[MURG] new Road Map draft

[king-yin yan]

Hi Jess and others

Newest version 3.3:
http://www.geocities.com/softuploading/RoadMap2.png
http://www.geocities.com/softuploading/RoadMap2.xml
http://www.geocities.com/softuploading/RoadMap-Explanations.txt

>-The visual representation of information in this diagram seems disorganized to
> me, but maybe it's just me.

It's not you, roadmap 3.1 was bad for many reasons. In v3.3 I've
made things much simpler and tried not to speculate on things I
have insufficient knowledge of.

>-This is listed as v3.1 of the roadmap.  What was v3.0?

V3.0 contained a few mistakes so I deleted it from the site. I prefer
keeping the current version only, is that OK?

>-What do you mean by "chronic implants?"  As opposed to what?

There're some BCI implants that don't last very long, for various reasons
such as chemical reactions at the electrode surface, or immuno-rejection,
etc. Chronic means the opposite of that, meaning long-term stability of
the BCI implant.

>-Why could more invasive and damaging techniques be considered if replacement
> occurs over the span of a year?

How long the complete replacement will take place will depend on
the status of several technologies, as shown on the roadmap. The
"chronic" issue is not so central in the current context, so I
dropped it in v3.3.

>-What are your concerns about "hard BCI?"  The only issue I can think of is
> that the electrodes might wander over time, but I'm not sure about this. See
>   http://www.sciforums.com/t8155/s8cee3ed6aab510939aad1cf724cf6fd2/thread.html

I think probing density is probably the most important concern. This
determines how much memory extraction can be achieved. Also it
determines how much interaction between brain and artificial brain
can occur. I think probing density may be improved, since current
technology not yet matches that of neurons in the brain, in terms
of density. There're other issues such as bio-inertness and as you
say, undesirable displacement, etc. A BCI expert will know much
more about this... There's a body of knowledge existing in the
industry as well as in academia.

>-Why does memory extraction necessitate bi-directional communication?  Are you
> thinking of trying to stimulate the brain to evoke all memories, and then read
> them out?  But how would you know whether or not you've gotten all of them?
> and if the materialist philosophy is correct, then memories must somehow be
> encoded in the brain's physical makeup, so as long as you duplicate the brain's
> physical makeup you will automatically have the memories.  *Interpreting* the
> brain's physical makeup - trying to figure out WHICH information is important -
> leads to complications, and should be avoided whereever possible; so we don't
> want to "read out" the memories from the brain, we want to read out the physical
> structure from the brain.  We want to mindlessly (no pun intended) record all
> possible information about the lower level of the physical, not go into the
> higher level of interpretation.

We don't have a thousand years to read out all that info. If we have an
understanding of how relevant information is stored in the brain then we
can achieve MU much sooner.

My early theory of memory extraction assumes that biological neurons
can be sufficiently approximated by the Sigma-Pi (or "sum of products")
model, which is basically a power series in (n) inputs that is capable of
representing any smooth function of those inputs, and thus, is more
general than the Perceptron and Integrate-and-Fire models. Then, memory
extraction is essentially a nonlinear regression problem (as noted by
Randall). *Note* that this is an assumption and currently there are a lot
of controversies about what exactly is single-neuron computation.
Because of this, I think it'd be much wiser to elucidate first the single-
-neuron mechanisms before developing any theory of memory extraction.

>-Is the "Moravec procedure" really required for gradual replacement?

By "Moravec robot" (sorry I conned many terms) I mean a neurosurgical
robot that implants electrodes to the brain. I think this is essentially the
same as the so-called Moravec procedure, except that my version doesn't
require nanotech. I think using a robot is actually MUCH *neater* than
doing the replacement in bigger chunks. Also a lot of surgeries will wear
out any human neurosurgeon.

>-What is "cell-type identification?"  If it means identifying which type of cell a
> given neuron is, why is it separate from the "Moravec Robot?"

Yes, it's part of the job of the robot. I make it separate so the robot's
function is solely to implant electrodes. Right now I don't have any idea
of how to do the cell-type identification.

>-You've identified a whole-brain connectivity map as essential to uploading, but
> identified vitrification/slicing/scanning as essential to that map.  I agree that
> a whole-brain map is necessary, but not, in my inexperienced view, in the same
> level of detail as a Slice N Scan (TM) would provide.  I would argue that all we
> need to know about the connectivity of the brain is what is common to all humans,
> which  I think (PLEASE correct if I'm wrong) we already know; I don't think the
> effort of the SnS is truly necessary.

You're wrong I'm afraid. We need the most detailed map (common to
all (or most) humans), but not more detailed than that. The Slice and
Scan is needed, I guess. Currently the maps are crude and my point is,
why ignore generic information when we could get more? The more
generic (= common to most humans) info we employ, the simpler and
less computation will be required for MU (ie LESS brute force).

>-Why is "Hard lesioning" necessary for short-term replacement techniques?  Why
> not "soft lesioning?"  Why is no lesioning necessary for long-term replacement?

Sorry I've said something that's just my own speculation, and I need
to dig up more info on it. A neurosurgeon would know more about this.
The statement of the problem is "how do we lesion a small peice of
neural tissue and also the blood vessels within, without causing damage
/adverse effects to other brain areas?" In v3.3 I just call it "lesioning
techniques" without speculating on the details.

[Siv: You can upload files to my web site so
we can see them. I'll give you the password
in email.]



____________________________________________________________
Get advanced SPAM filtering on Webmail or POP Mail ... Get Lycos Mail!
http://login.mail.lycos.com/r/referral?aid=27005