Time and Alternity by Computer

 Hans Moravec
 The Robotics Institute
 Carnegie-Mellon University
 Pittsburgh, PA  15213
 (412) 268-3829

Copyright 1986 by Hans P. Moravec


	Time travel is a familiar concept in science fiction, and
often brings with it the concept of alternate worlds.  The mechanism
of time travel is usually some extrapolation of modern physics -
certainly fertile ground, with special relativity allowing
communication to the past if faster than light particles could be
found, general relativity allowing spacetime to be warped and twisted
into temporal knots, and quantum mechanics seemingly founded on the
temporary superposition of alternate worlds. Yet, if tachyons really
don't exist, if Tipler vortex time machines are in principle
impossible to build and black holes lead only to oblivion, if the
alternate worlds in quantum mechanics are a mere mathematical
artefact, or are truly inacessible, are we stuck, helplessly drifting
down the one way river of time?  Is there some way out, other than
exotic physics?  Here's how to do it with a philosophical leap and a
lot of conventional future technology.

\sect{What Am I?}

	Let's suppose we have some method of reading out the contents
of a human mind into a computer controlling a robotic body, in such a
way that the machine behaves like the person it absorbed.  Science
fiction readers have encountered this concept many times, but often
the stories and articles have been humorous in tone, masking, I think,
a discomfort with the idea felt even by the authors.  This feeling is
sometimes articulated in statements like: ``Regardless of how the
copying is done the end result will be a new person. If it is I who am
being copied, the copy, though it may think of itself as me, is simply
a self-deluded imposter. If the copying process destroys the original
then I have been killed.  That the copy may then have a good time
using my name and my skills is no comfort to my mortal remains.''

	This point of view, which I will call the {\it Body Identity}
position, makes a mockery of many of the supposed advantages of being
``mind transferred'' to a new body.  I believe the objection can and
should be overcome by intellectual acceptance of an alternate position
I will name {\it Pattern Identity}.  Body identity assumes that a
person is defined by the stuff of which a human body is made. Only by
maintaining continuity of body stuff can we preserve an individual
person.  Pattern identity, on the other hand, defines the essence of a
person, say myself, as the {\it pattern} and the {\it process} going
on in my head and body, not the machinery supporting that process.  If
the process is preserved, I am preserved.  The rest is mere jelly.

\sect{Matter Transmitters}

	Matter transmitters have appeared often in the science fiction
literature, at least since the invention of facsimile machines in the
late 1800s. I raise the idea here only as a thought experiment, to
simplify some of the issues in the mind transfer proposal of.  A
facsimile transmitter scans a photograph line by line with a light
sensitive photocell, and produces an electric current that varies with
the brightness of the scanned point in the picture. The varying
electric current is transmitted over wires to a remote location where
it controls the brightness of a light bulb in a facsimile receiver.
The receiver scans the bulb over photosensitive paper in the same
pattern as the transmitter.  When this paper is developed, a duplicate
of the original photograph is obtained.  This device was a boon to
newspapers, who were able to get illustrations from remote parts of
the country almost instantly, rather than after a period of days by
train.

	If pictures, why not solid objects?  A {\it matter
transmitter} might scan an object and identify, then knock out, its
atoms or molecules one at a time.  The identity of the atoms would be
transmitted to a receiver where a duplicate of the original object
would be assembled in the same order from a local supply of atoms. The
technical problems were mind boggling, and well beyond anything
foreseeable, but the principle was simple to grasp.  If solid objects,
why not a person?  Just stick him in the transmitter, turn on the
scan, and greet him when he walks from the receiver.  But is it really
the same person?  If the system works well, the duplicate will be
indistinguishable from the original in any substantial way.  Yet,
suppose you fail to turn on the receiver during the transmission
process. The transmitter will scan and disassemble the victim, and
send an unheard message to the inoperative receiver. The original
person will be dead. Doesn't, in fact, the process kill the original
person whether or not there is an active receiver?  Isn't the
duplicate just that, merely a clever imposter?  Or suppose two
receivers respond to the message from one transmitter.  Which, if
either, of the two duplicates is the real original?


