today’s topic is self-replicating machines,particularly von neumann probes and berserkers. we will be looking at the basic concept ofself-replicating machines and some misconceptions about them, then moving in to discuss thosetwo specific types of them and some others. there’s also three things we should clearup about self-replicating machines straight from the outset. first, self-replicating machines do not haveto be small. second, we are arguably capable of makingone right now. third, self-replicating machines are not necessarilyjust machines, they can be alive. now in fiction self-replicating machines areusually implied to be small, tiny little nano-robots,
but they hardly have to be, nor do nano-robotsneed to be self-replicating. they are also usually portrayed as essentiallya single type, when you could easily have swarms of them composed of many sub-typesand sizes. often if something larger is desired theyjust clamp together to make it, but that’s not actually a good approach. for instance if we want metal to make moreof the little buggers we could send them in to go remove individual atoms of metal fromdirt, which is a regular portrayal but kind of silly, or we could have them form intoa conventional kiln to smelt metal, which is also pretty silly especially consideringthe whole point of the kiln is to smelt the
very metal they are made out of, and so themtoo. it would make a lot more sense to have themjust build a kiln, so this is our first strike against the classic concept of grey goo, someswarm of robots that looks like goo while eating up planets. the original concept for this is usually injohn von neumann’s concept of a universal assembler or constructor, something drexlerfollowed up with as a molecular assembler which is pretty much where all the conceptsfor nano-machines come from. but self-replicating machines way predatethat, to at least the time of rene descartes, who apparently told her majesty christinaof sweden that the human body was a machine.
the queen, being quite the scholar herself,apparently pointed at a clock and told him to make it reproduce. then not long after darwin’s notions begancirculating samuel butler toyed with the notion of self-replicating machines mutating andevolving consciousness. so this concept is a good deal older thanpeople tend to think and did not focus on tiny machines. on the second point, that we could probablymake one now, that size concept is important. if i have some automated factory rolling aroundeating up rock and spitting out new fully equipped factories that is still a self-replicatingmachine.
a 3d printer able to print itself is a self-replicatingmachine as well. it need not be able to make its own constructionmaterial though. after all, you and i are self-replicatingmachines and we not only do not typically convert matter into food directly ourselvesbut actually use a lot of integrated but independent life forms to keep ourselves alive. we don’t just eat other organisms, we havelots of little buggers hanging around in our guts helping us digest them. we even have mitochondria in every cell ofour body that reproduce themselves using their own genetic code.
useful hitchhikers for eons who we’ve formeda genuinely symbiotic relationship with even though we do not actually have code for themin our dna. so a giant factory that replicated itselffor instance would still be a self-replicating machine even if it had other self-replicatingmachines inside it that it was dependent on but did not actually make itself. since we are comparing them to organisms,let us examine that claim that self-replicating machines are alive. now there is no universally agreed upon definitionfor life but generally we would include the ability to eat, grow, excrete, replicate,and to adapt to and interact with its environment.
the reason it is often hard to nail down asolid definition is that almost any of those can be done away with while leaving a validclaim to life, at least in theory, but that is an important qualifier when we are discussingbuilding and tinkering with life. i have never seen a definition for life thatcomfortably contains all the normal examples but which would also exclude a self-replicatingmachine, though it might exclude some types of them. and i do not mean in a sort of vague way likewe might say fire is alive or crystals are alive. this is not semantics, the typical self-replicatingmachine we envision would have some ability
to eat, replicate of course, and have someequivalent to dna it used for that. now it need not necessarily have the abilityto grow or repair itself, so long as it is able to make another full-grown copy of itselfin decently less time than it usually takes for it to breakdown. biological organisms do not use that method,of just manufacturing something separate and fully-grown, they get bigger and subdivideinto two or make small copies of themselves that grow up. a self-replicating machine could be builtto do the same but it has that third option to construct a fully grown version too.
but a self-replicating machine does need ablueprint to work off of, same as any other organism, and rather than invent a new termor just say machine dna, i will just call this dna even though it would almost certainlynot be dna in most cases. in some it might be. after all quick path to self-replicating tinymachines is just tweaking the dna or rna of existing cells or viruses to do a job. gmos, genetically modified organisms, arean example of self-replicating machines, and again something we have now. so this brings up what we use these for.
