[ExI] Computational equivalence (Was: FTL drive)

[Anders Sandberg]

Stefano Vaj wrote:
> We have known for a while now that more or less everything (starting 
> from the level of complexity of Wolfram's automata) computes things, 
> and that the real difference between the original IBM PC, a Chinese 
> Room, a human brain and a godlike computronium Jupiter brain is 
> essentially one of performance at a given task.

However, that performance matters a lot. My chair is in a sense 
implementing my mind (and yours), but not well enough that any of us 
would sacrifice our lives happily knowing we will survive in the chair. 
The lack of performance in the real world is so complete that we do not 
care at all about the mental states implemented in random objects.

There are also major differences in the abilities between computational 
systems; I am not convinced at all by Wolfram's reasoning - he is 
essentially just arguing that everything is equivalent to 
Turing-machines, and since any of them can compute any other (with a bit 
of overhead) they are the same. The snags are that 1) the overhead can 
be *enormous* - constant factors in theoretical computer science are 
often amazingly large numbers, the kind that cannot even be expressed 
with exponent notation. 2) That a Turing machine can compute program X 
and Y doesn't mean X and Y are equivalent in power [*], and one might be 
tremendously slower on the machine despite being faster on some other 
machine.

So implementing ones mental processes on the right substrate matters: do 
it on the wrong substrate and you cannot interact efficiently with the 
real world, or even think within the history of the universe. Solid 
state societies seem to have advantages here (variable speed, 
compactness, resiliency, resource use) in many domains that matter if 
one wants to go to space.


[*] A nice new paper by Shane Legg about an approximate universal 
intelligence measure,
http://www.vetta.org/2011/11/aiq/
shows that even for the small cases we can compute readily, there are a 
clear ranking of intelligences between algorithms (anecdotally, Shane 
can outperform AIXItl on this task so far...)

It is also the first *serious* application of the Brainfuck language I 
ever seen.


-- 
Anders Sandberg,
Future of Humanity Institute
Philosophy Faculty of Oxford University