[extropy-chat] Analyzing the simulation argument

[Eliezer Yudkowsky]

Dan Clemmensen wrote:
> 
> However, we now go back to your premises A and B. Unfortunately, they 
> are falsifiable only in the containing "real" universe, not in the sim, 
> so we are still in trouble. You have committed a level shift from the 
> rule set of the sim to the rule set of "reality." This in turn requires 
> yet another unfalsifiable assumption: that statements of probability 
> derived from observations in the sim somehow apply to the containing 
> world.

The Simulation Argument says that, under certain premises, the conclusion 
that this world is probably real is *inconsistent* because this world taken 
at face value would contain more simulations of ancestors than ancestors. 
Whether the Red Pill Universe matches our own is not specified by the SA; 
what SA just says is that if you accept the premises as 
probably-true-at-face-value given the face value of our universe, you are 
forced to the conclusion that something other than face value applies - 
that we are a computer simulation within an enclosing world that may or may 
not resemble our own; or that our world is base-level and some other, 
unknown and unpostulated force forbids all simulations or sharply reduces 
their frequency.

> I must therefore apply falsifiability and Occam's razor to your 
> entire resulting logical structure, including its premises rather than 
> to the conclusion, I reach the same result: your additional assumptions 
> increase the complexity without changing the result, so I will stay with 
> the null hypothesis.

That doesn't help if it leaves your "null hypothesis" (the world at face 
value) inconsistent with itself.  Occam's Razor cannot rescue an 
inconsistent hypothesis.

Drawing on the terminology from "A Technical Explanation of Technical 
Explanation":  The Simulation Argument confesses its ignorance of the 
enclosing universe and leaves the Simulation Hypothesis vague.  SA says 
that, if the SA premises are implied by the "face value" hypothesis ~SH, 
then ~SH is *stupid* with respect to the data.

It is a theorem in classical logic that if not-P implies P then P.

Similarly, the Simulation Argument is that ~SH implies SH, therefore SH.

More precisely, the Simulation Argument is that the Face Value Hypothesis 
FV implies a Simulation Hypothesis SH1 which implies ~FV, therefore ~FV. 
We cannot conclude SH1, but we can conclude ~FV.  We can't conclude that 
our world is a computer simulation; but we can conclude our world cannot be 
taken at face value because it is inconsistent to suppose that our world is 
probably the bottom layer of reality.

Still more precisely, the Simulation Argument says that to deny this 
conclusion you must deny one of the premises of the Simulation Argument.

-- 
Eliezer S. Yudkowsky                          http://singinst.org/
Research Fellow, Singularity Institute for Artificial Intelligence