Stephen Hawking: Why Isn't the Milky Way Crawling with Life?

[Wyrm]

Would a self-replicating probe actually need to be adapted to such a wide variety of environments? It would most sensibly use asteroids and comets/KBOs as raw materials for its replication, and one solar system's Kuiper Belt or asteroid belt does not seem to me like it would be so radically different from another's. True, the probe might need equipment to exploit a variety of objects of very different compositions (equipment to extract the carbon and other useful materials from a dirty snowball vs. equipment for doing the same for a dry hot carbonaceous asteroid might be rather different), but aside from issues of temperature and radiation from proximity to the local star we're talking about fairly similar working environments (hard vacuum).

You skipped an imporant step: Is any of these places suitable for a VNM to set up shop? You need a place where you have everything necessary to refine the raw resources into suitable parts and expendables. Most industries require some water (universal solvent), and if you're assuming that the VNM is made of mostly metal, then the best place to find the water is not going to be the best place to find the metals. You'd have to do some trucking around, which will take time and propellant.

A self-replicating probe might also be programmed to skip replication in systems that did not contain resources it could readily exploit. Even if you only take equipment to work in an environment like our asteroid belt, say, a large number of solar systems will probably have similar asteroid belts. Replication would be slower, but you could still take advantage of exponential growth. Especially if you programmed your probe to replicate like crazy whenever it found a suitable solar system, instead of a simple one solar system surveyed one replication strategy.

The balance is not clear-cut here. If marginally useful systems are overwhelmingly common, the it may be worth it to replicate in a marginal system even if you can ruthlessly exploit an ideal system.

And where are you going to get the fuel and the oxidizer for the rockets?

I don't see why this would be a big problem. Find a suitable small icy object (comet, KBO, icy moonlet), grab some water, crack it into hydrogen and oxygen with electrolyses, and you have your rocket fuel and oxidizer.

With what? Rainbows and unicorns? You have to supply energy to crack water — the fact that you get energy out of burning hydrogen and oxygen should tell you that. In the Kuiper belt, you would have to expend your own resources or gather it up from your surroundings. The Kuiper belt is not known for its heavy elements, so any energy gathered would have to be from the star. The far away star, where the solar radiation is so weak that it would take ridonkulously huge solar panels to even work the electronics.

Of course, chemical rockets are about the worst thing you could use for interstellar exploration, but it's not hard to envision lots of propulsion systems for which "wilderness refueling" on icy bodies would be practical.

Water in and of itself is useful as reaction mass, but that's it.

The reason we can mine so easily on this planet is because we have an active geology, where the heavier elements get churned up mechanically and separated chemically. Asteroids are significantly lacking in that geology.

Actually asteroids would be much richer in metallic elements than planets precisely because they never had active geology. In a planet almost all the juicy heavy elements sink to the totally inaccessible core early in the planet's history. With asteroids the heavy elements remain evenly mixed in with the silicates and other chaff. Even C-type asteroids might be as much as 22% iron by mass, and you could probably scoop easily processed nickel-iron granules right out of the dirt.

Given that, in the particular section you were responding to, I was after radioactives, iron is not going to do me much good. Even if I could use iron, I'll still need to refine it and fabricate parts, which will take energy.

But let's examine that for a minute. First off, your observation that nickel-iron nugets can be scooped out of the dirt indicates one of the problems with asteroids in terms of mining. The minerals are evenly distributed through the asteroid's mass, and the local concentrations of the minerals not that large (granules) and mixed with a large amount of impurity (dirt). In order to get at most of the nickel-iron granules, you'd have to dig through a great deal of the asteroid, and you have to put the waste somewhere.

The situation on a planet is much different. There's one overwhelming advantage to planets: planets are big. This means that, even if minerals are distributed completely randomly through the mass, there will be statistical deviations from the average composition which are, while small on a relative scale, on an absolute scale are huge. And this is just with pure randomness. The distribution of deposits on Earth is not random, as they are dictated by geology and chemistry, and we've found most of the accessible mineral resources over the course of a few centuries. With a deposit, you can simply settle down in one place for bulk mining and refinement, and there's plenty of room to put tailings. The tailings will be smaller because the usable fraction is so concentrated.

For a first place to settle —when energy is at a premium— a planet is the way to go. While an asteroid belt will contain the most usable material overall, it's not all in one place, and grouped into easily-accessible and sizable deposits as it is on a planet.

Again, you theorize the existence of an entity that can replicate itself, has intelligence, is smart and clever enough to construct the underpinnings of a technological civilization, and at the end must surrender itself, its children, and their combined labors to usurpers of dubious worth and authority.

Since its creators program its goal system I don't see why this is a problem. The "robots will rebel" concept is based on the idea that robots will have tendencies similar to organic life (they will want to preserve themselves and their descendants, will value their own existence, will value their self-determination etc.), but there's absolutely no reason an intelligence designed from the ground-up by us would be like that. We have those traits because natural selection has selected for them; there's no reason for a designed intelligence to have them.

Uh, yeah. You realise that the reason we have these traits is because they're useful in an unpredictable environment. A truly autonomous machine will have these traits too in some fashion.

I'd also point out Peter Watts's take on consciousness in Blindsight, which is that it may not be necessary for an intelligent mind. If that's true an "idiot savant" AI might be every bit as smart as a sapient one, it simply wouldn't be self-reflective.

Not reflecting on what one has already done is a danger to long term plans. A truly autonomous machine needs to be able to reflect on its past performance and take that into account as it proceeds towards its goals. It needs to be able to know what it's doing.

Of course, even a sapient AI would have no reason to object to its "slavery", assuming it's designed competently, since its goal system would be completely designed by us (not by natural selection, which would naturally create self-interested intelligences).

The thought of creating such a being fills me with disgust and revulsion.

That aside, mutants are a danger to us. Our own sapience is a bolt-on to a package of more primitive instincts designed to keep us intact and successful in an uncertain environment. A truly independent machine will have something like them, with sapeince as a bolt-on. Although the goals may be initially subservient to humanity, a mutant can turn into a monster by simply uncoupling that subservience from the sapience.

[Samuel]

For a first place to settle —when energy is at a premium— a planet is the way to go. While an asteroid belt will contain the most usable material overall, it's not all in one place, and grouped into easily-accessible and sizable deposits as it is on a planet.

No. While there will be good deposits of minerals, they will be seperate from each other and getting from one to another requires larger expenditures of energy than in space.

You realise that the reason we have these traits is because they're useful in an unpredictable environment. A truly autonomous machine will have these traits too in some fashion.

False. We have those traits because those who do not fail to reproduce.

The thought of creating such a being fills me with disgust and revulsion.

Why?

[Wyrm]

For a first place to settle —when energy is at a premium— a planet is the way to go. While an asteroid belt will contain the most usable material overall, it's not all in one place, and grouped into easily-accessible and sizable deposits as it is on a planet.

No. While there will be good deposits of minerals, they will be seperate from each other and getting from one to another requires larger expenditures of energy than in space.

While the minerals in an asteroid belt will be... separate from each other by the fact that its spread throughout an entire asteroid belt, which will require large expenditures of energy and propellant.

You realise that the reason we have these traits is because they're useful in an unpredictable environment. A truly autonomous machine will have these traits too in some fashion.

False. We have those traits because those who do not fail to reproduce.

And they fail to reproduce because they... suffer from mishaps that cripple or kill them. In an unpredictable environment you need to avoid hazards — that's why you have these instincts, so an autonomous robot will have 'instincts' for self-preservation. If it doesn't, its a waste of resources.

Also in an unpredictable environment, you need to be able to take advantage of opportunities as they avail themselves. An autonomous robot will need to be able to do the same thing to efficiently exploit its environment. Some kind of 'ambition'.

Why?

Creation of an intelligent, sentient being whose sole purpose is to be our slave. If that thought doesn't give you some pause, I don't know what to say.

[Simon_Jester]

Barring technology indistinguishable from magic, I'm not optimistic about the concept of von Neumann machines. I suspect that the minimum size for an interstellar von Neumann machine is "a large group of sentients with machine shops, stockpiled raw materials, and extensive computer databases." And at that point, the difference between sending a von Neumann probe and sending a colony drops to zero.

Over interplanetary distances (where we can scope out the local resources more efficiently before sending the probe), von Neumann machines make more sense. As do limited pseudo-von Neumann machines that can make other machines, but cannot make themselves: if all we want is ball bearings, we might as well send a robot that can make ball bearing factories but not "ball bearing factory" factories.

But now that I really think about it, the engineering challenge of colonizing the galaxy with VNMs seems a bit prohibitive. Maybe a lot of intelligent species failed to solve that engineering problem even if they are otherwise advanced and long-lasting.

Yes, but we can say that as a simple matter of statistics the more technological civilizations there are the less plausibility explanations like "nobody cares enough" have, because larger populations are more likely to have deviations from the norm and it only takes one deviant civilization to fill up the galaxy in very short timescales compared to its lifespan. This is why I favor a number of filters instead of a Great Filter. You don't need to worry about how every Earthlike world in the galaxy must produce a civilization apathetic to interstellar colonization; very few of such civilizations arise because other filters catch most of them before they can get that far, and there may be more filters ahead the catch the tiny handful that slip past this particular one.

Agreed.

Maybe by 'remarkable' he means 'remarkably stupid'. It's not just bad authors who follow the train of thought 'aliens not like people = aliens not like contemporary culture = aliens are like xyz ancient culture = aliens are contrived political soapboxes'.

In fairness, if we meet aliens that aren't totally alien in the Cthulhu sense, they're going to have at least some features that we can look at and go "Hey, yeah, that reminds me of XYZ culture." The way those elements get put together may not resemble any Earthly culture that ever did or ever will exist, but humans are quite good at drawing tenuous analogies that don't work when expanded to cover a larger subject area.

