ms@pogo.ai.mit.edu (Morgan Schweers) (02/27/91)
Greetings, How much of nanotechnology is vaporware/dreaming? Can anyone point me to a solid and *REALISTIC* exploration of PRESENT DAY research on this topic? I've gotten a *LOT* of $#!t from people, when I try to explain nanotech to them. The things that got me the most strange looks were: * Nano-Dissassemblers - The idea that something can actually be programmed at that size, and then ACTUALLY HAVE AN INFLUENCE on other items seems to be a sticking point for a lot of people. What sort of materials are REALLY dissassemblable? * Nano-Assemblers - The same problem, really. Even when people manage to accept the idea of dissassembly, they rarely accept the idea of reassembly. * NanoProgramming - Is it REALLY possible to actually *PROGRAM* something that small? What *IS* the size that we are talking about? * Movement - How does something that small MOVE? * Power source - Obvious. What's their power source? * ETA - What are the optimistic assessments of when this technology will be available? The pessimistic? Or is all this just a joke? Any other information on the *REALITY* of nano-hacking would be greatly appreciated. -- Morgan P.S. I've read Blood Music, and consider it nonsense. I've also read a book named something like Down The Sea Of Stars (or something-similar) and it's nano-techs seem to make a *LITTLE* more sense. (but not much) +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | I DON'T UNDERSTAND!!!!!!!!!!!!!! | I understand perfectly, | | This makes *NO* SENSE! | You simply don't comprehend| | I'm *SCARED*!!!! | my genius. | | -- Morgan | -- Nagrom | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ms@ai.mit.edu OR ms@albert.ai.mit.edu(preferred) OR ms@woodowl OR ... [Read Engines of Creation by Eric Drexler (Anchor Doubleday), now out in a second printing. Look on the /nanotech/papers FTP directory on planchet.rutgers.edu. I could wish there were more discussion of these "lower-level" more technically oriented questions on this newsgroup. Please, anyone with anything to add to these questions, don't hold back just because I've given them a "lick and a promise". With that caveat, the short answers: Assembly/disassembly: Cells (both bacteria and cells that are part of larger organisms) do this all the time. Nanotechnology simply presumes that we can translate mechanical design concepts to the same scale. Programmability: Similarly, cells have a "program" in their DNA. We are assuming that the structures of formal computation can be reproduced at a molecular scale. The nanocomputer, in a crude mechanical version, is actually the most complete actual nanotech design so far. Movement: In many envisioned applications, the nano-robot floats around at random in some solution, making desired contacts with raw materials and other nanobots stochastically. Otherwise it could crawl or have propellors. Power: Most commonly suggested is chemical fuels in solution. Other schemes include tuned antennas converting some wavelength of light to electricity. ETA: Optimistically, 200x. It does depend on the amount of effort expended to that end. Could we land a man on the moon before 2000 (starting from where we are right now)? Will we? --JoSH]
bill@braille.uwo.ca (W.B. Carss (519) 438-0344) (03/01/91)
Let me preface what follows by saying that my ignorance on this topic could fill volumes. I am neither a "scientist" nor a "theoratician" (nor probably as you can tell, a speller of any great note). Even so, I have a few ideas which I thought I would add to the discussion. Probably the best thing I have going for me is a good imagination which, I have often been told, is more than a little over-active. As Josh<?> mentioned, there are several examples of nanotechnology in existence in the world around us right now - bacteria, cells in our bodies, process involved in digestion and many many more. The big question, as far as I can tell, is whether we will be able to get enough of a handle on EXACTLY how things are done to CAUSE THEM TO BE DONE OURSELVES. Certainly, many people would say that we are already doing that with genetic engineering. This is only partly true. From the little I know of the topic, it seems to me that what we are doing is CAUSING existing systems to make changes for us. There is a big difference between telling a calculator to find the nth root of a number and knowing how to do it yourself. I think we are still pretty-much at the 'calculator' stage of our nanotech development. That isn't to say that we won't eventually get there, only that we have an awful lot to learn before we are even generally knowledgeable enough to make any SERIOUS of MEANINGFUL attempts. I think one of the major dangers, (without wishing to squelch any dreams or rain on anyone's parade), is that we may try to do more than we are ready for too soon and botch it. From my own experience, I have done this several times in all kinds of situations. Certainly, the only real result in my own personal case is a fail to accomplish what i have set out to do. In the case of nanotechnology, however, it isn't inconceivable that something may be created which we