what is their task or mission? obviously they do not have to have one besidescopying themselves but a machine is built with a purpose in mind. now you could use such devices for any numberof things but the two big ones tend to be use off the planet or use inside a human beingor other complex device. this is because the most appealing qualityof them is that they can help repair things, like help a human heal from injury or fixa piece of equipment so you did not need to throw it out or take it in for repairs. that’s very handy for things like spaceprobes since it means you could send a probe
out at relatively slow speed and expect itto still be working when it arrived at its destination solar system thousands of yearslater. now before proceeding i want to go ahead andkill the common myth that self-replicating machines invariably mutate. we mutate, other organisms mutate, and indeeda bunch of little robots could too, but they do not necessarily have to. even on astronomical timelines. mutation is an absolute necessity for evolvingfrom that most simple organism that presumably once assembled itself into more complex ones,but mutation is not a desirable trait if you
are building to a specific purpose. i do not want my probe heading off to theandromeda galaxy to mutate in the millions of years it takes to arrive. if i hand someone a book and tell them tocopy it word for word we know they will fail at that task, they will make a couple mistakes,and if they hand that to someone else that person will probably copy those mistakes whilemaking a few new ones, and so on until you get a copy that is just nothing like the original. that is mutation. if that is my only way of copying and preservingdata, say i am some old king and i want to
make sure my scribes maintain my memoirs properly,i can order three copies made of my memoir by three different scribes, that way if theoriginal is destroyed they can compare those three copies word by word and if they comeacross a word that is different in one than in the other two, they know it was probablythe two that are correct not the one, and can fix that. of course it is always possible those twoscribes made the same error, or that all three copies disagree, but both of those eventsare less likely. they are still likely enough though that itcan happen and with potentially millions of lines of code and millions of copying eventsyou would often get two identical errors or
spots where all three disagreed. if you add in a fourth copy, these odds decrease. if you add in five or six it gets even lesslikely, and you can increase that to the point where while it is still possible, the oddsof it happening even once over the age of the entire universe is less likely than not. so for instance i could have a setup so thatto make a new machine it required multiple machines to get together, just as with sexualreproduction but more polygamous, as it were. say, twenty other ones had to assemble togetherinto a dodecahedron, a platonic solid with twelve faces and 20 vertices, with each atone vertex, and they’d build the new one
in the middle. before adding each new bit they check andagree. if all twenty agree on a given chunk of theblueprint, all is well. if not, then the odds of less a majority ofthem having the right bit are freakishly tiny. we have done these kinds of extreme improbabilitiesbefore but the brain does not tend to work well with them and i am sure some of you arethinking right now, “sure, but it is still a chance so it will happenâ€. this is technically true but reaches a pointof absurdity when you begin dealing with improbabilities so high that they would be less likely thannot to occur if every atom in the universe
were turned into the things and we sat aroundwaiting until the stars all burned out. so you could for instance tell every machineit needed to match back up with 19 of its buddies once a year for a check where theycompare data, or to shut down if they can’t find 19. you could also have all sorts of differentspecies and protocols for dealing with unexpected events, and you can always ‘what if’ yourway into some bizarre circumstance but that’s not really the point. you might want your robots to mutate, youmight not, but if you want to send your robots off somewhere confident they won’t mutatebefore arriving, that is an option.
this one of the objections to what is calleda von neumann probe, so i wanted to clear that up before jumping in. john von neumann came up with the idea ofuniversal assemblers, often just called von neumann machines or grey goo, and this sparked5 major concepts for how this might be used in regards to deep space. one is the basic version and i’ll just callthat a von neumann probe even though the others are too, here are those 5 categories. 1) von neumann probe2) bracewell probe 3) terraforming swarm4) berserker swarm
5) grey goo swarm a basic von neumann probe is simply an interstellarprobe able to maintain itself with little robots and stop over place to repair, refuel,and reproduce copies of itself to explore more places. now in practice if it can self-repair thisway you’re better off launching all your probes from our own solar system. even if you needed to budget a hundred tonsfor each probe, about ten times what the hubble telescope masses, you could still get awaywith manufacturing a trillion of them, more than the number of stars in the galaxy, withthe available mass of just one medium-large
asteroid. those could all go cruising off from our solarsystem and arrive at their destination far sooner than if you send out a handful of probesthat slow down at the nearest stars, build more, launch those, which slow down againto build more, and so on. you’d be better off using that automatedproduction ability to have a small probe arrive, grab a small asteroid, and convert it intoa bigger monitoring station that can also act as a relay for information from ones furtherout. now the second type, the bracewell probe,you would be most familiar with from the movie 2001.
