X-Message-Number: 3326
Date: Fri, 21 Oct 1994 22:47:22 -0400
From: "Keith F. Lynch" <>
Subject: CRYONICS Brain backup proposal

What's the easiest way to back up a living brain?  On Cryonet, Yvan
Bozzonetti suggests this can be done in the near future with x-ray
holography, MRI, or quantum correlations.  Brian Wowk has debunked
these, and suggested that it would require nanotechnology.  Ralph
Merkle's brain analysis paper (ftp://parcftp.xerox.com/pub/merkle/
brainAnalysis.html) doesn't discuss analysis of a *living* brain.

By brain backup I mean a method of recording the state of a living
human brain, such that that person could later be recreated using the
information in the backup, even if the brain were unavailable.

There are several advantages to this over cryonics as a route to extreme
longevity.  Cryonics cannot help if the brain is destroyed by injury or
disease, or if it can't be found (e.g. if you're lost at sea).  Also,
cryonics only preserves the one copy, which is always susceptible
to natural disaster, vandalism, theft, or failure to keep at liquid
nitrogen temperatures.  Multiple backups can be stored at multiple
locations, and don't require liquid nitrogen or other special care.

Here's what I believe is the simplest method of brain backup.  It does
require nanotechnology, but in a fairly simple form, much simpler than
would be required for biostasis or for brain repair.

I'm assuming that a person's personality, memories, aspirations, and
identity (self-circuit, in Robert Ettinger's words) are completely (or
at least, sufficently) defined by their neural connectivity diagram.
This diagram would consist entirely of information on what the odds
are of each neuron firing, given the current state (firing vs. not
firing) of all the adjacent neurons, and given the current levels of
each neurotransmitter chemical in the brain.  If this assumption is
wrong, my backup method won't work.  (Does anyone have any good evidence
for or against it?)

The subject would be injected with about 10^12 nanoprobes.  Each probe
would circulate in the bloodstream until it found itself touching a
neuron.  Then, it would attach itself to the neuron.  It would emit a
chemical signal whenever the neuron fires.  That's all it does.

It doesn't measure or report the neuron's location, or the particular
geometry or chemistry of the neuron or its synapses, as that information
is assumed to be important only to the extent that it influences what
the probe *is* measuring and reporting.

A tiny proportion of these probes would emit a chemical signal which
reports the current levels of various neurotransmitters.  (I'm assuming
that the level of every neurotransmitter is roughly the same across the
whole brain at any given time.  Does anyone know if this is true?)

The chemical signal consists of two binary numbers.  The first is the
ID number of the probe.  Each probe has a unique 40 bit number which
never changes.  The other is a timestamp.  The probe contains a clock
which counts milliseconds since the epoch.  The epoch is the same for
all the probes, and is just before they were injected.  (I'm assuming
a one millisecond time resolution is sufficient, given that neural
firing frequencies are typically on the order of 10 Hz.  If it's not
sufficient, this proposal can be reworked with the higher frequency.)

How would the clocks be kept synchronized?  I think it would be possible
for each probe to pick up the shortwave time stations WWV and CHU,
using the inside of the neuron as an antenna and the outside as a
counterpoise.  The signal wouldn't be strong, but it should be stronger
than anything else at those frequencies.  A probe wouldn't have to be
able to pick up these stations continuously, just often enough that its
clock doesn't drift more than a millisecond.  If any probe's clock does
drift too far, its data won't correlate with anything else, and will be
discarded in the analysis phase, unless it can be gotten back into sync
by trying various fudge factors.

I'm assuming the longest runtime would be one year, or about 3*10^10
milliseconds.  Thus a 35 bit timestamp would be sufficient.  The total
number of bits would be 75.  I'll round it up to 100 for parity and
CRC check bits, and possibly a subject identifier (in case data from
several subjects somehow becomes mixed), and possibly a backup sequence
identifier (in case data from several backups of the same subject
becomes mixed).

This 100 digit number could be encoded in an aliphatic hydrocarbon
chain, by alternating single (saturated) and double (unsaturated) bonds.
Since there can't be two adjacent double bonds, this would require 200
carbons.  On one end would be a radical which the body will ignore.

This chemical is similar to a fatty acid, except that fatty acids always
have a COOH radical.  We wouldn't use that, since the body would then
metabolize our chemical.  Naturally occuring fatty acids would be used
as a source of material for making these chemical signals.  Without an
appropriate radical such as COOH, the body's enzymes have no way to
break into and metabolize the molecule.

Along with these nanoprobes, the subject would also have been injected
with nanomachines which lodge themselves in the kidney, and cause the
information molecules to be excreted in the urine, rather than remaining
in the blood.  The subject collects all his urine for the duration of
the backup, and for a few days after, and the information molecules are
filtered out of the urine and stored.  This waxy solid should be stable
for centuries at room temperature.

If for some reason, this is unworkable, an alternate method would be
to alternate carbons and flourines on an aliphatic hydrocarbon, thus
resulting in a material similar to teflon.  I'm pretty sure this would
be non-toxic.  The disadvantage of this is that you'd need to keep these
probes supplied with flourine.  Perhaps this could be done with regular
injections of more of the same kinds of teflon-like molecules.

Once the backup is over, some more information will be added to the
mix, in the form of information chemicals with a zero timestamp.
These will have the current date and time, the subject's full name,
the name and location of the lab doing the backup, the time of any
events during the backup that weren't representative of the subject's
usual consciousness (e.g. concussions, seizures, fainting, anesthesia,
etc.), and a complete description of the formatting.  This information
would be encoded in English in straight ASCII.  Thus, even if the
external label and all documentation were to be lost, a sufficiently
advanced civilization should be able to figure it all out, given that
they know English and ASCII.