can't control. I don't wish to be an alarmist nor anything like that, but I do believe that GREAT PAINS SHOULD BE TAKEN TO INSURE THAT BEFORE ANY NEW "MACHINES" GET CREATED BY US WE KNOW WHAT THE HEC WE ARE DOING!!! Certainly the ever-popular trial and error method is just about the only way we will know for sure whether something works, and I have no problem with that except to say that when the trial takes place let's make darn sure it is in a situation where we can control WHATEVER happens!! AND I MEAN CONTROL!!!!!!!!!!!!!!!!! All we need is some "rogue" machine running loose doing who knows what as a result of mutation to who knows what. I am thinking specifically of machines that are able to self-replicate. In any such situation millions if not billions (thanks Karl) of these machines would be necessary to really accomplish anything of significance on our size scale. Therefore, the machines would have to be self-replicating and the "offspring" would be prone to "mutation" (for lack of a better term). What would these mutations be "designed" to do? Would the mutations increase the rate at which the self-replications occur? Would this be a perpetually compounding problem? It could be argued that there could well be "guard" machines to oversee the machines that are self-replicating. In such a manner some "control" could be exerted over the situation. To do this would require millions or billions of "guards" and the problem recurs. A related concern is indeed the "fuel" that these machines use. What if the mutations "require" a different fuel than we planned? What if the mutations "took a liking" to organic matter? Never mind us, plants, useful bacteria, the list isn't endless, but I am sure you get the point. Without wishing to sound too foolish, this is the stuff of which "The Attack of the Killer Tomatoes" is made. And whether anyone wants to actually discuss these ideas, they ARE within the realm of possibility. All I am saying is that we should go slowly and carefully. What we are really talking about here, in essence, is the creation of life from lifelessness. At least I believe that nono-machines could be considered "alive". We don't have any real idea what the release or even exposure of "beings" such as these into our environment would do. We don't know how the other organisms in our environment would react to these newly arrived "intruders". Could we end up with a similar situation to rabbits in Australia, where no "natural" predator (or in our case controls) exist? -- Bill Carss bill@braille.uwo.ca [This is essentially what is referred to as the gray goo problem, from the concept that unchecked replication could lead to the entire biosphere being consumed by nanobots and there would be nothing left but a "goo" consisting of them. This scenario is considered quite unlikely and overdramatic by most who have studied it seriously. The reasons are several. First, the only reason we have to believe that we can build a nanobot more efficient than a bacterium, for example, is that it would be built like a machine: it would be specialized, it would have precise components built to atomic precision, it would have a highly sophisticated design. By this very assumption, it COULD NOT MUTATE. The inefficiencies in cells are the very thing that allow mutation, and for lifeforms, that's good. But you couldn't build a nanobot to mutate unless you tried very very hard to achieve that specific goal. On the other side of the assumption, of course, is that if your nanobots are not more efficient than say, bacteria, they won't win out over bacteria when taken out of the laboratory environment. Fears of accidental gray goo scenarios are less comparable to rabbits in Australia, than to a story where feral automobiles run wild, mutating into herds of grass-eating vans hunted by carniverous pickup trucks. I would worry instead about what people do with them on purpose; I was in Australia recently and I saw a hell of a lot more sheep than rabbits. --JoSH]
barryf@rpi.edu (Barry B. Floyd) (03/01/91)
I am no expert, though my interests are more than passing... In my quest for relevant information I find Science News (MIT? weekly?) and Scientific America (monthly) to be accessible (intellectually). Each has run one or more stories in recent months describing independent efforts to "move" individual atoms using normal-scale machines. Researchers have successfully "written" names and designs by manipulting surface and sub-surface atoms (details and references are not on hand). I am less atuned to the biological approach (e.g. protien machines, enzyme machines) etc. though I have read "Blood Music" et al and find it plausible. Several genetic engineering firms written up in financial newspapers seem to be predisposed to advances and commercial applictions along these lines. To the extent that such companies (vs Universities) exist, prospects seem positive. barry -- +--------------------------------------------------------------------+ | Barry B. Floyd \\\ barry_floyd@mts.rpi.edu | | Manager Information Systems - HR \\\ usere9w9@rpitsmts | +-Rensselaer Polytechnic Institute--------------------troy, ny 12180-+