Again, if we meet totally alien aliens, all bets are off... but totally alien aliens are hard to work with in fiction. By definition, we can't understand what the hell they're thinking or doing.

[Narkis]

It is quite possible that we'll meet aliens that we can understand, and yet are totally different to us. I remember reading a short story about humanity meeting an insect-like alien species that gave birth to vast amounts of children, but ate 9/10 of them due to a lack of enough food to feed them all in their prehistoric times. Their culture considered baby-eating the most virtuous thing one could do. I doubt any human could draw a connection, however tenuous, between them, and any culture that's ever existed.

[His Divine Shadow]

And it might be only one or two civilizations that do it. Even on Earth in most cultures reclusive hermit kingdoms and extreme lack of interest in exploration are the norm.... The Age of Exploration and the Ming Dynasty were aberrations. The unusual thing about humanity is not that we are intelligent but that we are industrialized. The Romans for example produced examples of steam engines and developed highly sophisticated knowledge of the universe... And kept using slave labour for nearly everything, because there was no motivational pressure to rely on for industrialization. The development of industrial civilization was a unique combination of the geological conditions of the British isles, the cultural conditions of western Europe generally, and the historical circumstances of the Black Death wiping out a large fraction of the population, happening at just the right time to put the necessary technology in the hands of people afterwards that they began to exploit it to begin a process of industrialization.

What we are probably going to find when we venture to explore other planets is a hell of a lot of Qing Chinas--vast centralized Empires that have lasted for ten thousand years in which life proceeds through exploitation of the local ecosystem through physical labour, whatever that ecosystem and whatever their exact structure may be, a common theme we'll find repeated in ten thousand variations.

Even sapient life, where it exists, is quite simply almost certainly not very sophisticated. Our random luck of the draw was in finding a section of the globe laden with coal and steel and other such resources, to find itself with a severe labour shortage at a time when it had the tools to remedy that labour shortage mechanically, and the inclination culturally to do so... Remove any one of those single components and it never happens.

I find this a remarkable notion. Would you object if I use it as the premise for original fiction? It seems to have immense story potential.

You should read Guns Germs And Steel by Jared Diamond which explains in detail why some people developed advanced technological civilizations and others remained at more primitive levels. It's a really good book, it comprehensively goes into the various reasons and factors.

You need a lot of factors and luck to get beyond just the hunter gatherer stage, such as geographical factors. Having access to steel and iron ore and such is secondary to having the right kind of fauna, suitable native plants and animals that can be domesticated, both are exceptionally rare in the animal- and plant kingdom. And it's very rare for any civilization to invent all the stuff on it's own, a lot is received by neighbors, herein natural obstacles and climate also factor in.

For instance eurasia was the perfect spot to spread plants required for agriculture (and also technology), from western europe to china you have a huge uniterrupted span from east to west of pretty much the same climate.

Not at all like say the americas which where divided up mainly by north/south and had lots of climatic changes therefore making climatic barriers that prevented plants from north america to migrate to south america. And I am just scratching the surface here of all the factors that come into play to get an advanced civilization running.

[Starglider]

A 'von Neumann machine' has a specific definition, and organic life fits that definition like a glove.

Semantically, no they do not, because they are not machines. That makes bacteria useful for establishing absolute lower limits - in that anything evolution can make, intelligent designers can make at least as well. Furthermore while macroscale organic life exists all of its reproductive processes and nearly all of its resource processing operates at the molecular scale, so it can tell us nothing at all about how machines designed to employ macroscale replication will perform. Finally organic life can of course tell us nothing about functioning in interplanetary space.

In fact, the way real organisms replicate themselves is a pretty good match to the way von Neumann machines replicate themselves.

No, it is not. Unicellular organisms reproduce by fission. The reproductive mechanisms of multicellular organisms are developed from that; they start with a single cell which slowly grows into a new organism. It is required to maintain basic metabolic function throughout the entire process and is required to be a viable independent organism from birth onwards.

Mechanical self-replication is nothing like that - in fact it would take massive effort to emulate this grossly inefficient system, for no gain at all. Robotic self-replicators build full-size copies of themselves which are generally inert until completed and then fully functional upon activation. This includes mental capabilties; robots can trivially copy their mind-state, organisms cannot.

Hint: von Neumann's chief insight was that the blueprints are their own specification.

As with science and technology in general, that insight on its own was pretty obvious. von Neumann's real achievement was creating a (basic) functional model showing how such a machine could actually operate.

Even if you were correct, yours would be an irrelevant point, because organic life faces exactly the same problems dealing with their environment as VNMs would.

Organic life actually faces far fewer problems, because evolution targets simple reproduction, not interstellar travel. Even spreading to other biomes on the same planet is a side-effect, not something that is directly selected for. However this is irrelevant, because the characteristics of the solution are not solely dependent on the shape of the problem, they are the product of that and the tools available to solve it, and the toolset for self-replicating robots is entirely different (and vastly more capable).

You appear to be imagining a single, static design of machine that must reproduce itself exactly.

I do not. A lifeform that is able to aggressively utilize a particular resource will not be able to survive on a different resource, as it is too optimized for that resource to effectively utilize another. A constructor that is optimized for using one resource would be ill-suited to using another. That's a basic engineering tradeoff.

How is that in any way a response to my point? 'A constructor', as in a single physical organism or machine, is optimised. Robots can build completely dissimilar robots and even self-modify as required. Organisms cannot. You refuse to recognise this basic distinction.

I'm basing my reasoning on an exant model

You are assuming all technology is exactly like bacteria, a ludicrous and transparently false position.

while you're basing yours on fanciful, hypothetical machines with hyperbolic capabilities

We already have industrial robots that build other robots. They are completely dissimilar models. Being able to construct an exact copy is actually a rare case; occasionally useful, but most of the time it is more efficient to have a small number of robots dedicated to building more robots with the other robots all optimised for different tasks. Even organic life, limited as it is, hit on a very limited version of this with some insects; worker ants form the vast majority of the colony, but do not reproduce. Most mammals have a very limited degree of task specialisation keyed off the gender distinction that evolved to support sexual reproduction. Of course with the horrid limitations of available selection pressure, chemical cascade expression mechanisms (for differentation during growth) and unstable DNA as the information storage medium, this differentation could not proceed very far.

In short, they are not 'fanciful' capabilities, they are not only already demonstrated in industrial settings, they are how any sane engineer would expect robots to work.

Even a simplistic von Neumann design would realistically carry a library of blueprints, and it would build machines appropriate to the local environment.

How many different blueprints will you need? How large is the universal constructor you need to carry? What's the resulting weight of this library and constructor?

Again you are acting as if the limitations of organic life were relevant, ignoring even the (relatively primitive) computer you are using to access this BBS. DNA is reasonably compact at the bit level (only a few of orders of magnitude off optimal), but it is horribly redundant at both the message level (lots of junk DNA, lots of duplicate copies of proteins, no data compression) and the organism level (humans have 100 trillion copies of their own blueprints... for no good reason at all). A robot of equivalent size and complexity to a human, storing a conservative 100 copies of its own blueprints with lossless compression in moderately efficient optical storage, would use approximately one quintillionth the mass and volume that a human uses for the same purposes. You could include a million different robot designs for different purposes and environments, and that's still a trillionth of the mass a human devotes to DNA.

Construction limitations are more relevant, but not much. The mechanical side is straightforward; the same basic manipulators and bending/cutting/joining methods are used for the vast majority of machines. The real limitation that stops us from building self-replicators right now (other than AI) is chemical processing, but we are making rapid progress on this. Since you're so fond of organics, look how biology does it; DNA is only capable of directly coding for polymers of amino acids, transcribed and synthesised by a single relatively straightforward nanostructure. All the other materials in an animal are produced by chemistry carried out by molecular tools (catalysts with complex molecular structure, mostly) created by those ribosomes. Molecular nanotechnology would use a similar bootstrap process, only with more steps and higher efficiency since it can use a much wider range of chemistry and techniques such as channeled transport (don't have to do everything in the same flexible bag of water). However molecular nanotechnology is not a prerequisite for VNM, because the same principle works at the macroscale; building the tools to build the tools to achieve the goal. It's less efficient, to the extent that your probe may well become a multi-thousand tonne automated factory rather than the tiny probe you could produce with nanotechnology, but it is certainly workable.

Remember also that this library and constructor has to be fault-tollerant against many centuries of hard interstellar and cosmic radiation, as well as sheer time.

While these are not easy problems, they are well within the scope of existing engineering skill, both as extensions of techniques we already use on contemporary spacecraft, and the numerous proposals you can find in both academia and the sci-fi community. They also apply to all interstellar craft, not just VNMs.

Also, you're imagining a universal constructor that could construct absolutely anything, when even in organic life, no such damn thing exists.

'Even in organic life'. That says it all really. You actually believe that organic life is superior to technology - not just our current technology, all technology. I believe we call that 'biowank' around here.

Organic life doesn't possess a universal constructor because not only is evolution incapable of making such a thing, it is not even 'motivated' to do so, in that there is no selection pressure to develop it. Life on earth settled on DNA and protein chemistry because it was the first mechanism that blind search stumbled upon that was robust and flexible enough for the purpose of making self-reproducing chemical blobs, which had to do nothing more strenuous than float around in shallow temperate pools and replicate. Once the basic mechanism was in place and significant complexity began to accrue on top of it, it became effectively impossible to change (other than some minor details e.g. exactly which amino acids are used), because that would require multiple-point simultaneous novel alterations that evolution is simply incapable of, even with sexual recombination. Intelligent designers can refactor, and can usually change out implementation layers without throwing away everything above them.

