From: Anders Sandberg (asa@nada.kth.se)
Date: Tue Aug 26 1997 - 18:05:15 MDT
"Eliezer S. Yudkowsky" <sentience@pobox.com> writes:
> Anders Sandberg wrote:
> > "Eliezer S. Yudkowsky" <sentience@pobox.com> writes:
> > > 1A. Programmer writes a piece of code to do something.
> > > 2A. Code evolves into something small, dense and tight.
> > > 1B. Code leaps into new evolutionary space and discovers new algorithm.
> > > 2B. Programmer expands algorithm and generalizes it.
> >
> > Sorry, step 2A doesn't occur that easily. GAs tend to produce
> > bloated, redundant and weird code; their strength is that they are so
> > good at finding flexible solutions.
>
> WHAT? What about all those famous stories where a GA does in 28 steps what a
> human could only do in 31? GAs produce code that's tight where ours is
> bloated, and bloated where ours is tight. Hence the double cycle.
OK, a clarification. What you refer to is likely Tierra, where long
initial organisms evolve into shorter organisms due to competition for
scarce memory. I was referring to commonly used GA algorithms (the
Lisp-like trees invented by Koza), where there usually are no size
constraints.
If you make small programs more fit, small programs will evolve, but
if you keep the size contraint they will not evolve well (that needs
a large genome/code). So a better solution is to first evolve big and
baroque programs, and then make size more and more relevant, creating
shorter, less correct but more efficient programs. I read
some amazing results where they got 2-3 instruction programs that solved
a myoelectrical pattern recognition problem rather well this way! Of
course, in the case of immune systems you would not care much about the
code size.
> > Building a nuke: you need around 10 kilograms of uranium 235 (or whatever
> > isotope it was). There is around 2 grams uranium / tonne in the crust of the
> > earth, of which 0.72% is U235, so to get 10 kg you need to process around
> > 7000 tonnes of crust. I'm not sure how much energy is required to reduce
> > the UO2 to pure U, but it is a noticeable amount (are there a chemist
> > in the house?). Assuming the nanites cover a large patch with solar
> > collectors, they can get around 500 W/m^2, which has to cover their
> > replication, search through the crust, reduction, isotope separation and return
> > to the "base"; how much energy this is is a bit hard to tell right now
> > (it is 1.30 in the morning here :-), but it looks like it will take
> > a while for the bomb-mold to blow up. A wild guess would be around a
> > week.
>
> Well, there's a couple of questions I'd pose about nuclear nanotech:
>
> 1) Can mini-assemblers build mini-thermonuclear devices? If you have 10
> grams instead of kilograms, but you still put the deuterium at the center and
> jacket with U-238, could you get the same effect? What about custom-assembled
> crystals that will collapse easily and tightly... maybe laced with water to
> slow neutrons down?
No. A fission bomb is based on the creation of a critical mass, where the
growth of neutron flux due to fission is larger than the loss into the
surroundings. Unless nanotech could build a neutron reflector (which I
doubt is physically possible?) it would still need several kilograms of
fissible material.
> 2) What about direct fusion devices? Inertial confinement would be my guess.
> Take the deuterium and imprison it inside a fullerene-bracketed block of
> diamond, then apply a large energy pulse, perhaps from an enormously efficient
> quantum-well laser.
Doesn't sound very likely, although I'm more iffy on the physics here. First,
the entire structure would have to be macroscopic in order to do any damage,
so there is no real need to use fancy fullerenes to keep the hydrogen in.
And you would need to store enough energy for the laser, and somehow compress
almost all of the hydrogen, not just a small pellet (ah, thats the problem!)
which seems to be nearly impossible to do using laser light.
> 3) Given either of the two above, could you build a REALLY HUGE thermonuclear
> bomb, perhaps by surrounding the item with a huge U-238 crystal? Direct the
> blast in a specific direction using the properties of crystals, maybe with a
> bit of quantum mechanics mixed in? Use the blast to pump a graser?
