From: Eugen Leitl (eugen@leitl.org)
Date: Tue Nov 05 2002 - 04:07:32 MST
On Mon, 4 Nov 2002, Jeff Davis wrote:
> The problem with the SHRDLU model?
>
> It's inadequate as a model because it lacks sufficient
> correspondence with that which it is intended to
> model. It may be instructive. It may be a jumping
> off point. But an aerial photograph isn't the
> territory, and a map even less so. And the SHRDLU
> model doesn't rise even to the level of a map.
Hmm, I don't think so, but let's see.
> First, Gene, you set as a condition, no system noise.
Actually, there _is_ system noise. That's why I brought in the video
camera, the machine vision front end and the robot arm. You can also
inject random or pseudorandom noise into the framebuffer or the robot arm
control. It doesn't matter what the origin of the noise is and where
exactly you inject it.
> It appears your intent was to construct a model which
> was, a priori and by design, deterministic. This to
No. I selected a simple yet nontrivial framework which can be
deterministic or nondeterministic.
> prevent forking of identical starting states, and thus
> prove that identical states stay identical. In doing
> so you beg the question. You do not construct a model
No, I intended to handle both synchronization and the forking. I have the
impression that some people have trouble understanding that evolution of
two synchronized discrete deterministic system results in identical
trajectory. I wanted to achieve consensus at this stage before moving on.
> like you do a product, to perform in a certain known
> and useful way. You construct a model to set it in
> motion and find out how it performs. You may not
> impose on a model a condition alien to the character
> of that which you are supposed to be trying to model.
> The real world is inherently and inescapably noisy,
> and consequently unpredictable, indeterminate.
Yes. That's why I included the option of using the world as noisy input. I
also intended to demonstrate that you can fuse forks with trajectory
forcing in practice.
> Secondly, the model is too simple. Gene, you didn't
> say just how simple it was, but you insisted that one
> must start with a simple system. But emergent
> properties, unpredictable behaviors, and multiplied
> sources of indeterminacy arise out of complexity. Yet
> you specify simplicity. Again, this works as a place
> to start, but lacks correspondence with the (vastly
> complex)system you are seeking to model.
>From the simulation point of view there is no difference between a worm
(fly, mouse, human) upload and a SHRDLU world but the scale. It's just a
question of more bits and a more complicated transformation function. This
is another important point to achieve consensus on. Apparently when we are
talking about uploading here people assume we're talking some nebulous,
fantastic technology we have no idea how to implement. Nothing could be
farther than the truth. It's just a discrete deterministic system. It's
governed by the same laws as software in the computer you're reading this
on.
> Finally there is the inherent difference between a
> designed system and an evolved one. The designed
So, use an worm emulation from first principles, then. It's clearly an
evolved system.
> system--the model you are building--is based on the
> notion of a 'designer' and his/her capabilities. One
> can imagine a deterministic and reliable design
> outcome if one supposes both a
> highly(thoroughly?,perfectly?) knowledgeable designer.
> We can even dispose of the problem of mistakes made
> along the way, by giving the designer sufficient time
> to work out all the bugs. And in a simple system this
> might be achievable. But a designer creates a system
> using a limited inventory of materials and techniques.
> The final design is unavoidably bounded by this
> limited skill set. The universe on the other hand,
> evolves, rather than designs. It throws a set of
> starting materials together and shakes them up and out
> pops reality. Everything that exists takes part.
> Expertise, and the known and unknown, don't constrain
> the outcome, because evolution doesn't deliberate, it
> just 'operates'.
I think I smell vis vitalis in above passage. I guess the only way to
dispel it is to repeat Woehler's feat with urea synthesis in the digital
domain. Seeing is believing, apparently.
> (Even more extreme in its inapplicability is John
> Clark's use of a Bose Einstein condensate to make the
> point that historical distinctions can be made, ex
> post facto, to disappear, and to suggest somehow from
> this that those distinctions never existed in the
> first place. Huh?!! Just how does the very special
> quantum peculiarity of a small assemblage of atoms,
> isolated in a vacuum at a nano-fraction of a degree
> above absolute zero, bear the remotest correlation to
> two hypothetically indistinguishably-identical vast
> and complex assemblages of atoms, chemically,
> physically and informationally evolving at room temp,
> possessed of and manifesting levels of complexity both
> partly grasped and suspected-but-as-yet-unassessed,
> which, despite rare moments of proportionate humility
> and justifiable awe, we nevertheless name "living",
> "sentient", "self-aware", "intelligent", and
> "conscious", and in so doing pretend an understanding
> both vastly greater than we possess and vastly less
> than we imagine? Forgive me for going adjectival on
> you, but I thought the BEC argument a bit more than a
> stretch. Later, I came back to rebut, on my own, my
> thesis of the utility of historical distinctions.)