A true universal constructor in the sense of being able to make anything an established industrial society can make, in a compact mobile package, would require quite sophisticated nanotechnology. However this capability is far in excess of that required by a VNM. We can accept quite stringent limitations on materials and construction methods, and still have vastly more options than a single evolved (or even genetically engineered) organism can.

We would only be able to construct such a thing because we use our tools on resources produced by our extensive infrastructure. The kind of machine you're fancying must be a mini-civilization onto itself, or a seed to a technologically advanced civiliation. The former case is a very, very Hard problem.

Correct, but it is almost certainly design problem. Drexler and his successors have fairly conclusively proved that there is no physical reason why nanoscale processes of relatively low energy cannot achieve everything we can with bulk processes, in a fully programmed way - energy efficiency may be lower or higher than the bulk equivalent depending on the situation.

The later case is what you assume when postulating a VNM designed by a civilisation lacking mature nanotechnology (or in all likelihood, general AI). I would note that the actual variety faced by such designs is not that great. Typically we are talking about extracting material from asteroids, which come in a few basic compositions, and small planets and moons, with trace atmospheres and weak gravity wells. The only major variable is relative abundance of elements.

In the latter case, you're enslaving a race of intelligent creatures to your own end.

'Slavery' inherently implies unwilling service. If someone voluntarily does something for you, we would not call it 'slavery' - we don't call it 'slavery' when we domesticate animals either. Thus non-sentient VNMs aren't slavery, and the only way to make sentient ones loyal is to design them so they want to help you, so those aren't slaves either. The debate on 'enslavement' of sentient AIs is actually rather more complex than that, in that there are some kinds of mind which you might reasonably say have 'free will' and a human-comparable self-image, and others that don't, and the later can act as if sentient without actually being sentient in the morally relevant sense. Or at least, I am on the side of the debate that believes that to be the case, it is still an open issue, but it is beyond the scope of this thread.

In practice a society advanced enough to build von Neumann machines will almost certainly be able to equip them with a decent level of AI,

Why?

I confess that this is a personal judgement on likely technological development paths, but IMHO AI is a considerably easier challenge than all the combined physical issues of interstellar VNMs. The level of AI required to recognise interesting self-organising structures (and we don't care if the probes err on the side of caution when detecting life, in fact we very much prefer it) is substantially lower than sapient and probably not that much higher than the programming required to handle the usual contingencies of interstellar flight and finding suitable rendevous targets (to being self-replication). As such on average civilisations are unlikely to be able to produce the physical self-replicators while lacking a decent level of controlling intelligence.

A true self-replicator would have to be about the order of macromolecules to manipulate matter with enough precision to be a true self-replicator.

You appear to be using a completely arbitrary definition of 'true'. Do you mean 'a robot that can make copies of itself and also any artifact or material a highly technological civilisation could make, unassisted?'. That's nice, but we've already established that it isn't actually necessary, the minimum requirement is just a development path from the initial probe to a large-scale infrastructure capable of making and launching more probes (and optionally, doing other things, e.g. preparing a planet for colonisation), that works in at least a good fraction of the solar systems that probes end up in. This is much easier and probably possible for civilisations lacking molecular technology.

You can't fit very much intelligence in even a million atoms.

How is that relevant? Even the smallest self-replicating units in organic life have about a trillion atoms - human cells have a couple of orders of magnitude more. A ribosome doesn't need to have the genetic code incorporated into it; cells run many thousands of ribosomes from one nucleus. That fan-out ratio is limited by the horribly slow and unreliable mechanism of information transport; letting mRNA diffuse around the cell. Sensible designs for nanoassemblers have large numbers of active mechanisms electrically or mechanically connected to processors, which network to handle larger problems. Even designs where the individual assemblers move freely (which is considerably harder and probably unnecessary for nearly all practical applications) put a small amount of CPU and memory on each assembler and datalink new instructions to them en mass for each stage.

It sounds like you have been arguing against some woefully misinformed grey goo proponent using the Hollywood model of nanomachines, e.g. magic macromolecules that are simultaneously general assemblers, highly efficient energy producers (photoconversion, general oxidisers, if they even bother to specify a method of powering them at all) and at least insect-level intelligence (networking into an inevitably-evil sapient AI technically optional but very popular). The real nanotechnology community is not attempting anything so ridiculous.

And where are you going to get the fuel and the oxidizer for the rockets?

Electrolysis is not hard, but it's hardly relevant, realistic probes would use ion drives at minimum for everything outside low orbit. I was merely pointed out, once again, that what you seem to be proposing as an upper limit, or at least reasonable average, is in fact an extreme lower limit (uselessly so in fact).

Also, 10x Voyager speed is not a substantial fraction of c. No fast exploration for you!

For the purposes of the Drake equation it could take a billion years to cross the galaxy (presumably in numerous ten thousand year hops) and the basic question would still be there.

Asteroids are significantly lacking in that geology.

Uranium-rich asteroids are rare, but they can be surveyed relatively easily (fire tiny slugs at them, use spectroscope on resulting flash) and mined at low energy cost (again, assuming you're in no particular rush). Even for this edge case of a civilisation trying to make a VNM while lacking even mature fusion technology, it is only going to slow things down a little, because fissionables are only essential for powering follow-on probes. That said such technology is going to be limited to operating in inner solar systems, which is a significant though not crippling issue.

Why is a 50 kiloton spacecraft sufficient to carry a decent von Neumann probe? You assume that a real von Neumann probe with a sufficiently adequate universal constructor can fit in the payload. You will justify this assumption now.

Sure.

e.g. the Project Daedalus study (halving the top speed to allow for decceleration to relative rest). That was somewhat ambitious, but you could get nearly half of that just using a nuclear pulse drive based on late 50s technology (as detailed in the Project Orion design study). Both of these are quite within the capabilities of civilisations like ours, if the political structure made interstellar missions a priority - they would be trivial for a planet covered in a robotic civilisation built out of a von Neumann seed.

All of which require refined materials and fabricated parts — to wit, an exant civilization — to build. That means that there's only going to be one of these spacecraft no matter how many times your von Neumann machine replicates, unless it's able to refine materials and fabricate parts just like a civilization could. Furthermore, unless these machines can replicate the bombs/fuel pellets for the rocket, it's going to be good for only one trip. In order to take advantage of the VNMs' advantages, the spacecraft and its fuel must also be within the machines' manufacturing capabilities.

Given that life is made of the most common elements in the universe, it's obvious that, for creating a VNM to be used in unpredictable environments like remote alien planets, the low-hanging fruit is an artificial life-form.

It is not 'obvious' at all. If you mean a genetically engineered life form, it can only manage mere survival in narrow earth-like temperature ranges, with protection from vacuum boil-off and either sunlight or a pre-existing chemical fuel compatible with known biochemistry. That's just reproducing on a planet. How is covering a planet in genetically engineered bacteria going to produce and launch new interstellar craft?

If you mean something else by 'artificial life form', you're going to have to be more specific.

Anything with enough AI to function as a reasonably general replicator can easily be programmed to scan for evidence of life and avoid colonising such planets. Appropriate behaviour in this situation would be to unfold a big antenna and radio or laser news back home, then possibly consuming some asteroids to build orbital sensor platforms and/or robotic lander probes.

Again, you theorize a universal replicator with no limits on raw materials and end products.

What? How is replication generality related to having the sensors, processing and comms required to perform a normal interstellar probe mission? Detecting life and radioing home about it is completely unrelated. The VNM aspect is simply to remove the need for a) the original planet to build and launch millions or billions of probes and b) incredibly long flight times for direct transgalactic missions.

Also, you're assuming that the AI can recognize life in any form. That to me suggests some form of imagination, given that 'life' itself is an extremely slippery concept — it's hard to find a definition that is general enough to include any potential life form we've never seen without including things that aren't life that we know about.

Including non-life in the definition is no problem. It doesn't matter if the VNMs have even a large fraction of false positives. They could self-rep in only 10% of the available systems and still spread quickly. You're likely to lose far more systems to lacking appropriate resources than to having things that might be life. Even when there is possible life, in most cases it is a highly localised phenomenon that does not preclude use of the rest of the system for self-replication.

What's to prevent a mutant VNM from killing the embryoes and simply forming the Cluster, a civilization by robots, for robots, and will crush the puny fleshlings?

Ah of course, you can't imagine anything that isn't chained by the horrible limitations of organic life can you?

Do you live in mortal fear of your Windows computer 'mutating' into an evil AI that drains your bank account to pay for server hardware (for its inevitable attempt to usurp humanity)?. I imagine you must.

Back in reality, software does not 'mutate'. If the storage medium is damaged beyond the capability of the error correction to fix, the system fails checksum and shuts down. Even if it somehow lacked such basic safety systems, it would almost certainly just crash. There are real debates regarding 'mutation' of self-modifying AIs, but that is entirely unrelated to bit-level errors or biological mutation, and it is not relevant to VNMs based on non-sentient control software.

[Starglider]

I didn't want to get into Junghalli's argument as well as my own but this bit was just too hilarious to pass up.

Our own sapience is a bolt-on to a package of more primitive instincts designed to keep us intact and successful in an uncertain environment. A truly independent machine will have something like them, with sapeince as a bolt-on.

How did you put it, ah yes, 'You will justify this assumption now'.

[Samuel]

It is quite possible that we'll meet aliens that we can understand, and yet are totally different to us. I remember reading a short story about humanity meeting an insect-like alien species that gave birth to vast amounts of children, but ate 9/10 of them due to a lack of enough food to feed them all in their prehistoric times. Their culture considered baby-eating the most virtuous thing one could do. I doubt any human could draw a connection, however tenuous, between them, and any culture that's ever existed.