Well, you could always pile up as much U and deuterium as you wanted around
it. The extra U would not really do much except make it dirtier (it will
mainly be blown away by the explosion instead of fissioned), but the
deuterium would add to the blast (that is why you can make fusion bombs
as large as you want).
Directed nuclear explosions doesn't sound likely, since at present we
are relying on random nuclear fission with no preferred direction. If
we could make neutron mirrors things might get different.
> In other words, nanotechnology operates as close to the level of nuclear
> reactions as we operate to the level of chemical explosives. I can't help but
> think that there will be similar improvements in technology.
That is likely, if there is an interest. But remember that nanotech isn't
much closer to the nuclear level than we are - it still cannot interface
directly with it, just move the atoms around. There are fundamental
physical problems here we need to look into to get the nano-nukes.
> > Doing the same work as a nuke with nanites (i.e. disassembling everything
> > within a few hundred meters and blasting everything within a few kilometers)
> > is rather tricky, since it is extremely energy intensive. You need plenty
> > of energy to do the disassembly (essentially you have to break most molecular
> > bonds), and nanites are bad at making blast waves.
>
> I don't see why. First of all, they'll be able to construct explosives
> completely to order.
But where does the energy in the explosives come from? Remember conservation
of energy - when you make explosives you have to add energy beside the
chemical energy of the raw material to make the energetic but unstable
TNT molecules (electrical energy -> chemical energy). The same is true for
the nanites - if they want to turn a lawn into a bomb, they need more
energy than will be released in the eventual blast.
> I'm far from an expert on explosives... but couldn't you get a bit of an
> improvement by detonating the ENTIRE bomb with electronic synchronization, so
> that the entire blast arrives time-on-target? A shock wave is just that, a
> wave. So modern TNT is a huge splash made up a lot of little uncoordinated
> splashes, while nanotech TNT is the precisely synchronized sum of all those
> splashes. Almost exactly like the distinction between a light bulb and a laser.
I think something like this is done already, although not with the nano
precision you suggest. You might improve the current state a bit, but it
is likely not a quantum leap.
> Again, I'm trying to point out that a bit of
> imagination is necessary to see the enormous destructive potentials here. I
> think we can safely extrapolate a *little* bit beyond current capabilities.
What I am trying to do is theoretical applied science - using known laws
of physics to discuss what can and cannot be done. You can always say that
unknown quantum effects will circumvent all that, but it is an empty
speculation until you show that these unknown effects really do exist.
Note that we should speculate beyond current *capabilities*, not current
*physics*.
> Most of what you're saying has nanotech being less dangerous than modern
> technology. I can't help but compare this to the old saw about "Atomic
> technology may someday equal the capabilities of present bombs, but it is
> unlikely to produce anything more dangerous." [Paraphrase.]
Well, if that turns out to be history's judgement on my part in this thread,
so be it. But remember, there were people expecting a cure for the common
cold in the early 70's too...
> > Ever thought about the fact that the immune system isn't mutually supporting?
> > It is rather a balance of predators which do not predate on each other since
> > that destroys their fitness.
>
> Interrogative? Expand, please.
I would like to get some expert in on this, since my medical knowledge is
more neuroscientific than immunological (I'm not an immunologist, I just
pretend on the net :-). Basically, the immune system is programmed to
destroy, destroy, destroy - everything. But during the complex process
where white blood cells are bred the cells that would kill unsuitable
things (like each other, the body etc) are weeded out, as are the "meek"
cells. The rest are just barely tamed, and will with no hesitation destroy
any cell that doesn't show the right antigens, and everything in the
vicinity.
> > Good guys will be motivated to contemplate directed evolution since the bad guys
> > do it; and they tend to have more brainpower and money on their side. I
> > think the weapon of openness is important here.