There is considerable redundancy in the system.  There are about 10^11
neurons in the brain, thus the average neuron will have ten probes.
And each probe releases ten identical molecules per neural firing.  Thus
there should be about a hundred molecules conveying the fact of that
that neuron fired at that time.  Only one of those hundred need be
stored in order to capture all the available information.  Also, if
the proportion of events for which *no* molecules were stored (if any)
can be deduced, as I suspect it can, the neural firing probabilities
can simply be fudged by the appropriate ratio.  The system degrades
gracefully with loss of information molecules, becoming gradually
noisier rather than failing at once.

Because of this redundancy, the stored molecules can be divided among
several containers which are stored in different places, for protection
against fire, flood, theft, tornadoes, earthquakes, misfiling, and the
other ravages of time.

It's quite likely that, although the information molecules can't be
metabolized by the body, some proportion of them will end up being
stored in the body's fat cells.  This is a feature, not a bug.  The
body's fat stores will thus contain another backup, which should
be readable even if the body is badly decayed, or possibly even
skeletonized.

Please note:  There's no need for anyone to be able to analyze the data,
or to be able to do anything with it except store it, at the time of the
backup.  It's sufficient that it's known that the information is there,
and that a technology able to recreate a living brain from it is likely
to eventually be developed.  Neither is it necessary to determine the
length of the backup phase ahead of time.  I'm assuming we collect data
for a year.  Perhaps at restore time, all but one minute's worth will be
discarded as unnecessary.

So long as the recreated brain (whether it's made of flesh, silicon, or
something else, and whether there is some particular piece of matter
which acts as each neuron, or whether it's emulated in software) has
the same statistics as the original, its behavior should be the same
as the original subject's.  And given that its behavior is the same,
I believe its internal subjective experience would be the same.

How much material would be in these information molecules?  If each
probe is designed to release ten identical molecules for each neuron
firing, and if the average neuron fires ten times per second (does
anyone know how realistic an estimate this is?), and there are 10^12
probes, that's 10^14 total molecules per second, or 3*10^21 molecules
per year, or 6*10^23 carbons per year, or 8*10^24 AMU per year (assuming
all bonds are saturated), or 14 grams per year.  The density is about
one, so this should be about 14 cc's in volume. The naturally occuring
fatty acids used for the purpose would total about 130 dietary Calories
over the year -- about equal to the amount gained from eating one
cookie, or lost in walking one mile (1.6 km).

It's not clear to me whether, if a memory isn't recalled for a period,
that means that certain related neurons won't fire for that period.
If this is the case, that means that anything the subject doesn't think
of during the backup won't be picked up, and the restored subject will
have no recollection of them later.  That possibility is the main reason
I suggested a backup time of one year.  I'd much rather have *all* my
memories, but I can live without the ones I don't think of for a year,
if the only alternative is having no backup.  If I were making a backup,
I'd be sure to go over my old photo albums, my old usenet files, visit
places I used to live, and reminisce with old friends.

Another possibility would be to deliberately induce seizures during the
backup.  Especially if there's a way to do so that guarantees that every
neuron fires lots of times.  I believe electo-convulsive therapy does
this.  However, it's believed it may cause subtle brain damage, so it's
best avoided if possible.  Anyhow, it's not clear that a backup made
during a seizure would register anything useful.

There are other issues.  For instance, where do the sensory and motor
nerves hook up?  I'm hoping this would be clear from internal context,
together with a more advanced knowledge of how brains are typically
wired.  It should, for instance, be simple enough to not only locate the
visual cortex, but to reproduce any image that was seen by the subject
during the backup.  Analogously with the other senses.

How would a restore be done?  There are lots of very different
possiblities.  Here's one.  Keep in mind that this will be many decades
after the backup was made, and will involve much higher technology.

All of the information molecules would be read into a very large
computer memory.  Any whose data is corrupt, or which belong to a
different backup or a differnet subject would be discarded.  The
quality and quantity of information available will be measured as
a function of time during the backup.  A particular target period
during the backup will be selected, when the quality and quantity
were both excellent.  All information outside that period will be
discarded.  (Or rather, set aside from the restore -- no information
will ever really be deliberately permanently discarded.)

The data will be temporarily segregated by probe number.  For each
pair of probes -- all 5*10^23 ways to select two of 10^12 -- the
all-over correlation coefficient will be looked at.  If it's very
high and positive, they will be considered to be on the same neuron,
and will subsequently always be lumped together.  If it's high, positive
or negative, they will be considered to be on adjacent neurons.  If it's
close to zero, they will be considered to be distant from each other.
The relative timing, and the dependancy on neurotransmitters will be
looked at.

In any case, the wiring diagram can be made to look like this:
               _
             _|_
           _|_|_
         _|_|_|_
       _|_|_|_|_
     _|_|_|_|_|_
   _|_|_|_|_|_|_
 _|_|_|_|_|_|_|_
| | | | | | | |

Shown are eight neurons.  All of them run vertically upwards, then
make a sudden right angle turn, then run horizontally to the right.
Wherever two cross, there may be a synapse in one or both directions,
with some arbitrary delay built in, and some arbitrary strength, and
some arbitrary dependance on neurotransmitters.  All we have to do is
figure out what's at each intersection, then build the thing, then
hook it up to sensory and motor nerves (or their equivalent) in a
healthy body (or the equivalent) in the appropriate way.  Whether
it is actually built in this half-flyscreen geometry is a mere
implementation detail.

I eagerly await comments and criticism.

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=3326