Why did they consider it virtuous? I consider it a sign of poor planning and waste and that is leaving out the infanticide.

You need a lot of factors and luck to get beyond just the hunter gatherer stage, such as geographical factors.

There were 6 (?) places on Earth that came up with it independentantly.

and animals that can be domesticated, both are exceptionally rare in the animal- and plant kingdom.

Animals aren't required- they are helpful. Plants aren't rare, just good plants.

And I am just scratching the surface here of all the factors that come into play to get an advanced civilization running.

You are misrepresenting. It was about why Europe got advanced first. None of this stops the other places from doing the same- in fact he thought that Australia could have developed agriculture on its own with a couple thousand more years or that the Pacific could develop Imperial states with more time.

This is about having it happen at all.

[Narkis]

It is quite possible that we'll meet aliens that we can understand, and yet are totally different to us. I remember reading a short story about humanity meeting an insect-like alien species that gave birth to vast amounts of children, but ate 9/10 of them due to a lack of enough food to feed them all in their prehistoric times. Their culture considered baby-eating the most virtuous thing one could do. I doubt any human could draw a connection, however tenuous, between them, and any culture that's ever existed.

Why did they consider it virtuous? I consider it a sign of poor planning and waste and that is leaving out the infanticide.

Agreed. It's been a few months since I read the story, but I think the aliens regarded it as the biggest sacrifice a parent could make for the Greater Good in the old times, and after food shortages were a thing of the past it remained as a cultural artifact that was elevated to a central tenet when a faction that wanted it abolished was killed and eaten by the traditionalists. Those aliens, of course, were portrayed as more stupid than humans. And bizarrely, the humans in the ship that made first contact mainly questioned the morality of the act, and not its stupidity.

[Samuel]

but I think the aliens regarded it as the biggest sacrifice a parent could make for the Greater Good in the old times,

How is it a sacrifice on the parents part? If the whole community drew lots it would make more sense, but even then it isn't a sacrifice.

elevated to a central tenet when a faction that wanted it abolished was killed and eaten by the traditionalists. Those aliens, of course, were portrayed as more stupid than humans.

No kidding. Killing and eating your political opponents... lets just say the situation is more unstable than a tyranny.

In humans. And bizarrely, the humans in the ship that made first contact mainly questioned the morality of the act, and not its stupidity.

Well, that is because they are normal people. Still, you'd think they would conclude if something was a planet wide cultural practice, there would be a good reason for it.

[Simon_Jester]

It is quite possible that we'll meet aliens that we can understand, and yet are totally different to us. I remember reading a short story about humanity meeting an insect-like alien species that gave birth to vast amounts of children, but ate 9/10 of them due to a lack of enough food to feed them all in their prehistoric times. Their culture considered baby-eating the most virtuous thing one could do. I doubt any human could draw a connection, however tenuous, between them, and any culture that's ever existed.

You'd still expect to see occasional moments of "Oh, hey, that looks sort of like a parliamentary democracy they've got there!" or "Oh, hey, they write fiction with a fixed number of light bursts per phrase! Huh. Sort of like a sonnet, or haiku."

I fully agree that comprehensible aliens won't just be "[insert historical culture here] IN SPACE!" But I think that almost any alien we encounter will have at least a few points of similarity that we will notice if we're not overwhelmed by the points of difference.

You should read Guns Germs And Steel by Jared Diamond which explains in detail why some people developed advanced technological civilizations and others remained at more primitive levels. It's a really good book, it comprehensively goes into the various reasons and factors.

You need a lot of factors and luck to get beyond just the hunter gatherer stage, such as geographical factors. Having access to steel and iron ore and such is secondary to having the right kind of fauna, suitable native plants and animals that can be domesticated, both are exceptionally rare in the animal- and plant kingdom. And it's very rare for any civilization to invent all the stuff on it's own, a lot is received by neighbors, herein natural obstacles and climate also factor in.

Yeah, but if you wait long enough, a lot of those factors can be compensated for- for instance, Mesoamerica was putting together civilizations that looked quite a bit like late Neolithic or early Bronze Age Eurasia by the time the conquistadors showed up. They were thousands of years behind schedule, as it were, but they were still inventing and building stuff.

So while not having the right confluence of geographical and biological factors in play may set back the growth of civilization, it's not obvious that it will stop growth entirely. Certainly not on million-year time scales.

but I think the aliens regarded it as the biggest sacrifice a parent could make for the Greater Good in the old times,

How is it a sacrifice on the parents part? If the whole community drew lots it would make more sense, but even then it isn't a sacrifice.

Answer One: They love their kids and don't want them to die, but kill them anyway.
Answer Two: They're all crazy.
Answer Three: Alien minds are alien.

Take your pick.

[Narkis]

It is quite possible that we'll meet aliens that we can understand, and yet are totally different to us. I remember reading a short story about humanity meeting an insect-like alien species that gave birth to vast amounts of children, but ate 9/10 of them due to a lack of enough food to feed them all in their prehistoric times. Their culture considered baby-eating the most virtuous thing one could do. I doubt any human could draw a connection, however tenuous, between them, and any culture that's ever existed.

You'd still expect to see occasional moments of "Oh, hey, that looks sort of like a parliamentary democracy they've got there!" or "Oh, hey, they write fiction with a fixed number of light bursts per phrase! Huh. Sort of like a sonnet, or haiku."

I fully agree that comprehensible aliens won't just be "[insert historical culture here] IN SPACE!" But I think that almost any alien we encounter will have at least a few points of similarity that we will notice if we're not overwhelmed by the points of difference.

Oh yes, I agree completely. Even in my example, the humans commented that the aliens had poetry, stories of valorous mythical heroes, etc. One of the scientists even said that the aliens' practice reminded him of human birth control, only post-birth. The proposed diplomatic approach was to convince them that we committed pre-birth baby eating in a scale far greater than they ever imagined.

elevated to a central tenet when a faction that wanted it abolished was killed and eaten by the traditionalists. Those aliens, of course, were portrayed as more stupid than humans.

No kidding. Killing and eating your political opponents... lets just say the situation is more unstable than a tyranny.

The infodump that the alien ship sent to the humans portrayed their civilization as a big happy family, but it was doubted as an accurate portrayal even within the story. One of the more dominant ideas on what to do with them among the ship's crew was to invade, kick their asses, find those who are dissenting in secret, install them as leaders, and heavily encourage the abolition of the practice.

but I think the aliens regarded it as the biggest sacrifice a parent could make for the Greater Good in the old times,

How is it a sacrifice on the parents part? If the whole community drew lots it would make more sense, but even then it isn't a sacrifice.

Answer One: They love their kids and don't want them to die, but kill them anyway.
Answer Two: They're all crazy.
Answer Three: Alien minds are alien.

Take your pick.

It was number one actually. The alien infodump contained stories about loving parents that weren't virtuous enough to kill their children, and who were killed and eaten when inevitably discovered. Also, I fail to see a difference between options two and three.

[Intio]

1) Since this has been alluded to already, I'll get this out of the way. There is nothing to suggest evolution 'rewards' self-awareness. I know most peope here are aware of this but some comments seem, to me, to come dangerously close to assuming that evolution is a ladder which leads to sentience - as we would define that. It may be that self-aware species in most of the galaxy don't survive their own inventions.

The unusual thing about humanity is not that we are intelligent but that we are industrialized.

2) This touches on my proposed idea. Let me state from the start that I have no deep understanding of geology or planetary formation. Anyone who does, please correct any mistakes. What if Earth is freakishly abundant in minerals? What if the vast majority of other planets do not have the materials required for EM-generating technology? The galactic norm may be societies which are locked into a medieval state of living, and humans (for all our faults) are the developed ones. How many synthetic EM signatures did humanity produce whilst we were tilling the fields with handmade equipment? Without the materials needed, and the desire to use them (although natural curiosity may be enough) there would be nothing in this 'teeming galaxy' which produced the evidence we are looking out for. (I know there are more recent detection developments which do not depend on detecting generated signals from other planets or artifacts)



As a side note, and more directly referring to the quote, I've read one account of the industrial revolution in Europe that posited the "necessity is the mother of invention" idea. The examples given were Africa and China. China dominated enough land and neighbours to the point that there was no pressing need to develop full industrialisation. Africa was given as an example at the other end of the scale where empires may have risen and fallen, but tribes tended to become an established norm, thus becoming too disparate to develop the societal confluences needed for industrialisation.

Europe was then analysed through this prism, and believed to have the "just right" balance of nation-states (non-disparate) and competition (other nations) packed closely together enough that industrialisation became an obvious way to develop and compete. I have my own issues with this, admittedly simplified, view of things, but I though I'd bring it up.

[Narkis]

2) This touches on my proposed idea. Let me state from the start that I have no deep understanding of geology or planetary formation. Anyone who does, please correct any mistakes. What if Earth is freakishly abundant in minerals? What if the vast majority of other planets do not have the materials required for EM-generating technology? The galactic norm may be societies which are locked into a medieval state of living, and humans (for all our faults) are the developed ones. How many synthetic EM signatures did humanity produce whilst we were tilling the fields with handmade equipment? Without the materials needed, and the desire to use them (although natural curiosity may be enough) there would be nothing in this 'teeming galaxy' which produced the evidence we are looking out for. (I know there are more recent detection developments which do not depend on detecting generated signals from other planets or artifacts)

*cough*Google is your friend*cough*

In short, the four inner planets all share roughly the same composition. If we extrapolate from that, it is reasonable to expect up to 40% of the galaxy's planets being having a similar mineral abundance. A data sample of one solar system is pretty much worthless for any meaningful statistics though.

[Intio]

Yes, but I was thinking more along the lines of the minerals needed for specific implementations of technologies, and the occurence of these minerals at points where they could be harnessed. Not just the prescence of the minerals within the confines of the planets themselves.