\sect{Pattern Identity}

	The body identity position is clear. A matter transmitter is
an execution device. You might as well save your money and use a gas
chamber, and not be taken in by the phony double gimmick.  Pattern
identity gives a different perspective. Suppose I step into the
transmission chamber. The transmitter scans and disassembles my
jelly-like body, but my pattern (me!)  moves continuously from the
dissolving jelly, through the transmitting beam, and ends up in other
jelly at the destination.  At no instant was it (I) ever
destroyed. The biggest confusion comes from the question of
duplicates.  It is rooted in all our past experience that one person
corresponds to one body.  In the light of the possibility of matter
and mind storage and transmission this simple, natural, and obvious
identification becomes confusing and misleading.  Suppose the matter
transmitter is connected to two receivers instead of one.  After the
transfer there will be a copy of you in each one.  Surely at least one
of them is a mere copy - they can't both be you, right?  {\it Wrong!}

	Consider the message ``I am not jelly''. As I type it it goes
from my brain, into the keyboard of my computer, through myriads of
electronic circuits and over great amounts of wire, and after
countless adventures shows up in bunches of books like the one you're
holding.  How many messages were there?  I claim it is most useful to
think there is only one, despite its massive replication.  If I repeat
it here: ``I am not jelly'', there is still only one message. Only if
I change it in a significant manner: ``I am not peanut butter'' do we
have a second message. And the message is not destroyed until the last
written version is lost, and until it fades sufficiently in
everybody's memory to be unreconstructable. The message is the
information conveyed, not the particular encoding.

	The ``pattern and process'' that I claim is the real me has
the same properties as the message above.  Making a momentary copy of
my state, whether on tape or in another functional body, doesn't make
two persons.  There is a complication because of the ``process''
aspect; as soon as an instance of a ``person message'' evolves for a
while it becomes a different person. If two of them are active, they
will diverge, and become two different people by my definition. Just
how far this differentiation must proceed before you grant them unique
identities is about as problematical as the question ``when does a
fetus becomes a person?'' But if you wait zero time, then you don't
have a new person.  If, in the dual receiver version of the matter
transmitter, you allow the two copies to be made and kill one (either
one) instantly on reception, the transmitted person still exists in
the other copy. All the things that person might have done, and all
the thoughts she might have thought, are still possible.  If, on the
other hand, you allow both copies to live their separate lives for a
year, and then kill one, you are the murderer of a unique human being.
{\it But}, if you wait only a short while, they won't differ by much,
and destruction of one won't cause too much total loss. This rationale
might, for instance, be a comfort in danger if you knew that a tape
backup copy of you had been made recently.  Because of the divergence
the tape contains not you as you are now, but you as you were: a
slightly different person. Still, most of you would be saved should
you have a fatal accident, and the loss would be nowhere near as great
as without the backup.

	Intellectual acceptance that a secure and recent backup of you
exists does not necessarily protect you from an instinctive
self-preservation overreaction if faced with imminent death. This is
an evolutionary hangover from your one-copy past.  It is no more a
reflection of reality than fear of flying is an appropriate response
to present airline accident rates. Inappropriate intuitions are to be
expected when the rules of life are suddenly reversed from historical
absolutes.

\sect{Soul in Abstraction}

	Although we've reasoned from strictly reductionistic
assumptions about the nature of thought and self, the pattern identity
position has clear dualistic implications. Though mind is entirely the
consequence of interacting matter, the ability to copy it from one
machine or storage medium to another gives it an independence and an
identity apart from its machinery.  The dualism is especially apparent
if we consider some of the variations of encoding possible.

	Some supercomputer designs call for myriads of individual
computers interconnected by a network that allows free flow of
information among them.  An operating system for this arrangement
might allow individual processes to migrate from one processor to
another in mid computation, in a kind of juggling act that permits
more processes than there are processors.  If a human mind is
installed in a future machine of this variety, functions originally
performed by particular cell assemblies might be encoded in individual
processes. The juggling action would ensure that operations occurring
in fixed areas in the original brain would move rapidly from place to
place within the machine.  If the computer is running other programs
besides the mind simulation, then the simulation might find itself
shuffled into entirely different sets of processors from moment to
moment.  The thinking process would be uninterrupted, even as its
location and physical machinery changed continuously, because the
immaterial pattern would keep its continuity.