the monoliths in that were bracewell probes. a bracewell probe is designed to communicatewith other life forms, so it needs to be a good deal smarter and more adaptable. the most simple form would just be one thatwas able to self-repair and identify planets with a decent probability of life, and thensetup shop nearby there and sat around transmitting a repeating radio loop of how to make contactwith us and some basic info about us. sort of giant flashing neon sign saying hello,here’s our phone number along with a rosetta stone for how to talk to us. typically this is envisioned as a human-levelor greater intelligence though.
something with actual brains and decision-makingcapability. now technically a bracewell probe doesn’thave to be a von neumann machine but considering the timelines for interstellar travel andhow long it might need to wait when it arrived, you would need to be building the componentsunbelievably tough to expect it to survive for those kind of duration without the repaircapability of either being a self-replicating machine or able to make use of them to fixitself. also again it is probably more advantageousto ship these all out from our own solar system and just have them unpack and build themselveson arrival at the target. this has the advantage that it could set itselfup on some small asteroid and send in satellite
surveillance and even ground probes to collectdata. or make contact rather than have to sit aroundbroadcasting. if it can manufacture on site and has humanlevel intelligence it could pick up enough data to send in androids that looked likethe hypothetical primitive young race of aliens and chatted. for instance someone doing this to earth centuriesago might start with satellite surveillance, then stealthy aerial drones, then little androidbirds or mice for close looks and to get the language and customs observed, then send inan android to ask questions it could not get from just listening and watching, or to givethem information.
obviously you could take that into the ethicallygrey realm of trying to teach them or pretending to be a god. incidentally both of these approaches workjust fine for building manned spaceships too. you can build larger spaceships for peopleto be on with self-replicating machines helping in the building and maintenance but the assumptionis you can always build these automated versions cheaper and faster, both in terms of constructiontime and their velocity in interstellar space. now a terraforming swarm is essentially thenotion that you are sending out von neumann probes in advance to scout places for humansettlement and that those probes either have terraforming capability or would be followedup by ones that did and followed up by manned
ships later. the probe arrives at the destination and setsto work expanding itself so it can take on terraforming a planet. or even turning the entire solar system intohabitats as we have discussed before when talking about dyson spheres. this is another morally grey one because ifyou do not include an intelligence in that terraforming probe it might just go and terraforman inhabited planet, cheerfully disassembling the local flora and fauna in the process. that would seem pretty ghastly even if therewas no intelligence on those planets, on the
other hand you might think it was fine toterraform a place that only had amoeba on it. individual views range from this being wrongonly if there was intelligent life to it being wrong even if there was a decent possibilitylife might arise on that planet one day. of course some might feel it is fine evenif it had intelligent life on it. the berserker is usually seen as somethinga civilization might build if it had those kind of views, or made a terrible mistake. the name comes from a series of novels byfred saberhagen where the robotic ships or probes explore the galaxy to seek out newlifeforms and blow them up.
the ones from the series are not actuallyvon neumann machines individually but the collective whole was. a berserker is essentially a malevolent bracewellprobe, in that the bracewell probe’s mission is to seek out new life and meet it, the berserkerprobe is to seek it out and kill it, and our last type, grey goo, just wants to eat it. grey goo is occasionally called a hegemonicswarm, which i think originated from author iain m. banks. this doesn’t have to be robots or tiny littlemachines, the specific grey goo case, so i will use hegemonic swarm as a maybe more appropriateterm.