erich@eecs.cs.pdx.edu (Erich Stefan Boleyn) (03/01/91)
ms@pogo.ai.mit.edu (Morgan Schweers) writes: > How much of nanotechnology is vaporware/dreaming? Can anyone point me >to a solid and *REALISTIC* exploration of PRESENT DAY research on this topic? [...questions deleted...] > Any other information on the *REALITY* of nano-hacking would be greatly >appreciated. What do you mean by "nanotech"? I've seen some quite fascinating ideas spun on this group, some with considerable technical sophistication to it, but I wonder how realistic people are about bringing it about (are they too hooked up on the dreams?). There are some of us who are at least interested in working on the *REALITY*, and are preparing. IMHO, there seems to be a surprising number of people on this group interested in interdisciplinary studies, which seem to me to be the best way to get there. I have few illusions as to my own ability to understand all the complexities involved. Just a little molecular genetics can dampen one's spirits fast unless one is willing to stick it out. > P.S. I've read Blood Music, and consider it nonsense. I've also read > a book named something like Down The Sea Of Stars (or something-similar) > and it's nano-techs seem to make a *LITTLE* more sense. (but not much) David Brin gives a rather pessimistic, but plausible (given other precedents set in history) view of nanotechnology in "Earth". In short, it says that they are very specific and need to have absolutely *pure* nutrient baths... and even then can only produce repeating units of some sort, like a crystal. Apparently the rest is from the editor [JoSH...], I guess: [...encouraging comment deleted... ;-)] > With that caveat, the short answers: > Assembly/disassembly: Cells (both bacteria and cells that are part > of larger organisms) do this all the time. Nanotechnology simply > presumes that we can translate mechanical design concepts to the > same scale. Ack! Have you studied any molecular genetics? (well, you probably have ;-) IMHO, there seems to be a definite lack of mention just how radical a *transformation* of concepts would be necessary to achieve such a translation of scale. What I know of molecular genetics seems to clearly indicate that the mechanisms involved are many orders of magnitude more complex, even in the prokaryotic case (single-celled organisms). Many people resist the notion of parallel computing (at least doing the parallelizing work themselves), much less having to work with complex automata-like systems. Now, of course, this doesn't stop some of us die-hards from trying anyway ;-). > Programmability: Similarly, cells have a "program" in their DNA. > We are assuming that the structures of formal computation can be > reproduced at a molecular scale. The nanocomputer, in a crude > mechanical version, is actually the most complete actual nanotech > design so far. A *program*?!? Arghh... although I will grant you that it *can* be called a "program" per se, this has no reference as to the encoding of this program. Needless to say, it is neither linear, nor easily decodable. This doesn't account for the fact that these programs are perhaps meant to have more (and different) long-term functions than anything we currently have. Lately, I have been considering the concept of what I call "minimal encodings", sort of like packing the most information possible into a set of instructions. It seems that an information theoretic-like attack on this problem might have some interesting leads. Again, the concept has to undergo radical revision. > Movement: In many envisioned applications, the nano-robot floats > around at random in some solution, making desired contacts with > raw materials and other nanobots stochastically. Otherwise it > could crawl or have propellors. This seems reasonable. There is a question of how generally the forms of the "raw materials" would take... remember that our own assembly systems are easily fooled by look-alikes in building proteins... stochastic systems, even in their most specific forms, can still in many cases be fooled, especially as they get sufficiently small. > ETA: Optimistically, 200x. It does depend on the amount of effort > expended to that end. Could we land a man on the moon before 2000 > (starting from where we are right now)? Will we? This very much depends on what we want to work for... and how much we do about it. Discussion is, of course, encouraged. I also am very interested in getting the discussion into details/ideas of substance. Erich "I haven't lost my mind; I know exactly where it is." / -- Erich Stefan Boleyn -- \ --=> *Mad Genius wanna-be* <=-- { Honorary Grad. Student (Math) }--> Internet E-mail: <erich@cs.pdx.edu> \ Portland State University / >%WARNING: INTERESTED AND EXCITABLE%<
toms@fcs260c2.ncifcrf.gov (Tom Schneider) (03/01/91)
