Re: on simply being an SAS (and UDA)

From: Marchal <marchal.domain.name.hidden>
Date: Wed Feb 9 02:37:20 2000

> However, I do invoke computation (as opposed to just stucture) to
>explain why we should not expect just a random state. Even if all
>computations occur, most strings would just be random (e.g. junk code
>after a program ends), but those involved in computations would not.

False, you have a majority of programs with junk strings as *input*,
and those inputs are involved in the computations.
This show that 'our program' should be stable for the randomness
of the data descrbing implementations of ourselves below our
substitution level.

> Bullshit. You are clearly relying on infinitely long programs as
>that is the only way you could get such numbers with your "equivalence"
>assumption. So you will have lots of white rabbits.
> By contrast, in my view, the shorter programs dominate because
>they have a number of copies that goes as 1/2^l, which is due to the fact
>that shorter programs leave more room for junk code. This eliminates WRs.


You eliminate past/objective WRs. You still don't eliminate
futur/subjective WRs. You eliminate 3-WR, not 1-WR.

Of course you are denying the distinction between first and
third person. So I guess you are vaccinated to the conclusion of UDA.

I am not relying on infinitely long programs. But I am relying a priori
on the state of finite programs in front of arbitrary complex inputs.

I have not the choice: this comes from the fact that the evolution of the
first person mental state linked to a computational state depends on
the infinite set of possible reconstitutions in UD*.

If we are both correct (which is possible although I doubt it) then
the solution of your implementation problem will lead you to a
very low level implementation: like having the universe as (generalised)
brain.

>> But how many steps make a computational history?
>
> One.

So a single computational state constitutes a comp. history !?!
(Or perhaps you mean one step = two computational states).

In any case you are in trouble, at least with COMP + CUD (there
exists a concrete UD running integraly in ``our cosmos").
Because that concrete UD will build all implementations
of your actual states + any inputs. That is at least k^q, with
k= the number of inputs any sensory cells can distinguish, and q =
the number of censory cells (and censory cells defines the border
of your concrete implementation relatively to a neighborhood.
And without abandoning comp I don't see how 'new constraints' on
implementations will help, because the concrete UD will run all of
these.
(This reasoning works also for 'abstract cells' defining our relative
implementation, like position of our electrons for instance).

Bruno
Received on Wed Feb 09 2000 - 02:37:20 PST