this could be grey goo, where self-replicatingmachines just fling themselves across the void stopping at stars and turning all thematter into more of themselves, but it could also apply to something like the borg fromstar trek or a crazy machine intelligence that has decided it needs to turn the wholeuniverse into paperclips. alistair reynolds even had a particular varietyof this in one of his books that began as a terraforming swarm and through bad designwas racing around turning everything into habitats around a star, into dyson swarmsbasically, already filled with flora and fauna, but was attacking colonized planets and spaceshipstoo, to add them to the habitats. this is one of the reasons i wanted to takesome time on the mutation issue because it
is often assumed any self-replicating probeswould turn into berserker armadas, grey goo, or hegemonizing swarms given enough time tomutate. once genuine mutation is in play, especiallyon machines that were not sentient, it is reasonable to assume they would start mutatingtoward strictly darwinian goals like survival and replication. from that comes an assumption that left looseto run around the galaxy unsupervised for long times most of the nice and benign orhelpful types of these von neumann machines will turn all evil. so it worth remembering there are ways toprevent mutation, but something often overlooked
is that mutation does not change somethingfrom a to b, it turns a into a whole alphabet and then a library, given enough time. there is a reason right now that even thoughmy billions of times great grandfather was an amoeba i am sitting here at the moment,and there’s also a reason why there are still tons of tiny and simple micro-organisms. so you would expect a runaway mutating self-replicatingmachine to result in a whole ecosystem. at the solar system level you would expectto see the bottom of the food chain probably be self-replicating machines that swarmedin the trillions around a star sucking up its light then other things which came byand ate them and got eaten in return.
probably complete with detritus eating versionsat the far end of the food chain and ones swimming around kupier belt grabbing cometsand small asteroids freshly arrived from deeper out. such things are not an example of machinelife, they are just an example of life, its kind of silly to think of it any other wayat that point. and it is worth remembering that is how lifebegan on earth too. our planet did get gray-goo’d by those earliestlife forms and probably more than once. i occasionally find it amusing to think ofintelligent life as an adaptation of grey goo to produce a new wave of it that can leavean atmosphere or solar system, since classic
evolutionary processes do not lend themselvesto those kind of jumps. okay some final notes about self-replicatingmachine and nano-machines in general. i already mentioned that mutation does nothave to be an automatic feature of these, but with that reminder that we basically descendfrom grey goo i should address the misconception some get that tiny little robots swarmingaround can just disassemble whole planets in days. i mentioned speed limits a couple videos backwhere 3d printing machines are concerned and tiny little robots have them too. just for conceptual purposes, keep in mindthat bacteria can reproduce quite quickly
compared to us, often able to double on atimeline of an hour or so. meaning if you start with one you could havea million the next day and a trillion the next day and a quintillion the day after that. on paper anyway, in practice exponential growthtends to get dampened by other effects. however there is a clear evolutionary advantageto being able to grow and replicate quicker, and to be as omnivorous as possible aboutyour sources of food and fuel. yet bacteria do not split in two every second,even the fastest viruses, which are incredibly simple critters if they are organisms at all,are not that fast. complexity has a cost, it takes longer toassemble.
now a machine should be designable that doesreplicate faster than biological life, but it is not very likely to be many orders ofmagnitude faster than organisms of the same size. it is also worth remembering that chemistryand construction tend to produce a lot of heat, there is a reason bread dough or compostor other things bacteria go nuts in tend to get hot. you can only go so fast replicating beforethe heat would get so bad it destroyed the machines doing it. it is also a lot harder to snatch and placemolecules in something you are building when
it is hot and all the molecules are bouncingaround much faster and also bouncing into whatever you are building. we tend to forget that at the molecular levelhotter temperatures mean stuff moving quickly, but building at that scale in a hot environmentwould be like trying to pitch a tent in the middle of a hailstorm. heat, as we have seen in a lot of our topicson this channel, tends to be a big bottleneck on a lot of processes. you also have to remember that tiny thingsare very fragile and that each component adds more material, more time to build it, andslows the process.
how are your tiny little robots getting power? solar energy? not very useful for anything but the surfacelayer when its sunny and a solar panel can only be so thin before it would be incrediblyfragile. you are not getting a nuclear power source,fusion or fission, that small. that just is not how that works. that leaves either battery power, which itneeds to go recharge somewhere and could get quite bulky, or using existing fuel in whateverit is disassembling. that is great for medical nanotechnology,you can design it to run off your own power
supply in your cells, but tiny and universalchemical fuel eaters are not exactly doable. you would almost have to give it a separateengine for each type of fuel supply so it could run on oxygen and methane in one place,solar in another, sugar in another, etc. adds more complexity, adds more bulk, addsmore replication time. want to shield it from an emp blast? add shielding, adding more bulk and more replicationtime. also slows down how quickly it can do otherthings since it uses more energy to move its greater bulk and has more of its mass devotedto things other than moving and assembling and disassembling things.