Although I accept that these dispositions might well be similar to our own planet.

In addition, moons tend to be very different, and it is thought by some that life would be more likely to occur on the moon of an extra solar planet rather than an actual planet (I seem to remember a reference to the diversity of contitions which can exist on moons), this might be another way in which we could be unusual. Personally, I'm not so sure, and I suspect that the moon idea was referring to any form of life rather than intelligent life.

[Simon_Jester]

Oh yes, I agree completely. Even in my example, the humans commented that the aliens had poetry, stories of valorous mythical heroes, etc. One of the scientists even said that the aliens' practice reminded him of human birth control, only post-birth. The proposed diplomatic approach was to convince them that we committed pre-birth baby eating in a scale far greater than they ever imagined.

Can't remember that; maybe it was in the discussion thread attached to the story?

How is it a sacrifice on the parents part? If the whole community drew lots it would make more sense, but even then it isn't a sacrifice.

Answer One: They love their kids and don't want them to die, but kill them anyway.
Answer Two: They're all crazy.
Answer Three: Alien minds are alien.
Take your pick.

It was number one actually. The alien infodump contained stories about loving parents that weren't virtuous enough to kill their children, and who were killed and eaten when inevitably discovered. Also, I fail to see a difference between options two and three.

Not quite. You see, I would argue that all three answers are true; it depends on your frame of reference. (1) is an attempt to understand their behavior in their own frame of reference, with a result that seems schizophrenic to our mind. (2) is an attempt to describe them in a human frame of reference, in which they really are crazy- humans who acted like that would be widely considered insane by other humans. (3) is an attempt to explain the situation in the general case- alien minds are alien to each other, and will often contain concepts that are mutually exclusive. Trying to make sense of Species X's behavior in terms of how Species Y would react to that behavior isn't going to work.
_________

2) This touches on my proposed idea. Let me state from the start that I have no deep understanding of geology or planetary formation. Anyone who does, please correct any mistakes. What if Earth is freakishly abundant in minerals? What if the vast majority of other planets do not have the materials required for EM-generating technology?

That seems a little unlikely. Simple stellar physics suggests that materials like iron and silicon will be abundant on rocky planets throughout the universe. Other chemical elements like copper might be more or less common, but it's unlikely (though possible) that Earth just happened to win the lottery on so many of the elements vital to technology.

Moreover, in a lot of cases substitutions can be used. While substituting Metal A for Metal B may not give us all the properties we'd like, it's usually at least possible.

It's definitely relevant that a planet locked in something like the medieval period won't be producing signals we can detect, but it seems unlikely that Earth is a freakish exception in having the resources it takes to escape the medieval period.

[Samuel]

Trying to make sense of Species X's behavior in terms of how Species Y would react to that behavior isn't going to work.

There are behaviors that are entirely insane, from any reference frame. For those we blame shitty writters. My favorite is an alien species that never bothered with the concept of medicine because of cultural prohibitions.

[Narkis]

Oh yes, I agree completely. Even in my example, the humans commented that the aliens had poetry, stories of valorous mythical heroes, etc. One of the scientists even said that the aliens' practice reminded him of human birth control, only post-birth. The proposed diplomatic approach was to convince them that we committed pre-birth baby eating in a scale far greater than they ever imagined.

Can't remember that; maybe it was in the discussion thread attached to the story?

You're right. I reread the story, and it was a comment in the discussion thread indeed. No one proposes a concrete diplomatic approach in the story. I should've checked it before giving my example, probably.

[Junghalli]

For those curious the story about the baby eating aliens can be found here. The deal is that they began culling their young when they were still pre-sapient; the need to do so became the foundation of their entire system of morality, to the point that their word for "good" literally translates as "to eat children".

You skipped an imporant step: Is any of these places suitable for a VNM to set up shop? You need a place where you have everything necessary to refine the raw resources into suitable parts and expendables. Most industries require some water (universal solvent), and if you're assuming that the VNM is made of mostly metal, then the best place to find the water is not going to be the best place to find the metals. You'd have to do some trucking around, which will take time and propellant.

This is completely impossible to answer without knowing the technological capabilities of the machine's builders. For instance, whether the machine needs heavy elements or whether it can be built almost entirely from carbon like organic life will make a big difference (the latter will probably allow exploitation of a greater variety of environments). So does whether it needs fissionables for power or whether it can simply use deuterium fusion. In the latter case it could easily obtain its fuel from water, ammonia, methane, or any other hydrogen-bearing compound (albeit with heavy filtration). In the former things will be much more difficult. I'm sure this line of thought can be continued for a long time.

The balance is not clear-cut here. If marginally useful systems are overwhelmingly common, the it may be worth it to replicate in a marginal system even if you can ruthlessly exploit an ideal system.

Again, this requires defining the technological capability of the builders, and what sort of solar systems are and are not common.

With what? Rainbows and unicorns? You have to supply energy to crack water — the fact that you get energy out of burning hydrogen and oxygen should tell you that. In the Kuiper belt, you would have to expend your own resources or gather it up from your surroundings. The Kuiper belt is not known for its heavy elements, so any energy gathered would have to be from the star. The far away star, where the solar radiation is so weak that it would take ridonkulously huge solar panels to even work the electronics.

I can think of several possibilities. The most obvious is that the machine could run off deuterium fusion, deuterium being readily obtainable from a wide variety of ices. Of course, this requires a certain minimum level of technological sophistication on the part of the builders (they must be able to make such a practical D-D fusion reactor). The next is that it could snatch a comet passing through the inner system, where sunlight is more available. The last is that it could set up shop in the Kuiper Belt, use solar power, and simply take a very, very long time to refuel (and replicate). If you are exploring the stars with chemical rockets you had better be very, very patient anyway.

Water in and of itself is useful as reaction mass, but that's it.

A fusion drive using deuterium or hydrogen, as I have said already, could easily be wilderness refueled. So could nuclear thermal rocket, although to be truly self-sustaining it would probably have to use fusion for power (getting fissionables from asteroids may require processing infeasible amounts of material). A Bussard ramjet/scramjet could, of course, be fueled indefinitely off the interstellar medium and would by virtue of design necessity have to be able to use extremely abundant simple hydrogen for both propellant and fuel. Finally, there is the possibility of a propulsion system that requires no fuel or propellant at all: solar sailing. An advanced solar sail making a close pass to the sun could be accelerated to some low fraction of c by the simple light pressure of the local star (this paper suggests a probable upper limit of .012 c), enough to allow for a slow but steady exploration of the galaxy with no worries about rocket fuel at all.

Given that, in the particular section you were responding to, I was after radioactives, iron is not going to do me much good. Even if I could use iron, I'll still need to refine it and fabricate parts, which will take energy.

Ah. According to this chart a typical C-type asteroid might contain perhaps 10 parts per billion of uranium, so while obtaining fissionables from asteroids might be possible it would require processing a lot of material. If you were looking for an engine for a self-replicating probe a fission reactor would be close to the bottom of the list of things I'd consider. Even a chemical rocket would probably be better, because hydrogen-oxygen rocket fuel would be much more easily obtainable.

But let's examine that for a minute. First off, your observation that nickel-iron nugets can be scooped out of the dirt indicates one of the problems with asteroids in terms of mining. The minerals are evenly distributed through the asteroid's mass, and the local concentrations of the minerals not that large (granules) and mixed with a large amount of impurity (dirt). In order to get at most of the nickel-iron granules, you'd have to dig through a great deal of the asteroid, and you have to put the waste somewhere.

PERMANENT has an extensive discussion on processing asteroidal metals here. I think I'll just quote it.

Asteroidal material in general is exceptionally good ore requiring a minimum of processing, since it has free metal already.

Only basic ore processing need occur at the asteroid*, producing free metal and volatiles (usually stored as ices), and perhaps selected minerals, glasses and ceramics. The required equipment is quite simple.

The following is a sample ore processing system, but is not the only one proposed to date.

At the input chute, the ore will be ground up and sieved into different sizes as the first step of a basic ore processing system. Most asteroids probably offer far more crumbly material than we could consume in one mining expedition.

Simple mechanical grinders, using a gentle rocking jaw arrangement for coarse crushing and a series of rollers for fine crushing, could be arranged in a slowly rotating housing to provide centrigufal movement of the material. Vibrating screens are used to sift the grains for directing them to the proper sized grinders.

The streams of material are put through magnetic fields to separate the nickel-iron metal granules from the silicate grains. Alternatively, the streams can be dropped onto magnetic drums, whereby the silicates and weakly magnetic material deflect off the drum whereas the magnetic granules and pebbles stick to the magnetic drum until the scrape off point. Repeated cycling through the magnetic field and perhaps additional grinders can give highly pure bags of free nickel iron metal.

An optional additional piece of equipment is an "impact grinder" or "centrifugal grinder" whereby a very rapidly spinning wheel accelerates the material down its spokes and flings it against an impact block. Any silicate impurities still attached to the free metal are shattered off. It's feasible to have drum speeds sufficient to flatten the metal granules by impact. A centrifugal grinder may be used after mechanical grinding and sieving, and before further magnetic separation. In fact, most of the shattered silicate will be small particles which could be sieved out.

The nonmagnetic material is channelled into a solar oven where the volatiles are cooked out. In zero gravity and windless space, the oven mirrors can be huge and made of aluminum foil. The gas stream is piped to tanks located in a cold shadow of space. The tanks are put in series so that the furthest one away is coldest. This way, water condenses more in the first one, carbon dioxide and other vapors in the tanks downstream.

*This part, of course, is not true for a self-replicating probe; it must do all the processing into final products itself.