In article <Feb.26.17.12.27.1991.23759@athos.rutgers.edu> ms@pogo.ai.mit.edu (Morgan Schweers) writes: > How much of nanotechnology is vaporware/dreaming? Can anyone point me >to a solid and *REALISTIC* exploration of PRESENT DAY research on this topic? > * Nano-Dissassemblers - The idea that something can actually be > programmed at that size, and then ACTUALLY HAVE > AN INFLUENCE on other items seems to be a sticking > point for a lot of people. What sort of materials > are REALLY dissassemblable? Enzymatic digestion is an example. The proteins are joined by a 'peptide linkage; add water and they come apart. So the rebuttal is: Eat your dinner: it gets disassembled at the molecular level! > * Nano-Assemblers - The same problem, really. Even when people > manage to accept the idea of dissassembly, they > rarely accept the idea of reassembly. DNA is copied into RNA in the cell. That is, for the set of the chemical "letters" in DNA (the bases a,c,g,t separated by deoxy ribose sugars and phosphates), the set of RNA "letters" (the bases a,c,g,u separated by ribose sugars and phosphates), is created. Then sets of three RNA bases are read to insert amino acids in a growing chain in the process of "translation". The string of amino acids folds up into proteins that are involved in all these steps plus many other wonderful things (like sensors and feedback control systems!) So the answer to your friends is: what you eat makes you grow! > * NanoProgramming - Is it REALLY possible to actually *PROGRAM* > something that small? What *IS* the size that we > are talking about? This one is harder to give an example of, because 'program' has the specific meaning of using a high level language such as C or Pascal on a general purpose computer. But consider all the instinctive behavior that animals do. It is 'programmed' by the genes. Temporary Rebuttal: teach fido to get the stick. > * Movement - How does something that small MOVE? Nobody knows how muscles work, but they do work at the molecular level! Rebuttal: a good punch in the nose should do the trick! > * Power source - Obvious. What's their power source? In biology, the triphosphate nucleotides, mostly rATP are energy sources. But to get the ATP, lots of other tricks are used, sunlight being the main one. Rebuttal: dab their bloody nose with cotton balls. The cotton was grown using solar power and constructed by molecular machines. > * ETA - What are the optimistic assessments of when > this technology will be available? The pessimistic? > Or is all this just a joke? Lesee - I forget. Just how long has life been around on the planet? At least a billion years. So the optimistic assesment of when we can use nanotechnology is about a billon years ago! Of course I understand you mean: when can we begin to direct it for our own use. Well, we've made bread since ancient times; we make drugs, we now modify enzymes... It's here folks! Of course the full general assembler idea of Drexler is not yet, but I wouldn't bet more than 50 to 100 years at the rate we are going. > Any other information on the *REALITY* of nano-hacking would be greatly >appreciated. Read lots and lots of molecular biology. A good source is: @book{Watson1987, author = "J. D. Watson and N. H. Hopkins and J. W. Roberts and J. A. Steitz and A. M. Weiner", title = "Molecular Biology of the Gene", edition = "fourth", year = "1987", publisher = "The Benjamin/Cummings Publishing Co., Inc.", address = "Menlo Park, California"} Well, I was planning on waiting to see when people would notice my latest two papers, but it looks like people aren't willing to post references or don't look where I put them... Anyway, all of you budding nanotechnologists (molecular machinists) will have a fun time reading: @article{Schneider.ccmm, author = "T. D. Schneider", title = "Theory of Molecular Machines. {I. Channel} Capacity of Molecular Machines", journal = "J. Theor. Biol.", volume = "148", number = "1", pages = "83-123", year = 1991} @article{Schneider.edmm, author = "T. D. Schneider", title = "Theory of Molecular Machines. {II. Energy} Dissipation from Molecular Machines", journal = "J. Theor. Biol.", volume = "148", number = "1", pages = "125-137", year = 1991} In these papers you will find plenty of references. (Note: figure 1 is on page 97, but should be placed just after page 84.) The second paper proves that it will be possible to create ACCURATE computers built out of molecular parts. > -- Morgan > P.S. I've read Blood Music, and consider it nonsense. Perhaps you should read it again. Tom Schneider National Cancer Institute Laboratory of Mathematical Biology Frederick, Maryland 21702-1201 toms@ncifcrf.gov
bill@braille.uwo.ca (W.B. Carss) (03/03/91)
[The arguments in this message are, fortunately for the future of life on Earth, flawed. See below. -j] >[This is essentially what is referred to as the gray goo problem, from {...lines deleted} > > Fears of accidental gray goo scenarios are less comparable to > rabbits in Australia, than to a story where feral automobiles > run wild, mutating into herds of grass-eating vans hunted by > carniverous pickup trucks. I would worry instead about what > people do with them on purpose; I was in Australia recently and > I saw a hell of a lot more sheep than rabbits. > To suggest that we can build machines that work perfectly EVERY time forever is (IMHO) just about as silly as the paragraph I have reproduced. All I said, essentially, was that considering our record so far, i.e. the