Re: Reversible Computation and Experience

From: Lee Corbin (
Date: Tue May 08 2001 - 00:18:28 MDT

Hal Finney wrote (at 06:59 PM 5/5/01 -0700)

>A reversible computer is as likely to take a step backwards as forwards.
>So even if it manages to complete a calculation from A to B, the process
>will be a random walk, moving forward and backwards many times over each
>portion of the path from A to B.

This thought is entirely new to me, yet makes perfect sense.
(After all, truly deterministic and spontaneous action occurs
only when entropy increases, no?) But in none of my readings on
reversible computation have I ever seen this point made. So what
is going on? Perhaps the authors of most works assume that you'd
need just a tiny minimal amount of entropy increase to drive a
calculation forward, and then an equally tiny increase in entropy
to drive it backwards.

If you then keep decreasing such a minimal entropy increase, then
"mistakes" occur, and the process takes a step backwards while its
long term progress is forward, and vice-versa. Random walks with
p arbitrarily close to .5 do eventually move in the right direction.
Perhaps Hal's point isn't as significant as I was beginning to think
(he implies that it's not terribly significant).

>Now if we interpret each step of the computer as instantiating a little
>bubble of conscious experience, then the consciousness will move forward
>and backwards in time just like the computer. You'll think thought 1,
>then thought 2, then you'll think thought 2 backwards, then forwards
>again, then thought 3, then thought 3 backwards, 2 backwards, 1 backwards,
>1 forwards, and so on.


>However of course the conscious mind is not aware of this. The structure
>of the computer is such that it would have no memory of having gone back
>and forth like this. In the end when it thinks back over what happened,
>it will remember thought 1, then 2, then 3, and so on, straight on to
>the end.


>So there is a real disconnect in this case between what we hypothesize
>happening from a functional perspective, and what we have evidence for
>happening based on the reports of the conscious mind. All we truly
>have evidence for is that the computation did exist, did complete.
>We don't know whether the mind really experienced going backwards and
>forwards like that.


>And we especially don't know what form the consciousness took during
>the reverse steps. What does it feel like to have memories erased
>rather than stored? There is no way to ask a person and find out
>(that I can see). Certainly if you ask them after a step is reversed,
>the answer will obviously be the same as if the step never occured.

I would like to insert here that "the feel", itself, is a rather long
computational process. Therefore, so far as I can see, it probably
makes no difference to the conclusion here that you and others (who
are also up to speed on this issue) arrive at. Thus it follows that
if you had to choose between an experience that expended entropy X but
backtracked at numerous points, and the same experience that went
straight to the conclusion, but expended 2X entropy, then you
should be indifferent.

But not me. In 1996, I started a thread called "Repeated Experience",
and I think that I won all the arguments about it---or at least didn't
lose---except for one point Robin Hanson got me on. I was (and still
am) convinced that in almost all thought experiments, repeated
experience is to be reckoned as highly as the "original" (identical)
experience. I mean (and meant) that to be taken quite literally.

So I, on the other hand, might prefer the longer "back-tracking"
experience, since I would gain more additional desireable runtime
from it (unless, of course, it was undesireable experience).

:-) For example, if you and I attend the same party and are
equally ravished, then I might be able to prove to you that
it "lasted longer" for me (even though I don't remember it),
and point out that certain exceedingly pleasurable transitions
that occurred during the party were experienced by me many,
many times, and by you only once.


This archive was generated by hypermail 2b30 : Mon May 28 2001 - 10:00:03 MDT