As for disposing of the slag and waste material, PERMANENT suggests simply putting it in bags or barrels (the latter would be easily manufacturable from asteroidal iron) and storing it away. Alternately, you could load it into a mass driver and shoot it into the sky at whatever paltry speeds are necessary to escape the asteroid's tiny gravity well. I don't see why it would be a big problem.

Uh, yeah. You realise that the reason we have these traits is because they're useful in an unpredictable environment. A truly autonomous machine will have these traits too in some fashion.

It would probably need some form of survival programming, but there's no reason its own survival and the survival of its kind has to be its highest goal as it is with humans. A competent AI designer would make serving its mission its highest goal, and survival merely a means to that end ("if I die I can't fulfill my mission so I want to live").

Not reflecting on what one has already done is a danger to long term plans. A truly autonomous machine needs to be able to reflect on its past performance and take that into account as it proceeds towards its goals. It needs to be able to know what it's doing.

This assumes that a zombie (in the philosophical sense) cannot effectively self-monitor. I am personally skeptical of this proposition, but it's hard to say one way or another until we understand more about the neurological mechanisms that create the feeling of consciousness in humans.

The thought of creating such a being fills me with disgust and revulsion.

Meh, I'm personally somewhat ambivalent on it. I see your perspective, but on the other hand it's not like this entity would suffer from being what it is. Personally I would prefer to use a (philosophical) zombie AI if possible, but this is really getting into matters not particularly relevant to the discussion. The subject is whether a self-replicating probe would be practical, and there's absolutely no reason to imagine its hypothetical creators would have to share your disgust and revulsion at the idea of creating a sapient entity from the ground-up to be a slave, find being a slave natural, like being a slave, and have no inclination whatsoever to not be a slave. The only thing that matters is here is whether that's feasible, and I see absolutely no reason it wouldn't be.

That aside, mutants are a danger to us.

That is very true. Before entertaining the thought of a self-replicating probe I would want to be sure that the error-proofing systems on the replication are very, very good. As in they have a safety factor of orders of magnitude beyond the number of times the thing will actually be programmed to replicate.

[Wyrm]

A 'von Neumann machine' has a specific definition, and organic life fits that definition like a glove.

Semantically, no they do not, because they are not machines.

Your semantics whoring does not impress me. Von Neumann was not concerned how the machine was created in the first place, only in how an entity following mechanistic laws could reproduce itself without some vital force or infinite regression of smaller self-copies.

That makes bacteria useful for establishing absolute lower limits - in that anything evolution can make, intelligent designers can make at least as well.

The current track record of intelligent designers does not bear you out.

Furthermore while macroscale organic life exists all of its reproductive processes and nearly all of its resource processing operates at the molecular scale, so it can tell us nothing at all about how machines designed to employ macroscale replication will perform.

You realize that there are some forms of life that are macroscopic, do you not?

Finally organic life can of course tell us nothing about functioning in interplanetary space.

Irrelevant. In broad strokes, the problems are the same: resource aquisition and avoiding harm in an environment not specified ahead of time. A fixed configuration of matter behaves in a particular way — it will be able to do some things and not others. Magic constructors violate everything we know about every physical process we have so far encountered.

No, it is not. Unicellular organisms reproduce by fission. The reproductive mechanisms of multicellular organisms are developed from that; they start with a single cell which slowly grows into a new organism. It is required to maintain basic metabolic function throughout the entire process and is required to be a viable independent organism from birth onwards.

Mechanical self-replication is nothing like that - in fact it would take massive effort to emulate this grossly inefficient system, for no gain at all. Robotic self-replicators build full-size copies of themselves which are generally inert until completed and then fully functional upon activation. This includes mental capabilties; robots can trivially copy their mind-state, organisms cannot.

Irrelevant details. Von Neumann was trying to specify how a totally mechanistic self-replicator could be possible without some vital force or infinite regression of self-specification, and he realized that only three things were necessary: a "universal constructor", a regulatory mechanism, and a blueprint of both. You could make the "universal constructor" and regulatory mechanism from the blueprints, and then copy the blueprints.

Translating this to biological organisms, we find the blueprints (DNA), the "universal constructor" (ribosomes, protiens, and attendant metabolistic cycles) and the regulatory mechanism (interactions between the DNA and the protiens), exactly as von Neumann predicted. That organisms mash together von Neumann's discrete steps and parts is irrelevant. That they cannot trivially copy their mind state is irrelevant. Life displays all the required properties to be called a von Neumann machine. Period.

von Neumann's real achievement was creating a (basic) functional model showing how such a machine could actually operate.

Wrong. Von Neumann showed how pre-scientific vital force or sperm humonculi woo woo were unnecessary; he provided the grounding of life/self-replication as a mechanistic process.

Organic life actually faces far fewer problems, because evolution targets simple reproduction, not interstellar travel.

Organisms have already mastered the characteristic property of a von Neumann machine: self-replication in an unpredictable environment.

However this is irrelevant, because the characteristics of the solution are not solely dependent on the shape of the problem, they are the product of that and the tools available to solve it, and the toolset for self-replicating robots is entirely different (and vastly more capable).

Why do you assume this? You can only pack so many tools with your VNM. Even if you include the tools as programs, you still have to have the tools to make the tools.

How is that in any way a response to my point? 'A constructor', as in a single physical organism or machine, is optimised. Robots can build completely dissimilar robots and even self-modify as required. Organisms cannot.

Fell asleep during biology class, did you? You developed from an undifferentiated ball of completely identical cells. A Trypanosoma modifies its own genetic program to put a different coat on itself every two weeks for decades. Parasites have some of the most fiendishly complicated life cycles on the planet. The kind of stuff life does is easily as complicated as anything we've done.

You refuse to recognise this basic distinction.

I refuse to recognize a distiction you fail to show exists.

You are assuming all technology is exactly like bacteria, a ludicrous and transparently false position.

Strawman. How is looking at life and realizing that an autonomous technology will have to overcome similar challenges "assuming all technology is exactly like bacteria"?

We already have industrial robots that build other robots.

But not themselves. Or anything outside of a narrow range of capabilities.

They are completely dissimilar models. Being able to construct an exact copy is actually a rare case; occasionally useful, but most of the time it is more efficient to have a small number of robots dedicated to building more robots with the other robots all optimised for different tasks.

A VNM has to eventually get around to building itself, moron. It's part of the definition. This capability has yet to be demonstrated in anything we have built.

Even organic life, limited as it is, hit on a very limited version of this with some insects; worker ants form the vast majority of the colony, but do not reproduce. Most mammals have a very limited degree of task specialisation keyed off the gender distinction that evolved to support sexual reproduction. Of course with the horrid limitations of available selection pressure, chemical cascade expression mechanisms (for differentation during growth) and unstable DNA as the information storage medium, this differentation could not proceed very far.

More proof that you fell asleep during biology class. Ants and mammals are multicellular organisms. Each one is a colony of self-replicators that specialize themselves to a particular task, sacrificing their further reproducion for the good of the greater organism it's part of. What you fantisize your wonderful VNM to do, organisms already have perfected.

In short, they are not 'fanciful' capabilities, they are not only already demonstrated in industrial settings, they are how any sane engineer would expect robots to work.

But not without the input of refined materials at one end, nor are there any examples able to reproduce themselves, nor work without the support of a civilization's infrastructure and their ultimate creator's intervention.

In a strange, uncontrolled, unpredictable environment without intelligent support, who has solved the problem of gathering resources and replicating themselves? The answer is LIFE!

Again, I have an exant model, while you have a hyperbolic fantasy.

Again you are acting as if the limitations of organic life were relevant, ignoring even the (relatively primitive) computer you are using to access this BBS. DNA is reasonably compact at the bit level (only a few of orders of magnitude off optimal), but it is horribly redundant at both the message level (lots of junk DNA, lots of duplicate copies of proteins, no data compression) and the organism level (humans have 100 trillion copies of their own blueprints... for no good reason at all). A robot of equivalent size and complexity to a human, storing a conservative 100 copies of its own blueprints with lossless compression in moderately efficient optical storage, would use approximately one quintillionth the mass and volume that a human uses for the same purposes. You could include a million different robot designs for different purposes and environments, and that's still a trillionth of the mass a human devotes to DNA.

All that text and you still haven't answered the damn question: HOW MUCH DO YOU FUCKING NEED!!

Construction limitations are more relevant, but not much. The mechanical side is straightforward; the same basic manipulators and bending/cutting/joining methods are used for the vast majority of machines.

That's ridiculously simplistic. Even without going to chemical processing, the number of ways to manipulate material is enormous. There are literally hundreds of ways to cast metals and ceramics, each raw material requiring specialized equipment and supplies. Even a seemingly simple process of bending has hundreds of specializations, such as stamping, or rolling, again, each raw material requiring different equipment.

The world we live in right now is the product of all those different manufacturing techniques, designed to control the quality of the goods quite finely. If you limit yourself to only three ways to manipulate any given material (or worse, three ways for all materials), you are going to have some serious engineering issues on your hands, and your stored blueprints are going to be quite a bit more complicated than you anticipate.

Molecular nanotechnology would use a similar bootstrap process, only with more steps and higher efficiency since it can use a much wider range of chemistry and techniques such as channeled transport (don't have to do everything in the same flexible bag of water).

Explain "channeled transport". And a more varied chemistry needs varied reagents. What happens if some reagents aren't present at your destination?

However molecular nanotechnology is not a prerequisite for VNM, because the same principle works at the macroscale; building the tools to build the tools to achieve the goal. It's less efficient, to the extent that your probe may well become a multi-thousand tonne automated factory rather than the tiny probe you could produce with nanotechnology, but it is certainly workable.