want it smarter, more material, more energy,slower replication times. probably a lot faster than biological lifebut not likely to be superfast. scale always been hard on people, lots offolks think your typical biological cell is made up of atoms and molecules like they werelego bricks, in practice cells do not have dozen of atoms or hundreds or thousands, theytend to have trillions. if you’re thinking of atoms or small moleculesas building bricks of cells, don’t picture a house, which has several thousand bricks,picture a major metropolis, or even an entire planet, that’s the scale of most cells comparedto atoms and simplest molecules. even a typical virus tends to be in the hundredsof thousands, more akin to a skyscraper than
a house. your average mammal like a human or a puppywould be more like a galaxy in this atom equals brick analogy. so yes you could build machines that werequite tiny compared to bacteria, probably were pretty sturdy, versatile, clever, andfast to reproduce but we do not want to get carried away and think of them as invincibleswarms of grey goo moving over a planet like a tidal wave or possessing magic powers. we also want to remember they could come ina lot of sizes, from on par with a virus to quite larger than a person.
so when will we have these? i would guess quite soon, again we arguablyalready have them and it’s an important area of research. automated construction, both for factoriesand for use in space, would be very useful, and is already very useful. at the microscopic scale the medical applicationsare huge and so is the convenience angle of having colonies of specialized tiny robotsthat hung around our appliances fixing them, particularly since you can make things smallerif they do not have to be durable enough to survive minor damage which could instead berepaired.
you’ve got two options on something likethat too, you don’t have to have the machines able to self-replicate, you could have themwithout that and produced somewhere and you buy a vial of non-replicating ones with anexpiration date. back at the factory for them slightly biggermachines churn out millions of the tiny ones constantly, which is advantageous since youcan save mass and energy by making them more specialized and without all the extra bitsfor reproduction and have some smarter bigger control bots that can issue them orders. they’d probably be more like a solutionof various species of nanobot. ditto you could still go the replication routejust have that done a couple steps up, with
smarter bacteria sized mobile factories thatbuilt them as needed and could get updates and issue updates. lot of potential uses and a big game changingtechnology. one i suspect most of us will live to see,but also not a magic wand or instant doomsday device. their value for medicine is huge, potentiallymaking us biologically immortal, and their value for space exploration and colonizationis equally huge. that leads to our topic for next week, spacecraftpropulsion, where we will be doing a quick survey of all sorts of spaceship propulsionideas either in use now or on the drawing
board, including the em drive. that topic was selected by drew mctygue, thefirst winner of the topic contest we had over on the patreon account for this channel. i’ll be drawing the second winner in a fewdays on sept 20th, coincidentally my birthday, so you can still go over to patreon, becomea channel patron, and submit a topic which will follow dark energy, our poll winner fromlast week, and either come right before or after crypto-currency, the runner up, dependingon how much time i need to develop that topic. incidentally we will be shifting all futurepolls to the channel’s new facebook group, science & futurism with isaac arthur, andthose will now be ongoing and we will have
one set of polls to suggest ideas and anotherto pick the winner from the most popular, and again that will just be getting constantlyupdated. since this is the debut week for that facebookgroup, i will be hanging out there the day this video comes out to answer questions,though i’m quite active on facebook normally anyway. my thanks to all the folks who agreed to helpmod and admin that, since that lets me leave it a lot freer for general discussion of othersubjects too like other channels you like or general discussion of science and sciencefiction. i do get asked a lot what other channels iwatch and its mostly comedy and scifi on youtube,
science-wise i have a few recommended channelson my channel page for joe scott, grant thompson the king of random, and cody don reeder fromcody’s lab, cody in particular has given me some great advice in the last couple weekson running the channel for which i’m quite grateful, but more importantly he makes somegreat science videos, so i’d suggest checking them out if you haven’t already. there’s not nearly enough science, i meanreal science, out there on this medium and it’s nice to see other folks trying to correctthat too. last note, there is now a sub-reddit for thechannel though we will be launching that officially next week to give tim a bit more time to polishit up and let me focus on the new facebook
group this week, and of course you can followme on twitter, visit the website, isaacarthur.net, listen or download the audio-versions of theepisodes from soundcloud, or find me on patreon. again next week is spaceship propulsion concepts,followed by dark energy, and i hope you will swing by a join the facegroup page, say hi,and vote in the episode polls and stop by to do that regularly. until next time, thanks for watching, andhave a great day!