>
> It could go either way. Maybe the bad guys kidnap all the scientists and then
> use pain-center stimulation to achieve faster results. Idealism may win
> battles and break ties, but it is not a defense to be trusted.
Or maybe the good guys spread a nanite which give all bad guys migraine
when they think evil thoughts. Get real, this sort of unrealistic Hollywood
speculations is the last thing we need ("... if you do that, I'll send in
my *dragons*!" "Ha! I have an army of invisible pink unicorns that will
eat your dragons!").
> Alternatively, it might only have to take over a single power base. Your
> question assumes that the goo, once detected, is doomed. My version assumes
> that the goo, once established, is undefeatable. This is pretty much what
> we're arguing.
It seems so. But as others have pointed out, your claim that destruction is
fundamentally more efficient than creation suggests that if I know about
your goo I can destroy it.
> How can I make quantitative comparisions between systems that don't exist?
> I'm just trying to point out the qualitative causal forces at work. When
> we're finished cataloguing, we could try to set up a quantitative comparision,
> bearing in mind that it would be utterly false-to-fact and useful only for
> speculative purposes.
We can't do any cataloguing based on a priori ideas (qualitative causal forces),
we need something more solid than that. I prefer theoretical applied science,
it allows us to compare not-yet-existing systems.
> > The body is surrounded by an inert skin (say diamond);
>
> Isn't diamond something that can be set on fire at sufficient temperatures?
Yes, but so is almost everything. I never said I would sketch an
*invulnerable* system, just a sufficiently strong system. For any immense
shielding I can come up with you could always invoke an even greater
cosmological disaster ("But your immune system can't stand a supernova!").
This exercise is trying to look at defenses against gray goo, which
is the real problem, not macro-level warfare.
Diamond is an useful start, but as an outer skin I would actually use something
more inert.
> Isn't the crystal fragile when properly angled forces are directed against it?
This is a mere matter of design, I don't think diamond crystal is the best
way to go, more likely a flexible matrix of jointed diamondoids.
> How much of an insulator is it for temperature? What about gamma rays?
It conducts heat well, and as far as I know is transparent for gamma rays.
But you are not seriously suggesting grey goo radiating gamma rays?
(that would BTW be very bad for the goo too, since nanites are radiation
sensitive; and I could have radiation shielding in the skin while the
goo would be unprotected (since grey goo carrying around lead is easy
to find and exterminate by taking it away)).
> > attempts to
> > physically breach it can be detected from the inside and the surrounding
> > region sealed.
>
> Sealed by what? If it's better than diamond, why not surround the whole body
> with it? If the sealant is worse than diamond, or is diamond, why wouldn't
> anything capable of penetrating the external wall go right through the sealant?
Diamond does quite nicely here, although even more inert materials would
do better. The breaches can be of two kinds, a strong
external force making its way in, and in that case we are talking about
macro level defense and offense again, or nanites forcing their way through
the walls. In the later case, they would need energy to breach the
walls if they are inert, and it would be exothermic for the defenders to
thicken their walls. If they turned the surrounding compartments into
a "diamondoid scar" the energy in the infected compartment would not
be enough to breach the barrier, and the whole infection would require
a lot of external support (which could be broken).
> What detector mechanism are you using? Can I falsely trigger it over the
> entire body, thus making the "detector" useless and redundant?
I have been thinking of having nanometer "pipes" criss-crossing the barrier,
containing linear molecules which break or vibrate when interfered with. You
cannot detect them from the outside (just diamond, and then an opening -
oops!), and the signal into the compartment will give information about
where something is going on.
> If sealed-off
> areas are easier to attack, wouldn't the false alarm let me peel off an entire
> layer? And why couldn't I just repeat the action until you were the size of
> an apple core?
Sounds like something that could be designed away. OK, I haven't yet
tried to sketch how the control systems ought to look (I haven't got the
time even for doing this discussion properly! :-( ). The software
part is of course even more important than the hardware part.