I think you sorely underestimate the size of the factory you need. The modern manufacturing infrastructure we have now is hideously complicated, with many, many interdependencies. Take apart your computer sometime and think about how each of the little parts was fabricated and assembled and you'll begin to grasp the scope of the problem — and this is just for a relatively primitive computer. Don't expect it to be any easier with a sophisticated VNM.

While these are not easy problems, they are well within the scope of existing engineering skill, both as extensions of techniques we already use on contemporary spacecraft, and the numerous proposals you can find in both academia and the sci-fi community. They also apply to all interstellar craft, not just VNMs.

Fair enough.

Also, you're imagining a universal constructor that could construct absolutely anything, when even in organic life, no such damn thing exists.

'Even in organic life'. That says it all really. You actually believe that organic life is superior to technology - not just our current technology, all technology. I believe we call that 'biowank' around here.

It's not 'biowank' to point to the exant capabilities of organic life — the fact that it can create a wide variety of chemicals that would make a chemist green with envy — and to point out that even life has its limits. The point is that every material we've created requires specialized equipment to both create and manipulate. There is no such damn thing as a "universal constructor." Life is the closest we know of, and even it's not universal.

<snip>We can accept quite stringent limitations on materials and construction methods, and still have vastly more options than a single evolved (or even genetically engineered) organism can.

You have more options, but they do not come without cost. You must carry the tools with you, or the tools to create the tools from the surrounding materials of unknown composition and unpredictable quality.

Correct, but it is almost certainly design problem. Drexler and his successors have fairly conclusively proved that there is no physical reason why nanoscale processes of relatively low energy cannot achieve everything we can with bulk processes, in a fully programmed way - energy efficiency may be lower or higher than the bulk equivalent depending on the situation.

The later case is what you assume when postulating a VNM designed by a civilisation lacking mature nanotechnology (or in all likelihood, general AI).

I cannot extrapolate from a vacuum. I suggest you refrain from doing the same.

I would note that the actual variety faced by such designs is not that great. Typically we are talking about extracting material from asteroids, which come in a few basic compositions, and small planets and moons, with trace atmospheres and weak gravity wells. The only major variable is relative abundance of elements.

All the statements about the richness of asteroids assume a space infrastructure designed to take advantage of them. A probe coming in from a distant star will not have this infrastructure. It will need, in broad strokes, a refinery for every single processed material it needs. Such a probe will require hundreds of distinct materials... you begin to see the problem now?

'Slavery' inherently implies unwilling service. If someone voluntarily does something for you, we would not call it 'slavery' - we don't call it 'slavery' when we domesticate animals either. Thus non-sentient VNMs aren't slavery, and the only way to make sentient ones loyal is to design them so they want to help you, so those aren't slaves either.

'Willing' and 'unwilling' implies choice. If you've short-circuited the choice for the subservient sapeint AI to choose between servitude and non-servitude, how can it be a willing choice?

The debate on 'enslavement' of sentient AIs is actually rather more complex than that, in that there are some kinds of mind which you might reasonably say have 'free will' and a human-comparable self-image, and others that don't, and the later can act as if sentient without actually being sentient in the morally relevant sense. Or at least, I am on the side of the debate that believes that to be the case, it is still an open issue, but it is beyond the scope of this thread.

Of course. Let's leave it at that.

You appear to be using a completely arbitrary definition of 'true'. Do you mean 'a robot that can make copies of itself and also any artifact or material a highly technological civilisation could make, unassisted?'. That's nice, but we've already established that it isn't actually necessary, the minimum requirement is just a development path from the initial probe to a large-scale infrastructure capable of making and launching more probes (and optionally, doing other things, e.g. preparing a planet for colonisation), that works in at least a good fraction of the solar systems that probes end up in. This is much easier and probably possible for civilisations lacking molecular technology.

You seem to strap your assertions of full AIs on the back of nanotech. By definition, nanotechnology requires manipulators on the order of macromolocules. It's not unreasonable that the most common element in your scheme will be on the same order.

You can't fit very much intelligence in even a million atoms.

How is that relevant? Even the smallest self-replicating units in organic life have about a trillion atoms - human cells have a couple of orders of magnitude more. A ribosome doesn't need to have the genetic code incorporated into it; cells run many thousands of ribosomes from one nucleus. That fan-out ratio is limited by the horribly slow and unreliable mechanism of information transport; letting mRNA diffuse around the cell. Sensible designs for nanoassemblers have large numbers of active mechanisms electrically or mechanically connected to processors, which network to handle larger problems.

How? Electricity behaves very differently on the nanoscale. We can today build devices that work by electrons tunneling through insulators, and they are not nanotech devices! On the nanoscale, physical linkages behave more like rubber than metal rods. Thermal noise is going to be rife in your device, wrecking all sorts of fun havoc.

Even designs where the individual assemblers move freely (which is considerably harder and probably unnecessary for nearly all practical applications) put a small amount of CPU and memory on each assembler and datalink new instructions to them en mass for each stage.

And how are they supposed to communicate? Telegraph? Semaphore flags?

It sounds like you have been arguing against some woefully misinformed grey goo proponent using the Hollywood model of nanomachines, e.g. magic macromolecules that are simultaneously general assemblers, highly efficient energy producers (photoconversion, general oxidisers, if they even bother to specify a method of powering them at all) and at least insect-level intelligence (networking into an inevitably-evil sapient AI technically optional but very popular). The real nanotechnology community is not attempting anything so ridiculous.

Yet it's exactly this kind of magic you require from your nanotech. You don't specify what processes you propose to extract, refine, and fabricate nanoparts, instead handwaving over the entire problem and mumbling 'um, nanotech,' and hoping I won't notice. You ignore the very real problem of energy consumption of your nanodevices, instead handwaving over the entire problem and mumbling 'um, nanotech,' and hoping I won't notice. You ignore the very real problem of coordinating all these nanodevices, with each other or to a home AI, instead handwaving over the problem and mumbling 'um, nanotech,' and hoping I won't notice. I don't see a fucking difference between you and 'some woefully misinformed grey goo proponent.'

Electrolysis is not hard, but it's hardly relevant, realistic probes would use ion drives at minimum for everything outside low orbit.

Okay, where are you going to get the xenon or cadmium for these ion drives?

I was merely pointed out, once again, that what you seem to be proposing as an upper limit, or at least reasonable average, is in fact an extreme lower limit (uselessly so in fact).

Extreme lower limit of what may be capable of a technologically advanced civilization, you mean. We don't even know if Daedalus would actually work, and as far as Orion goes, they require a stockpile of atomic bombs — which are hard to create and require a large quantity of the almost vanishingly rare fissionable isotopes. Unless you're proposing that the VNM construct the equivalent infrastructure of a full-fledged civilization, it's going to have to make some compromises. At least hydrogen and oxygen need no special processing.

For the purposes of the Drake equation it could take a billion years to cross the galaxy (presumably in numerous ten thousand year hops) and the basic question would still be there.

Ah, we're finally coming around the the crux of the matter, the Fermi Paradox. There are only two ways out of the Fermi Paradox:

(a) Intelligent life is rare, or
(b) Distance protects us from other intelligent life.

Actually, it would be a combintation of (a) and (b). Intelligent life is rare enough such that the expected distance to the next civilization protects us from them.

Von Neumann probes (VNPs) able to construct fast STL drives destroy (b) if they really do have the capabilities you claim. That must mean that the galaxy is barren of technically advanced life, or the requisite VNPs and/or fast STL drives are impossible. At least one must be missing! Choose.

Uranium-rich asteroids are rare, but they can be surveyed relatively easily (fire tiny slugs at them, use spectroscope on resulting flash) and mined at low energy cost (again, assuming you're in no particular rush).

Why can we assume that you're in no particular rush? Every part of your VNM is going to wear out sooner or later. Solar panels have a limited life. Reactors have a limited life, too. Your micrometeroid shield will have a limited life. Your chemical extractor and reagents will have a limited life. Your cooling system will have a limited life. Your days are numbered.

Even for this edge case of a civilisation trying to make a VNM while lacking even mature fusion technology, it is only going to slow things down a little, because fissionables are only essential for powering follow-on probes. That said such technology is going to be limited to operating in inner solar systems, which is a significant though not crippling issue.

It can be, if you can't reproduce/replace fast enough to stave off the inevidable encroachment of your eventual mechanical failure.

Sure.

A careful reading of your source shows it to be long on promises and short on details. For instance, the basic 'chemist' component of the reproduction is very much lacking on the expected impurity of the extracted elements. The impurities will degrade the quality of the material and therefore the performance of the machines and secondary materials fabricated from it. Aerostats are another bit of fuzzy math, as they are simply assumed to have half the practical lifetime of an industrial factory, even though an aerostat would be operating under very different conditions from an industrial factory. You're going to have to do better than that.

Given that life is made of the most common elements in the universe, it's obvious that, for creating a VNM to be used in unpredictable environments like remote alien planets, the low-hanging fruit is an artificial life-form.

It is not 'obvious' at all. If you mean a genetically engineered life form, it can only manage mere survival in narrow earth-like temperature ranges, with protection from vacuum boil-off and either sunlight or a pre-existing chemical fuel compatible with known biochemistry. That's just reproducing on a planet. How is covering a planet in genetically engineered bacteria going to produce and launch new interstellar craft?

First off, you've obviously never heard of extremophiles, have you? We've found life munching on just about every energy source and living in every nook and cranny on the planet, including rocks and corrosive hot springs that would strip the skin off of you instantly. Tardigrades are the king of extremophiles, able to survive complete dessication and the space environment for a decade or more. If I was designing an organism from the ground up, I'd take a hint from these critters and design a whole spectrum of creatures able to survive a large variety of hostile environments.

Secondly, your disparaging of the narrow range of life assumes that machines don't live in equally narrow ranges. Your computer has a cooling fan precisely because it will malfunction if it heats up too much. Mechanical parts can suffer from vacuum welding, and so have to be specially designed for the space environment.