things we think we KNOW, I would be willing to bet that there will be a lot more POTENTIALLY serious screw-ups than successes. In your GRAY GOO summation you neglected to discuss our i.e. human capacity for error, our often frequent assumption or at least attitude that we know everything and the frequent occasions when we find out that we don't. We are a long long way from being anywhere near "programming" these nanobots, nor even doing much more than day-dreaming about them. For you to suggest that sometime between now and when ever we are actually able to build or design them we as human beings will exhibit a lot less arogance and a lot more sense is just about as silly as your trite dismissal of a potential problem that you didn't seriously address or even consider. The problem isn't nanotechnology, it is US. Our inefficiencies and stupidity which has been shown time and time again. Let me take this a step further ... We design nanobots that are either self-replicating or are built by other nanobots. Assuming that ever replication from which ever source is not perfect, some of the machines will be flawed. To check for flawed machines we design sel-test abilities it to the replicators. Assuming that the flaws don't develop in the self-test portion of these machines everything works fine. If, however, the flaws develop in the self-test portion of the machine, you will either have a machine which is scrapping good nanobots and/or passing flawed nanobots. What will these flawed nanobots create? We have no real way of knowing. What type of end-product would be built by flawed nanobots? Again, we have no real way of knowing. Could we test the testers with testers with testers ... We would end up with so many testers nothing constructive could be built because all of the available resources would be consumed by testing the test ... So now I state the ultimate question, and I mean this very seriously, CAN SOMEONE SHOW ME WITH SERIOUS SCIENTIFIC REFERENCES I.E. SOME RESULT FROM EXPERIMENTATION OR SOME OTHER TESTABLE MEANS WHY THIS CRITICISM IS FLAWED? I truly believe that it requires a more serious answer than the ever popular gray goo catch-all which you have used. I challenge you, in a friendly way to show me that I am wrong rather than attempting to sweep my criticisms under the proverbial rug. -- Bill Carss bill@braille.uwo.ca [The major flaw in the above argument is that it doesn't take any account of the difference between a machine and an animal. Bill has taken a major (and completely unsupportable) leap of faith: since we are quite likely to make mistakes (true) our machines will suddenly become supermachines able to take over the world in spite of all we can do to resist them. A machine built for some useful purpose will tend to be as efficient as we can make it for that purpose. This means that it will tend to be highly specialized, run on special fuels or power sources, require inputs preprocessed for their special purpose, etc. Consider a car. It runs on gasoline. For a car to turn feral, it would have to convert to some naturally occuring fuel, say wood. You could *design* a car to live off the land; it would come with saws and chippers for harvesting trees, some low-efficiency but highly robust motor able to burn anything in some broad range; it would trade speed for off-road capability, etc, etc. Spend a little time actually trying to design a self-fueling, self-repairing car. Now come back and tell me how you are accidentally going to build this amazing vehicle, without intending to, by making a MISTAKE in the process of building an ordinary regular car that runs on gasoline, and is fixed at service stations with parts manufactured by factories. If you don't understand why I'm talking about feral cars you've missed some important point and need to go back and try to explain why you think gray goo could happen in the first place. Early nanotech thinkers realized that nanomachines could be much more efficient than natural organisms and it is easy to jump to the conclusion that they could thus outperform them in an evolutionary struggle and take over the biosphere. That's wrong. It's based on completely ignoring what it is that makes natural organisms less efficient than machines. The answer turns out to be flexibility, adaptability, self-repairability, evolvability. The lesson to be learned from this is that using nanotechnology for some given task might be much safer than the alternative if that alternative was to use BIOtechnology. Modifying actual organisms does *not* have the safeguards of specialization, inflexibility, brittleness of design, and so forth that a mechanistic approach to nanotechnology has. With biotechnology you're talking rabbits instead of cars. --JoSH]
cphoenix@csli.stanford.edu (Chris Phoenix) (03/03/91)
