From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Mon Dec 17 2001 - 09:55:17 MST
Samantha Atkins wrote:
>
> There is "no reason" true enough except that it might simply be
> how the post-SI Mind decides to do things here for now. If I on
> this side of the Singularity can come up with plausible reasons
> for a choice of limited involvement then I see no way to dismiss
> the possibility out of hand.
I can also come up with "plausible" arguments for why a Mind would tamper
just so much and no further, although usually such arguments tend to lend
themselves better to a Mind running a simulation with minimalist
interference, rather than a Mind reaching back into the past.
The question is not plausibility or implausibility. The question is the
Bayesian Probability Theorem. My theory - the theory that says that
you're just imagining it - uniquely predicts that a certain level of
intereference will be alleged. If you consider the prior probabilities
for "permitted level of interference", these priors have no favored
tendency to work out at a personally noticeable but not experimentally
verifiable level of interference. Sure, it *could* end up that way, but
the prior probability of it ending up that way, given a Mind, is something
that I would estimate at 1% or less because of the exotic initial
conditions required. This is essentially the same argument against
noticeable-but-not-verifiable divine intervention - you have to postulate
a very unusual psychology for God, where most compact descriptions of
God's psychology will lead either to total uncaring or to a massive
intervention to reduce suffering. Now it is of course possible to
postulate a set of initial conditions that produce a divine psychology or
Mind psychology that engages in quiet hints for moral uplift but not
massive interventions. The problem is that this is extra complexity, a
considerable amount of extra complexity, which under Occam's Razor reduces
the likelihood of the hypothesis.
If someone is honestly considering the Bayesian priors for Mind
psychology, just trying out the different *a priori* plausible options and
extrapolating them without trying to generate the specific conclusion of
noticeable-but-not-verifiable interference, then I don't think you'll ever
run across a "natural" conclusion (generated without expecting the
conclusion in advance) of "Aha! This scenario predicts personally
noticeable but not scientifically verifiable intervention, which
intervention is morally uplifting to those who observe it!" On the other
hand, the "It's all a meme" hypothesis does predict, as a unique outcome,
very naturally and without the need for any exotic initial conditions,
that just this type of intervention will be asserted by a large group of
human beings.
Of course, it's hard to make prior predictions without cheating, because
we aren't AIs and can't close off from ourselves our knowledge of how
things actually turned out. Even so, I would expect that the clean
prediction from the prior of "It's all a meme" would assign at least a 30%
probability, probably more like 80%, to the observation of certain people
making the claim "I know that God / a Mind is engaging in personally
noticeable intervention which is not blatant enough to be immediately
falsifiable." The clean prediction from the prior of "God / a Mind
exists", to the Mind having a particular psychology, to that Mind engaging
in personally noticable intervention which is not great enough to be
socially verifiable, to people noticing and (correctly) saying "I know a
Mind exists which is engaging in noticeable-but-not-verifiable
intervention", seems to me to be a maximum of 1% and probably more like
0.1% or 0.01%.
And then you just run the Bayesian Probability Theorem and ask what the
chances are for the observation "People claiming
noticeable-but-not-verifiable intervention" being produced by people with
actual experience as opposed to being produced by people with imagined
experience. It's not that an interventionist Mind is unlikely. It's that
the memetic hypothesis predicts a claim of interference exactly balanced
between noticeability and verifiability, and the actual Mind intervention
hypothesis does not.
And that's just for the general, vague version of spirituality. If you're
talking about a textual religion such as the Judeo-Christian-Islamic
tradition, you have to explain why the Mind engaged in massive blatant
interventions a few thousand years back, but then suddenly kicked the
habit as soon as civilization started coughing up scientists and
reporters. In this case, there's basically no way you can make it
plausible, no matter what kind of Mind psychology you hypothesize - the
inconsistent description of past and present events forced by a text-based
religion has the unique signature of a human-invented false meme, and you
can't come up with a significant prior probability of that description
being correct, no matter what initial conditions you postulate.
