From: Anders Sandberg (asa@nada.kth.se)
Date: Sat Jul 19 2003 - 01:01:51 MDT
On Fri, Jul 18, 2003 at 10:02:56PM -0700, Lee Corbin wrote:
> Anders writes
> > > Sorry, but you are sounding like one of the usual apologists
> > > for pain and suffering here. There really isn't anything that
> > > can be computed with pain that cannot be computed without it.
> >
> > Yes, it does sound like the standard apology.
> > But evolution and similar time-asymmetric complexity increases cannot
> > occur unless 1) some information is erased to produce selection or 2)
> > there is always more resources to encode information in.
>
> It's not clear to me that 1) always has to be true; but you
> do a good job of delineating the possibilities here:
To get evolution you need replication, variation and selection. Unless
selection is all about number of offspring (case 2), some replicators
have to be removed. (Here replicators do not need to be beings, they
could just as well be memes or engineering designs). I think you can do
the variation part without losing information (just attach a changelog),
and if you have free memory space you can replicate with no loss (but
this only work if the memory is known to be zeroed; overwriting stuff
implies erasure and energy dissipation).
> > Second, pain might perhaps be replaced by non-pleasure. But in many
> > learning rules there is an inherent difference between non-pleasure and
> > aversive experience, and it could be that the most efficient learning
> > rules belong to this category. Then the issue becomes a choice between
> > pain or lesser learning abilities.
>
> It's been pointed out many times, I am sure, that the kinds of
> minds we would like to become would still encounter disappointment
> and frustration. Fine.
>
> The aversive experience I have in mind is the hideous burden that
> intelligent creatures continue to suffer under because they have
> not yet gained control of their internal workings. An extremely
> stupid beast needs to be reminded not to kick certain kinds of
> objects, for example, and extremely and long-lasting pain was
> the most efficient teacher. I propose to do away with all stupid
> creatures, and, in fact, probably most matter that is not itself
> intelligent and capable of getting along quite nicely without pain.
>
> I cannot imagine the case that you are discussing: namely, an
> intelligent entity whose learning is advanced by the kind of
> pain experienced by lower animals and, sadly still, humans.
Human thinking and behaving is right now extremely colored by that we
seem to be based on temporal difference reinforcement learning. We get
a bigger reward from when things turn out better than expected than when
things turn out just right. This makes us look for new
possibilities more than we would have done if we had another form of
reinforcement learning, even if that is just as good from a pure
learning standpoint. This and many other quirks of the learning rule
makes us who we are. Changing the rule might be possible, resulting in
subtle changes of the posthuman mindset.
But it is still not clear to me that an antisymmetric utility function
(X units of bad is felt as -1*(feeling X units of good) so we can
replace bad experiences with absences of good) will be a useful option.
Risks against survival have to change behavior far more strongly than
absence of a strong reward signal. Or is this prejudice from a world
where survival risks are common? Maybe it is not necessary in a world
where it is really hard to hurt oneself. But as _The Golden Age_ points
out, a world with mental technology is filled with subtle dangers we
really need to find ways of avoiding.
> David Pearce's message is so right and so important, that all
> quibbles need to be swept aside www.hedweb.com
Hmm, a rather dangerous sentiment. :-)
-- ----------------------------------------------------------------------- Anders Sandberg Towards Ascension! asa@nada.kth.se http://www.nada.kth.se/~asa/ GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y
This archive was generated by hypermail 2.1.5 : Sat Jul 19 2003 - 01:07:28 MDT