RE: clones/'perfect'-mates/self-image

Eugene Leitl (root@lrz.uni-muenchen.de)
Mon, 28 Dec 1998 23:02:22 +0100 (CET)

On Mon, 28 Dec 1998, Hal Finney wrote:

> It is not necessary to download; copying uploads should be much easier.
> Of course, you have to be able to upload in the first place, but that
> may not require advanced nanotechnology. It could be as simple as some

Absolutely. One of the major points of uploading is that you don't need to make your reconstruction verbatim (in the flesh), just run a number of (pretty smart) data filters over a (pretty large) dataset. One of the more trivial instances would seem to be the substitution of vitrification agents by pristine vitrified water purely in a computational model (such a filter does not require much intelligence, a human could certainly write it). Such an early filter stage is obviously useful in uploading, but absolutely crucial in nanoresurrection (vanilla cryonics), where a patient could be disassembled via micron-thin fronts of dissassembly/reassembly, with the filtering done interim. In a sense, the patient is washed over with material waves of processing, during each being piecemeal-wise yanked into the virtual realm and back.

> kind of high-resolution MRI, or perhaps it could be done by freezing you,

High-enough-resolution MRI sans sample destruction is physically infeasible. However, even now cryo AFM achieves molecular resolution on tissue cryosections, and this technique is in principle combinable with tip abrasion/freeze/UV etch. Making a molecular-resolution map of a vitrified tissue block is actually quite straightforward, i.e. you can already see all the neccessary technologies, and assess the difficulties. It looks indeed quite doable, imo.

> slicing you up and scanning in each slice, then running software which
> simulates the effect of undoing the freezing damage.

Yep. Such filtering stages obviously need to be neuronal. You could imagine simultaneously applying a ball-shaped "3d retina" to each area of the voxel dataset, simultaneously removing all kinds of artefacts, from low to high level.

> Any upload will probably require considerably more computing technology
> than is feasible today, but there are other paths than super nanotech.

Any upload will _certainly_ require considerably more computing technology than is feasible today, indeed the artificial reality renderer alone is considerably beyond the state of the art. So what? Molecular circuitry of any flavour is certainly feasible (look into mirror if you don't believe me), whether Drexlerian, or not. And recently the progress in Drexlerian nanotechnology is looking very good indeed. Suddenly, vitrification of macroscopic human organs appears within reach, while simultaneously research in fullerene and diamondoid autoreplication by mechanosynthesis is making giant strides forward. Interesting times, and all.

> > If uploading is possible at all, it should be feasible to re-engineer an
> > uploaded mind just like any other piece of software. You could run multiple

Difficult. An uploaded mind is not exactly 'software', just a giant blob of data, being directly 'executed' by dedicated hardware. It is far from clear how you can mutate that relatively opaque blob of data towards a certain goal. Perhaps in increments, over a population...

> > instances of your sensory-processing and motor-control software in parallel,
> > allowing you to control more than one body at once. For better fault

Err, there are no clean interfaces. You would have to restructure much of your mind. While I can see how to write a vitrification/scanning artifact removal filter, I cannot see how to constructively operate at such high abstraction levels. Either we learn the tricks necessary, or we need something quite beyond the human level to analyze, and to recast us in a new shape.

> > tolerance you could even turn yourself into a collective mind, with local
> > processing in each body so that they can function if your data links go
> > down. Spread yourself across a few hundred humanoid bodies and a similar

In that context, the body is just a servo. If you are talking about a cluster mind (borganism), the body is highly secondary.

> > number of useful robots and/or vehicles, and you become very hard to kill.
> > How's that for an interesting post-human mode of existence!
>
> It is interesting, but I question the assumption that uploading will
> automatically imply the ability to re-engineer minds.

Agree. One could imagine some incremental morphing, but this seems to be an invasive enough process to blow away the personality we so laboriously set out to conserve during the upload.

> Theoretically, uploading is a rather mechanical process: it is merely a
> matter of simulating a particular physical system (the brain/body) at a
> sufficient level of detail. The technical difficulties are a matter of
> sensing the object at that resolution, and having enough computer power
> to run the simulation. But perhaps these can be overcome.

If you don't need the quantum level, the amount of computation necessary is large, but containable.

> Re-engineering brains requires a wholly different level of understanding.
> Uploading is like painting a copy of a Rembrandt. Re-engineering is
> like being Rembrandt. It is a creative action, not a mechanical one.

I don't quite agree. In theory, you could imagine a purely mechanical process transferring an upload from some low-level encoding (say, around comparmental simulation level, or even MD) to some more abstract, nearer-to-hardware level in a mindless, purely darwinian process. I could imagine it working in increments, transmuting the wetware incrementally, block by block.

> You would need detailed understanding of how the brain works in functional
> terms. You have to know what to tweak and how to tweak it. You would
> have to understand consciousness and how it is related to brain activity,
> a matter which appears intractably difficult today.

Don't think so, for above reasons.

> True, having the ability to upload may help with our understanding of
> these issues, by giving access to neural function at a fine level of
> detail (and possibly allowing experimentation on uploads). But I think
> it is going to take a long time before we are able to re-engineer minds,
> even once we have nanotech.

How "long" is long? Remember, the time base ratio is certainly about 1:1 k, maybe even as high as 1:1 M, and these ALife golems won't certainly be long to wait for, nor likely to be terribly idle once there...

> Hal
>

ciao,
'gene