> Spirituality is quite falsifiable. Try it on. If it doesn't
> improve your life and open up your options in a way that is
> worth something to you then discard it. Don't expect to weight
> something that rich and intricately part of life with
> rationality and logic alone though. Not if you really want to
> know.
"Rationality and logic alone?" 'Alone' is a poor word to use for
something that powerful. If my mind really is incomplete, it is perhaps
plausible that my mind would not be able to *fully* understand the power
of spirituality as it stands. But to have my mind not notice it at all?
I don't buy it.
And I certainly will not "try on" a meme that appears to contain circular,
self-reinforcing logic, as for example if the meme asserts that its
wonderful effects are only visible from inside the meme. Who do you take
me for, Inoshiro?
> Why should the effects be blatant? Perhaps part of whatever
> this sim is set up for, or especially if it is the Mind's own
> past, will be harmed by anything too blatant. This especially
> works well with the time-travel model. Improve slightly but not
> so drastically as to jeopardize the timeline. Sufficent to do X
> and X being advisable within the constraints of the situation
> are quite different things.
Improve slightly but not so drastically as to jeopardize the timeline, I
could buy. "Improve just enough that a very large group of individual
spiritualists correctly notice it and are in basic correct agreement about
what it represents, but not improve so much that scientists can prove it,"
I don't buy. I don't think you can manipulate the physical conditions of
the technology, or the psychological conditions of the Mind, in such a way
that this result is a reasonable outcome. But the existence of people
incorrectly asserting this result is a very obvious and direct outcome of
human memetics.
> I would also point out that if you wish to move much of the
> population of the world in a more positive direction the
> language of spirituality is quite important.
Yeah, well, I don't just wish to move the population of the world, I want
to do it ethically. I can use a certain subset of the language of
spirituality because I regard that subset as an extension of rationality
by other means. I can't ethically use mystical language because I don't
believe in it myself. Even if mystical language was the easy and direct
way to convince much of the population of the world, I still couldn't use
it.
> And it is
> spirituality that gives rise to most living ethical systems
> today.
So? Most living ethical systems today are painfully incomplete.
> We have yet to acheive a strong ethical system using
> reason alone. Creating a world that is viable and forward
> moving requires ethical, political, psychological and
> sociological aspects as well as technological/scientific ones.
Ethical, political, psychological, and sociological aspects can be
understood using rational means. The choice of ethics may need to rely on
the human emotional set as part of the baseline (or not - currently I'm
figuring it both ways) but that doesn't compromise rationality.
> I don't believe the necessary work can be done with
> reason/rationality as usually understood alone. If you know how
> it can be I would very much like to hear it.
Use reason and rationality and (ethical) appeal to shared emotions.
> It is your construction that spritual speaking is "shorthand".
> There are spiritual images, concepts, experiences, teachings
> that simply do not translate into the language of
> rationality/logic/reason no matter how long you speak and how
> patiently intellectual you assume your audience is.
Can you give me an example?
> I believe in a transcendent Mind not limited to what we
> currently understand through science and that that Mind is
> immanent as well and is interested and under certain conditions,
> at least somewhat accessible. I have little choice but to
> believe this as I have experienced it firsthand. Yes, I could
> come up with rationalizations to "debunk" each such experience
> and the ones that I could possibly have in the future. But at
> some point the effort to "debunk" them becomes less believable
> and more contrived than accepting such as also valid. That is
> the point I have reached and passed in my own life.
I don't mean to sound insulting, but you do realize that unless you cite
the specific experiences (and I will understand if you are reluctant to do
so), I have to assume as a working hypothesis that this statement,
although originating from Samantha Atkins, is of much the same origin as
similar statements made by others for personal experience of astrology.
Of course I am far more inclined to grant you the benefit of the doubt
given your demonstrated rationality in other areas - but not, I'm afraid,
*that* much benefit of the doubt, not unless I know specifically what your
reasoning is.
-- -- -- -- --
Eliezer S. Yudkowsky http://singinst.org/
Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 08:12:37 MST