	A process that is described as a long sequence of steps can
sometimes be transformed mathematically into one that arrives a the
same conclusion in far fewer operations.  As a young boy the famous
mathematician Friedrich Gauss was a school smartaleck. As a diversion
a teacher once set him the problem of adding up the all the numbers
between $1$ and $100$.  He returned with the correct answer in less
than a minute. He had noticed that the hundred numbers could be
grouped onto fifty pairs, $1+100$, $2+99$, $3+98$, $4+97$ and so on,
each pair adding up to $101$. Fifty times $101$ is $5,050$, the
answer, found without a lot of tedious addition.  Similar speedups are
possible in complex processes.  So called {\it optimizing compilers}
have repertoires of accelerating transformations, some very radical,
to streamline programs they translate.  The key may be a total
reorganization in the order of the computation and the representation
of the data.  A very powerful class of transformations takes an array
of values and combines them in different ways to produce another
array. Each final value reflects all the original values, and each
original value affects all the results. An operation on a single
transformed quantity can substitute for a whole host of operations on
the original array, and enormous efficiencies are possible.  Analogous
transformations in time also work: a sequence of operations is changed
into an equivalent one where each new step does a tiny fraction of the
work of every one of the original steps.  The localized is diffused,
and the diffuse is localized.  A program can quickly be altered beyond
recognition by a few mathematical rewrites of this power. Run on a
multiprocessor, single events in the original formulation may appear
only as correlations between events in remote machines at remote times
in the transform. Certain operations that don't matter in the long run
may be skipped altogether. Yet the program is fundamentally
unchanged. You know what's coming.  If we thus transform a program
that simulates a person, the person remains intact. Soul is in the
mathematical equivalence, not in any particular detail of the
process. It has a very etherial character.

\sect{The Message is the Medium}

	If a mind can survive repeated radical restructurings,
infusion into and out of different types of hardware and storage
media, and is ultimately a mathematical abstraction, does it require
hardware at all?  Suppose the message describing a person is written
in some static medium, like a book.  A superintelligent being, or just
a big computer, reading and understanding the message might be able to
reason out the future evolution of the encoded person, not only under
a particular set of experiences but also under various alternative
circumstances.  Existence in the thoughts of a beholder is no more
abstract than as a transformed person-program described in the
previous section, but it does introduce an interesting new twist.

	The superintelligent being has no obligation to accurately
model every single detail of the beheld, and may well choose to skip
the boring parts, to jump to conclusions that are obvious to it, and
to lump together different alternatives it does not choose to
distinguish. This looseness in the simulation can also allow some time
reversed action - our superintelligent being may choose a conclusion
then reason backwards, deciding what must have preceded it. Authors of
fiction often take such liberties with their characters.  The same
parsimony of thought applies to the parts of the environment of the
contemplated person that are themselves being contemplated.  Applied a
certain way, this parsimony will affect the evolution of the simulated
person and his environment, and may thus be noticeable to him. Note
that the subjective feelings of the simulated person are a part of the
simulation, and with them the contemplated person feels as real in
this implementation as in any other.

	It happens that quantum mechanics describes a world where
unobserved events happen in all possible ways (another way of saying
no decision is made as to which possibility occurs), and the
superposition of all these possibilities itself has observable
effects. The connection of this observation with those of the previous
paragraph leads us into murky philosophical waters.  To get even
muddier, seriously consider the title of this section. If the
subjective feelings of a person are part of the person-message, and if
the evolution of the message is implicit in the message itself, then
aren't the future experiences of the person implicit in the message?
And wouldn't this mere mathematical existence feel the same to the
person encoded as a more substantial simulation? I don't think this is
mere sophistry, but I'm not prepared to take it any further for now.


\sect{Immortality and Impermanence}

	Wading back into the shallows, let's examine a certain dilemma
of existence, presently overshadowed by the issue of personal death,
that will be paramount when practical immortality is achieved.  It's
this: in the long run survival requires change in directions not of
your own choosing.  Standards escalate with the growth of the
inevitable competitors and predators for each niche.  In a kind of
cosmic Olympic games the universe molds its occupants towards its own
distant and mysterious specifications.  An immortal cannot hope to
survive unchanged, only to maintain a limited continuity over the
short run. Personal death differs from this inevitability only in its
relative abruptness. Viewed on a larger scale we are already immortal,
as we have been since the dawn of life.  Our genes and our culture
pass continuously from one generation to the next, subject only to
incremental alterations to meet the continuous demand for new world
records in the cosmic games.