Unfortunately, your paragraph, though nicely eloquent, is simply wrong.
Forget BEC. I keep repeating that you people look at the treatment of
equilibrium constant in Tipler's 'Physics of Immortality' appendices. I
have a reason for repeating that. When I say that quantum systems in the
same state must be indistinguishable orelse you'd die this very instant I
wasn't kidding. Equilibrium constants are something readily and quite
precisely measured. Life is chemistry, and equilibrium constants govern
every chemical reaction. If quantum systems in the same state were
suddenly to become distinguishable, the values of equilibrium constants
would shift, and the chemical networks that is life would instantly fall
apart. You would die.
> Segue time.
>
> A few days a go I stumbled onto something which seem
> an example of two things which share a single
> identity. And it seemed immediately helpful.
> (Perhaps I will seem a little slow for not having seen
> this sooner but, well, so be it.)
>
> Two copies of the same book. Two books that are two
> books. Two books that are one book. We can speak of
> them as two, and no one will dispute or complain. We
> can speak of them as one, and likewise no one will
> dispute or complain.
We've treated books before. They're trivial, because they don't evolve,
and they don't react to the outside world. Here's a model system that is
distinctly too simple th be of much value.
> If we then compare two copies of a book and two copies
> of a person, perhaps we can learn something about the
> identity question.
>
> The information pattern which is a book is static.
> The information pattern which is a person is dynamic,
> continuously evolving. The 'container' of a person is
> biological, the 'container' of a book is not.
>
> Finally, I was thinking about the 'transporter'
> business. It occurred to me that thirty-some years of
> familiarity with the concept, and with the attitude of
> fictional characters toward the concept, have
> influenced--I would say 'prejudiced'--our view of the
> matter. Most of the star trek folk use/used it
> routinely, didn't appear to consider the issues which
> concern us, and suffered no apparent ill effects.
> (Except for Dr. McCoy who "didn't like the idea"--does
> this mean Roddenberry was hip to the problem more than
> thirty years ago?--, and one fellow from TNG who was
> the host for a symbiont that would have been killed by
> transporter use.) Does this mean that we are required
> to accept the view of these fictional folks, or their
> authors, and everything is just fine? Setting aside
> the fact that we're dealing with fiction, let me point
> out that the accepting attitude is inseparable from
> the sanitary and conceptually inoffensive character of
> the process. But what's actually happening?
Apparently you forgot the transporter malfunction incident, which resulted
in forking.
> What if,... What if the transporter didn't neatly and
> painlessly destroy the 'original' at the start of
> transport? What if you had to messily bludgeon the
> original Capt. Kirk into unconsciousness, and drag his
> twitching and bleeding body down the corridor and
> stuff it into the garbage chute? The end result is
> the same, but now we're confronted with the unsavory
> nature of the business, which we are otherwise spared
> for the sake of a smoothly flowing narrative.
You're describing a fork. Transporters are badly treated in Star Trek
world, because they are not wormholes, they don't work instanteously. But,
did you ever observe the transported persons or objects move? I might be
wrong, but I didn't, ever. There's your loophole to prevent forking.
> Either there is no homicide, or the culture has
> persuaded its members that it's a case of no harm, no
> foul. But the former speaks against the facts,
> witness the blood trail down corridor, and the latter
> an abstraction we call tradition or cultural value.
>
> I wanted to be helpful, but don't seem to have made
> much progress.
That's because you rejected the model system I suggested. It is not nearly
as trivial as you initially thought. Instead, you go back to muddy waters
of physical systems, and imaginary technology (transporters), and faulty
physics (thinking that QM indistinguishability is pathological, while it
is in fact occuring in your body this very instant).
This archive was generated by hypermail 2.1.5 : Wed Jan 15 2003 - 17:57:57 MST