12:48:42PM +0000, Anders Sandberg wrote:
>> Hmm, depends on if you believe the self sampling assumption (SSA) or 
> A theory which starts with beliefs is not much of one.

Hmm. Please show me one without any beliefs beyond the empty theory. 
Remember that axioms are beliefs.

> "All other things equal, an observer should reason as if they are randomly selected from the set of all possible observers."
>
> Self-observation is the opposite of random just as selection
> is the opposite of mutation.

Exactly. This is why anthropics is so headache-inducing: you get biased 
probabilities just by being around, and the two assumptions give 
somewhat contradictory answers to how you should debias your probabilities.

> Out of two cases, where you're common as dirt or rare
> as Pt-190 diadem, the monoisotopic diadem will without
> fault find itself in its fine lustrous glory even if it's the
> only one in the whole universe.

Exactly. And now the diadem needs to think about what probability it 
should assign to the existence of other diadems. The diadem using the 
SSA should think there are a lot of diadems, while the SIA-believer 
diadem should favour theories where there are loads of observers (but 
not necessarily diadems). However, the unique diadem observer is also an 
"anomalous observer" - most observers are by definition common, and they 
would get things more right when reasoning with the SSA.

Reasoning from no data or a single, biased data point will of course on 
average produce much worse fits to reality than reasoning from more 
data. But that doesn't mean it is irrelevant - if you have no data. It 
is better to make the argument clear and the assumptions overt.

Same thing with Bayesian probability: yes, it involves subjective 
judgements (constrained by logic, data and  probability theory), but it 
allows you to think about somewhat important things like Pr(humanity 
going extinct|current actions) that frequentism cannot handle at all.

-- 
Dr Anders Sandberg
Future of Humanity Institute
Oxford Martin School
Oxford University