But basically, if you trigger all the surface cells, don't you think the
system will immediately (assuming some intelligence somewhere) guess
the strategy?
> > The rest of the organism (could be a transhuman, factory
> > or a city) is compartmentalized with similar walls; infected sections
> > can be isolated.
>
> Okay, so you've made the problem of immunity recursive on immunity for a
> single section. This may actually be an excellent plan of defense, allowing
> for partial breaches and reserves. But a design for an immune system it's not.
It is a partial solution, and makes the problem more manageable. The body
uses something fairly similar with its internal barriers (it is surprising
how strong the blood-brain barrier is, even adrenaline can't get through!).
> Okay, I see a trend developing in your plan. Detection is the important
> thing; it's assumed that pockets of infection, once detected, can be
> obliterated. Very much like the human immune system. But if viruses could
> outfight white blood cells, we'd be dead. I'll conceed that fooling detection
> might be made difficult, almost impossible, through the use of digital
> signatures. But again - how do you keep a single cell from being penetrated?
> Once a cell is penetrated, you destroy it - but does the destruction keep the
> infection from spreading? A single spore might be defeated by this strategy.
> A sea of gray goo won't be.
A sea of gray goo is also a macro-level problem. It is obvious, it can
be fought on the macro level (meaning that human or transhuman intelligence
have the time and space to implement defenses not just based on what their
ordinary immune systems can do, but also their other available resources).
For example, it wouldn't appear impossible to nanofacture "antibodies",
very simple nanites which disrupt the goo nanites, en masse once the goo
is understood.
I have noticed that your approach seems to be that when subtle attacks
doesn't work, you invoke maximum overkill instead. But that takes a lot
of resources, and can often be disrupted more easily than a subtle attack
(compare "AAARGH! The goo has my password!" to "Ouch, this goo is corrosive,
let's wash it away in a shower of hydrogen peroxide").
> No comment. Part of it isn't very complicated, given digital signatures.
> Nor, as far as I know, can a nanotech penetrate a wall invisibly. I'll
> concede you perfect detection. For this discussion, it's the fight I care about.
Thank you. Let's go down to the real battle - dismantling nanites.
> As far as I can tell, your strategy simply lets
> me destroy your entire complex cell by cell. Penetrate a cell, watch it go up
> in flame... repeat.
Hmm, like Stalingrad? ;-) You assume that the destroyed cell will no
longer be a problem. But what if it turned into a cube of inert diamondoid?
Then it would also be a hinder, and give me even more time to develop
countermeasures.
I see that I have been a bit too defensive (I usually am in strategy),
since you are right in that just slowing the advance of the goo won't
win the fight. So, to get back to how the body works, I'll better
develop antibodies and macrophages.
An antibody would be a very simple device that sabotages a certain nanite.
It could be a molecule that binds and then tends to stick to other
antibodies, trapping the goo nanites as dimers or polymers. A macrophage is
simply a selective disassebler, which destroys a certain kind of nanite
or structure. Assuming I have detected the infection the next step would
be to make antibodies and macrophages against it, and start spreading them.
Note that they would be produced faster than the goo since the goo would
have to both reproduce, breach security and defend itself while the
antibodies and macrophages would just be produced (although a goo-like
macrophage is an interesting and dangerous concept) and sent on their way.
> The real obvious one is this. Your immune system - any immune system
> conceived along these lines - requires that two nanosystems fight each other
> to a standstill. If two nanosystems fight and destroy each other, then any
> city in an island of gray goo is doomed.
I think you are thinking too much of ants fighting here. I see it more
as chemistry, especially protein interactions. It is much easier to make
something which binds goo into an inert mass than to build a goo eater.
> The goo simply makes repeated
> attacks, and each time, the city shrinks a little. We won't even speak of
> such horrors as cutting off that city's solar power.
