Re: Tolerance strategies (was: Two Essays on the violence...) (fwd)

From: Eliezer S. Yudkowsky (
Date: Thu Oct 25 2001 - 06:09:43 MDT

"Alex F. Bokov" wrote:
> To the Cc'd recipients...

You realize, of course, that you have now obligated yourself to forward on
my responses as well? <smile>.

> On Wed, 24 Oct 2001, Eliezer S. Yudkowsky wrote:
> > "Alex F. Bokov" wrote:
> > >
> > > Given that intolerant and absolutist memes can by definition exist
> > > within a tolerant and relativist society, but not vice versa, what
> > > strategies can the tolerant and relativist society use to survive?
> > > Will these strategies require it to relinquish some of its tolerance
> > > and relativism?
> >
> > The tolerant and relativist societies can maintain substantial military
> > superiority using superior economic and technological capabilities.
> > That's why Afghanistan hasn't conquered America.
> 1. This is criticized as a contradictory stance-- not only a
> relinquishment of tolerance and relativism, but a selective
> relinquishment.

Isn't this kind of like saying that "Tit for Tat" is bad because it's a
selective relinquishment of cooperation?

> In order to maintain this strategy, it is shielded
> from public debate and scrutiny.

Huh? Why does the use of a "Tit for Tat" policy need to be shielded from
scrutiny? And I wasn't even proposing that... just noting that
Afghanistan simply does not possess the military capability to wipe out
the US, no matter how much they may wish to do so, and that it is not a
coincidence that we are a tolerant democratic society and they are not.
Whether our current attack on them is justified is an entirely separate

We are not in danger of being wiped out by intolerant societies because
intolerance is an economic disadvantage is a military disadvantage. We
thus do not need to relinquish any of our tolerance to survive. Indeed,
such a strategy would be counterproductive.

> 1a. Furthermore, when an ends-justify-means calculus is employed, you
> end up with the problem of entrusting somebody with the priviledged
> position of deciding which ends justify which which means. Your
> friendly AI might do it Eliezer, but even the best intentioned humans
> inevitably lapse into "MY ends justify the means".

There are three settings on the dial here, not two. It's not "Hitler" and
"Gandhi"; it's "Hitler", "Churchill", and "Gandhi". The Churchill form is
widely recognized as not being equivalent to the Hitler form.

In a two-player game, then yes, both parties can claim to be Churchills
and bash away on each other using "ends justify the means". This is
unavoidable because each individual *does*, de facto, have the privileged
position of deciding what to do. In the larger multiplayer game, the
other players observe your actions and judge whether you're really a
Churchill or a Hitler.

As for your statement that the best intentioned humans "inevitably" lapse,
I think your statement is historically incorrect. Sure, some humans
lapse. Not all of them. The French Revolution failed as the
revolutionaries turned themselves into absolute rulers for the populace's
own good - but the US Revolution did pretty much okay.

> 2. This contradiction may percolate backwards as the relinquishment
> becomes less and less selective, ending in an intolerant and
> absolutist society.

What does tolerance have to do with military retaliation, much less
military defense? Being tolerant does not prevent me from responding to
an invasion of my own borders. You asked whether tolerant countries would
be wiped out by intolerant ones, not whether tolerant countries could or
should invade intolerant ones. My answer is that tolerant countries have
not been successfully invaded by intolerant countries due to the inferior
technology of intolerant countries, and that the relation between
tolerance and technology is more than coincidence.

If Afghanistan ever becomes as superior to the US as the US is currently
superior to Afghanistan, no amount of tolerance or intolerance will save
us; they will simply invade and wipe us out and that will be that.

> This reminds me of those sims that fuse Prisoner's Dilemma and
> cellular automata so you have neighborhoods of cooperators and
> defectors forming and reforming. Defectors can never be completely
> wiped out, because as long as they are surrounded by cooperators
> they'll thrive at their expense. However, they can never prevail,
> because a neighborhood composed only of defectors will starve and
> enterprising cooperators move in and recolonize it.

Tit for Tat wipes out defector colonies easily enough - and without
launching a first strike, either. We can do one better than this because
not only are tolerant democratic countries technologically superior, they
also tend to support each other. They will even support defender
intolerant countries against aggressor intolerant countries, albeit to a
lesser degree. Intolerant countries tend to make worse cooperators than
tolerant countries.

-- -- -- -- --
Eliezer S. Yudkowsky
Research Fellow, Singularity Institute for Artificial Intelligence

This archive was generated by hypermail 2b30 : Sat May 11 2002 - 17:44:15 MDT