Finally, the organisms would build a space ship the same way your nanobots would: by constructing larger and larger structures to help with gathering and manipulating resources. As for programming, if I'm building the organism(s) from the ground up, I can eliminate all the junk and code the instructions tightly, and genomes can code for a lot. The lowly Amoeba has 290 billion base pairs to play with, and if you work with communities of organisms, you can work with a library as sizable as any VNM you can build.

This is what I meant by the low-hanging fruit. If you're going to be building a machine, your best bet is to be building it out the most common elements of the universe. Once a community is established, you can then take your time to build the tools it takes to finally build a spacecraft. That life forms are not called "machines" is merely semantics.

Anything with enough AI to function as a reasonably general replicator can easily be programmed to scan for evidence of life and avoid colonising such planets. Appropriate behaviour in this situation would be to unfold a big antenna and radio or laser news back home, then possibly consuming some asteroids to build orbital sensor platforms and/or robotic lander probes.

Again, you theorize a universal replicator with no limits on raw materials and end products.

What? How is replication generality related to having the sensors, processing and comms required to perform a normal interstellar probe mission?

"Anything with enough AI to function as a reasonably general replicator". You've inexorably linked your AI to this "reasonably general replicator" of unspecified performance in your argument. If you don't want that, don't do it.

Including non-life in the definition is no problem. It doesn't matter if the VNMs have even a large fraction of false positives. They could self-rep in only 10% of the available systems and still spread quickly. You're likely to lose far more systems to lacking appropriate resources than to having things that might be life. Even when there is possible life, in most cases it is a highly localised phenomenon that does not preclude use of the rest of the system for self-replication.

Fair enough.

Ah of course, you can't imagine anything that isn't chained by the horrible limitations of organic life can you?

Computer errors are a fact of life, son. Some caution in handling machines with the ability to multiply exponentially is warranted.

Do you live in mortal fear of your Windows computer 'mutating' into an evil AI that drains your bank account to pay for server hardware (for its inevitable attempt to usurp humanity)?. I imagine you must.

The worst my laptop could do to me is erase my porn, then self-destruct. A swarm of hostile VNMs could loop back to our planet and erase us. Some caution for the latter is warranted.

Back in reality, software does not 'mutate'. If the storage medium is damaged beyond the capability of the error correction to fix, the system fails checksum and shuts down. Even if it somehow lacked such basic safety systems, it would almost certainly just crash. There are real debates regarding 'mutation' of self-modifying AIs, but that is entirely unrelated to bit-level errors or biological mutation, and it is not relevant to VNMs based on non-sentient control software.

Modern software is not designed to change its code on the fly, or more generally, the instructions are closed and fixed. A sentient has to be able to learn or it doesn't deserve the title. That means being able to change its instructions on the fly. This is a fundamental difference between the way we program now and the way we will need to program AI's, so past art is a dubious guarantee. Some caution is warranted.

Our own sapience is a bolt-on to a package of more primitive instincts designed to keep us intact and successful in an uncertain environment. A truly independent machine will have something like them, with sapeince as a bolt-on.

How did you put it, ah yes, 'You will justify this assumption now'.

Extrapolation from a single datapoint, plus the fact that a sapience is useless if the hardware is wrecked. Yeah, you may cry 'single datapoint', but it's better than extrapolating from 'no datapoint.'

[Samuel]

Holy shit those aliens are stupid. First rule of first contact- NEVER assume aliens are just like you. EVER.

Never trade anything pertaining to something that might be taken badly by any alien race. This means nothing about war, slavery or killing in general. You want it implied that you have a military in such a way that it is obvious to another group that does, but innocuous to everyone else (aka uniforms and marching).

Is it too much to ask for competant anthropologists?

[Junghalli]

Yeah, but if you wait long enough, a lot of those factors can be compensated for- for instance, Mesoamerica was putting together civilizations that looked quite a bit like late Neolithic or early Bronze Age Eurasia by the time the conquistadors showed up. They were thousands of years behind schedule, as it were, but they were still inventing and building stuff.

So while not having the right confluence of geographical and biological factors in play may set back the growth of civilization, it's not obvious that it will stop growth entirely. Certainly not on million-year time scales.

Civilization on Earth may well have been running against the clock to develop high technology. Barring technological intervention sooner or later (probably within a few tens of thousands of years) another ice age is bound to come, and it's an open question whether a preindustrial civilization could survive such a dramatic climate change.

Most Earthlike worlds would experience very dramatic periodic climate shifts because they'd lack a large moon like ours to stabilize their axis. On a world like that civilization might appear only to be knocked back to the stone age with the next major climate shift again and again.

Okay, where are you going to get the xenon or cadmium for these ion drives?

You don't need xenon or cadmium, just use VASIMR which uses hydrogen for propellant, or a plasma thruster which can use oxygen. Both of those can be obtained easily from any icy body. If you absolutely must use an ion drive PERMANENT recommends sodium.

[Starglider]

Extrapolation from a single datapoint

Yes, that is your entire argument isn't it; "Self-replicating robots must be exactly like life because... because... because I'm too stupid to actually consider any of the real technical issues, so I am going to pretend they're irrelevant!"

plus the fact that a sapience is useless if the hardware is wrecked.

So? How is that relevant to anything?

Yeah, you may cry 'single datapoint', but it's better than extrapolating from 'no datapoint.'

Since you have proven yourself incapable of understanding functional arguments, I will simply point out that subsumption architectures (which is what your fumbling characterisation is actually grasping at) comprise a tiny fraction of current general AI research. There is Rodney Brooks (who pioneered the design) and his followers, Mark Tilden's BEAM analogue dead end and a handful of similarly patterned NN designs. Together they make up about 3% of the publication output and projects underway. There are other heirarchical control designs and most architectures are layered in some way, but not in a way that puts goal driven behaviour functionally 'below' the main intelligence.

So, by your own favored method of utterly mindless extrapolation, we can say that 3% of VNMs built will have 'sapience as a bolt on to an instinctive intelligence'. You chew that over while the rest of us cheerfully accept that there would be no reason to chose this design strategy when there are better alternatives available (even Brooks only did it mostly out of frustration with seemingly slow progress).

[Wyrm]

So does whether it needs fissionables for power or whether it can simply use deuterium fusion. In the latter case it could easily obtain its fuel from water, ammonia, methane, or any other hydrogen-bearing compound (albeit with heavy filtration). In the former things will be much more difficult.

The latter ain't no picnic either, due to the extreme difference in abundancies between hydrogen and deuterium, and the very marginal difference in their physical and chemical properties. Deuterium is a pain in the ass to extract on Earth, where natural processes have enriched deuterium to 115 ppm for normal water. In contrast, Jupiter has only 22 ppm deuterium.

I can think of several possibilities. The most obvious is that the machine could run off deuterium fusion, deuterium being readily obtainable from a wide variety of ices.

If by "readily obtainable", you mean on the order of 22 ppm with long and energy intensive processing to enrich due to the very subtle differences between their physical and chemical properties. Yeah.

If you are exploring the stars with chemical rockets you had better be very, very patient anyway.

Of course, the bot is wearing out as you speak...

A fusion drive using deuterium or hydrogen, as I have said already, could easily be wilderness refueled.

22 ppm.

A Bussard ramjet/scramjet could, of course, be fueled indefinitely off the interstellar medium and would by virtue of design necessity have to be able to use extremely abundant simple hydrogen for both propellant and fuel.

You need startup rockets for that, dearheart.

Finally, there is the possibility of a propulsion system that requires no fuel or propellant at all: solar sailing. An advanced solar sail making a close pass to the sun could be accelerated to some low fraction of c by the simple light pressure of the local star (this paper suggests a probable upper limit of .012 c), enough to allow for a slow but steady exploration of the galaxy with no worries about rocket fuel at all.

Now all you need is the ridonkulously large (and flimsy) sail.

Ah. According to this chart a typical C-type asteroid might contain perhaps 10 parts per billion of uranium, so while obtaining fissionables from asteroids might be possible it would require processing a lot of material.

Remember that most of that is going to be infissionable U-238. You need to enrich that stuff to use it in a reactor.

If you were looking for an engine for a self-replicating probe a fission reactor would be close to the bottom of the list of things I'd consider. Even a chemical rocket would probably be better, because hydrogen-oxygen rocket fuel would be much more easily obtainable.

So we just need a ready source of power and we're peachy.

PERMANENT has an extensive discussion on processing asteroidal metals here. I think I'll just quote it.
<snip>

I see nowhere in that quote or in the referring paper the energy cost per tonne of material processed. You're operating on an energy budget, you know.

Uh, yeah. You realise that the reason we have these traits is because they're useful in an unpredictable environment. A truly autonomous machine will have these traits too in some fashion.

It would probably need some form of survival programming, but there's no reason its own survival and the survival of its kind has to be its highest goal as it is with humans. A competent AI designer would make serving its mission its highest goal, and survival merely a means to that end ("if I die I can't fulfill my mission so I want to live").

I'm having a hard time trying to think of a situation where the bot will fulfill its mission by being wrecked.

This assumes that a zombie (in the philosophical sense) cannot effectively self-monitor. I am personally skeptical of this proposition, but it's hard to say one way or another until we understand more about the neurological mechanisms that create the feeling of consciousness in humans.

I have no idea what being 'a zombie (in the philosophical sense)' means.

That aside, mutants are a danger to us.

That is very true. Before entertaining the thought of a self-replicating probe I would want to be sure that the error-proofing systems on the replication are very, very good. As in they have a safety factor of orders of magnitude beyond the number of times the thing will actually be programmed to replicate.

Agreed.