An isolated immune system in a world where nobody else has immune systems
is weaker than a system in a world where nanodefenses are common. The ultimate
defense would of course be to have immune systems everywhere, defending
not just the transhumans but the himalayas, squirrels and grass. A bit like
the current state in biology, really.
> In short, the immune system requires that *defense* have an advantage over
> *offense*. I can conceive of any number of ways that this could happen. I
> shall now proceed to list them.
Yes, this is the crux. I contend that defense is easier.
[Some good examples deleted for brevity; I get the impression we are
discussing nano-sized knights fighting on the battlements of a transhuman
city :-)]
> Moravec has a great discussion of exotic forms of matter(1). "Higgsium" is
> 1e18 times as dense as water; "Monopolium" is 1e25 times as dense. My
> impression was that Higgsium might be hard to penetrate, even using Higgsium
> projectiles; only Monopolium would suffice. With Higgsium and without
> Monopolium, then, any area taken might be easy to hold onto. With Monopolium,
> the reverse is true.
>
> Similarly - and here I speculate without knowledge - there might be a way to
> give a material an unbreakable "quantum surface tension". Or subbaryonic
> picotechnology - "chromatic technology" - might use "quark barriers", surfaces
> as unbreakable as a proton. (From what I know of quantum chromodynamics, this
> is incredibly unlikely. The bubble would have to be pushed out by truly
> inconceivable forces; pre-Plank-time Big-Bang-type forces. But hey, what do I know?)
OK, but at this level I simply escape into my own basement universe (Mitch
suggested that you could even hide backdoor wormholes inside atoms for
unexpected ambushes). Dragons vs. invisible pink unicorns again?
> Then there's the simple possibility, dependent on molecular details of which I
> know nothing, that defense on the molecular level has an innate advantage over
> offense. Certain molecules or nano-screens might simply be easier to form
> than to break, just like atoms or quarks.
Good point. There are certainly some molecules which are much easier to
break than to form, and hence allows me to make good detection systems or
nanites whose programming is erased if they are opened. I think there might
be some structures which are very troublesome to destroy on the nano level.
> Nanosystems are always faced with "destruction by induction". That is, you
> can always destroy one cell; therefore you can destroy the whole thing. To
> defend against this, it is required that the system expand faster than the
> destruction OR that it be impossible to destroy one cell.
Or that you can make the loss of a finite number of cells bearable. If their
loss removes the threat (for example by forming a nanoscar), then the
defense side will win.
> Speaking of escape, there's another workaround for "destruction by induction".
> If you run into space, the gray goo might hurl itself after you, but it won't
> have the overwhelming resources required for "destruction by induction". In
> addition, the effort of destruction might simply provide the city with a
> velocity boost. Certainly things in space will be different than on the ground.
Another good point. While I believe in active shields, space is IMHO the only
really proven form of defense against goo.
> So there are a lot of immune systems, just as there are many forms of attack.
> It's just that defense being favored over attack requires too much exotic
> technology; the exotic attack technologies are far more conservative than the
> exotic defense technologies. The majority of strategic considerations are in
> favor of attack; the only really hopeful strategic consideration is the city
> being able to quickly expand into rich zones of destruction.
Hmm, we still disagree. I don't see why you have to invoke exotic technology
for this. But I think we have touched on some of the most important
aspects of nano-warfare, and to quote Arnie, "I'll be back". Let's try
to get some real nanotechnologists, immunologists and strategists into
this discussion.
> Ergo, it is easier to destroy than create.
Om mani padme hum... :-)
> (1) "Harvard doesn't publish science fiction." New Destinies III, Spring
> 1988. Edited by Jim Baen.
What is this? Sounds interesting.
(now I have to get to bed - why do Elizer have to start a good argument
2 in the morning? :-)
-- ----------------------------------------------------------------------- Anders Sandberg Towards Ascension! asa@nada.kth.se http://www.nada.kth.se/~nv91-asa/main.html GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:44:46 MST