	In the very long run the ancestral individual is always doomed
as its heritage is nibbled away to meet short term demands. It slowly
mutates into other forms that could have been reached from a range of
starting points; the ultimate in convergent evolution.  It's by this
reasoning that I concle that it makes no ultimate difference whether
our machines carry forward our heritage on their own, or in
partnership with direct transcriptions of ourselves.  Assuming long
term survival either way, the end results should be indistinguishable,
shaped by the universe and not by ourselves.  Since change is
inevitable, I think we should embrace rather than retard it. By so
doing we improve our day to day survival odds, discover interesting
surprises sooner, and are more prepared to face any competition.  The
cost is faster erosion of our present constitution.  All development
can be interpreted as incremental death and new birth, but some of the
fast lane options make this especially obvious, for instance the
possibility of dropping parts of one's memory and personality in favor
of another's.  Fully exploited, this process results in transient
individuals constituted from a communal pool of personality traits.
Sexual populations are effective in part because they create new
genetic individuals in very much this way. As with sexual
reproduction, the memory pool requires dissolution as well as creation
to be effective.  So personal death is not banished, but it does lose
its poignancy because death by submergence into the memory pool is
reversible in the short run.

\sect{Back to Time Travel}

	In the continuing struggle for survival, we've already
acquired considerable control of time.  Memory - genetic, reflecting
our evolution, nervous, storing our experiences, or artificial,
recording events and thoughts, gives us some mastery the past. What of
the future?  A hallmark of intelligence is the ability to choose, from
many possibilities, those actions that accomplish certain ends. In
advanced robots, as in large brained animals, there is the possibility
of deliberation, in which alternatives are imagined and their outcomes
weighed, prior to the action.  However imperfectly, such planning
involves a prediction of the future, or, more precisely, of possible
futures.  The central goal of scientific inquiry has been the
refinement of this skill. A good theory {\it predicts}.  Theories and
their predictions come to life in simulations, particularly on
powerful computers. Such simulations have been especially accurate in
the programs that predict the course of the planets and of
spacecraft. More dramatically, if less accurately, modern weather
programs simulate the action of the atmosphere over the entire globe.
Increasingly powerful computation makes possible more accurate and
longer range predictions. In a real sense powerful simulators are time
machines, giving peeks into possible futures, and thus the power to
choose among them.

	The laws of physics are quite symmetric in time, and
simulations can usually be run in reverse as well as forward, and used
to ``predict'' the past, perhaps guided by old measurements or
archeological data.  As with future predictions, any uncertainty in
the initial measurements, or in the rule that evolves the initial
state, will allow for a variety of possible outcomes. If the
simulation is detailed enough, and is given all available information,
then all of its "predictions" are valid - {\it any of the possible
pasts may have led to the present situation}.  This is a strange idea
if you are accustomed to looking at the world in a strictly
deterministic, Newtonian, way.  Interestingly it closely resembles the
uncertain world described by Quantum Mechanics, and perhaps hints at a
mechanism underlying our world.  Now, imagine an immense simulator
that is able to model the whole surface of the earth on an atomic
scale, and that can run time forward and back, and produce different
plausible outcomes by making different random choices at key points in
its calculation. Because of the great detail this simulator models
living things, including humans, in their full complexity.  By the
arguments above such simulated people would be as real as you or I,
though imprisoned in the simulator.

	We could join them by linking up with the simulation through a
telepresence interface, that connects a ``puppet'' deep inside the
simulation with a physical ``helmet'' and "gloves" outside, allowing
us to experience the puppet's sensory environment, and to naturally
control its actions.  More radically, we could ``download'' our minds
directly into a body in the simulation, and ``upload'' back into the
real world when our mission is accomplished.  Alternatively, we could
bring people out of the simulation by reversing the process, linking
their minds to an outside robot body, or uploading them directly into
it.  In all cases we would have the opportunity to recreate the past,
and to some extent the future, and interact with it in a real and
direct way.  Realistically simulating the future is more difficult
because archeology cannot help, and because an advancing culture will
produce fundamental new knowledge, not found in the model, by research
into new physical arenas or exploration of new geography.  The same
techniques, of course, allow visits to entirely novel situations and
universes.

\end{document}