Josh writes: > The inefficiencies in cells > are the very thing that allow mutation, and for lifeforms, that's > good. But you couldn't build a nanobot to mutate unless you tried > very very hard to achieve that specific goal. From what I've heard, it is true that a nanomachine can be easily designed to avoid mutation. But I don't believe it would be that hard to build one that mutated. All you'd need is some encoding of the specification in a format such that a high (say, .01%) number of random changes to the spec produced something meaningful. Then program it to change 3 bits of the spec before it replicates itself. "genetic" algorithms work in finding good solutions to problems, and while I don't know much about them, it seems that there should be a way to code a machine spec so that it could be optimized in this way. I can't see why anyone would want to, though. Seems like once we get nanocomputers it would be easier to do a top-down design and simulation, and get a machine that does exactly what we want (we hope) rather than relying on chance. [Genetic algorithms are a good example of what I'm talking about. As an experiment, try writing a self-reproducing program in C that introduces random changes in itself, and still works. Genetic algorithms use highly inefficient production system mechanisms for the same reason cells do--because they are the only way (we know of) to make evolution actually work. --JoSH]
bill@braille.uwo.ca (W.B. Carss) (03/09/91)
In article <Mar.2.22.52.40.1991.21778@athos.rutgers.edu> bill@braille.uwo.ca (W.B. Carss) writes: >[The arguments in this message are, fortunately for the future of life > on Earth, flawed. See below. -j] >[The major flaw in the above argument is that it doesn't take any account > of the difference between a machine and an animal. Bill has taken a > major (and completely unsupportable) leap of faith: since we are > quite likely to make mistakes (true) our machines will suddenly become > supermachines able to take over the world in spite of all we can do to > resist them. No, I don't believe I have taken that leap at all. The point that you have apparently missed or refuse to acknowledge is a little aspect of industry called "quality control". If you machine building is so un-flawed, why do we need quality control? If our production methods were so perfect you wouldn't ever see people returning things that don't work properly. Computers (at least ours) make errors perhaps once every few million operations. Why is that? What is the result of the error? In the case of nanobots, it would take billions of them to create anything of an appreciable size i.e. something that is large enough for us to get any real use out of it. If you have billions of machines each making one mistake every ten million operations, that makes a lot of mistakes. Certainly, in self-replicating nanobots most of the mistakes would result in nanobots that are not viable. But in those cases where the mistake has not led to what we will call a fatal error what will be the result? We don't know. Let's say we have built machines that respond to the colour navy blue. We have designed these machines to replicate themselves and in the process an error occurrs so that some of the machines now respond to robin's eff blue not navy blue. the machine would still "work", it would just be responding to a different shade of blue. Whether this is a serious aboaration or not depends on what the machines "do" when they come into contact with the activating colour. Wuppose they are designed to break-down the navy blue item whatever it is. Those machines that are now responding to robin's egg blue would be breaking-down robin's egg blue items not navy blue items. What would be the result of all of that? As far as killer nanobots are concerned, certainly that may be the stuff of which science fiction is made. My point still is (and was) a question. How do we control what would essentially be mechanical errors? -- Bill Carss bill@braille.uwo.ca [Again: You are talking about two distinct phenomena: (a) [shades of blue] a machine with a specific, designed, function, performs that function on something slightly different than intended; (b) [gray goo] due to an error in copying, a plan for a ten-thousand part machine which does one specific function and runs on one specific fuel, becomes a plan for a 100 million part machine, able to perform hundreds of functions, recognize the circumstances under which each should be performed, run on a wide variety of naturally occuring energy sources, and survive the chemical attacks of the natural, highly adaptable microorganisms it will compete with. Think of a car again for a moment. Suppose we have a working car, and we come up with some improvement that consists of a new design for one of the mechanical parts of the engine. Can you design a car so that to incorporate the new part, I simply open the hood and throw it in? Well, guess what: cells work that way. If a copying mistake produces a better part, anywhere, it works, automatically. In a mechanical design, you have to change the whole design in a highly coordinated way to incorporate improvements. Almost all the copying errors in a cell are detrimental, i.e. they make it work less well. A tiny fraction improve, or simply change, its function. Almost all of even that tiny fraction, in a mechanical design, would simply cause it not to work at all. Try changing a bubblesort program into a heapsort, one character at a time, with the constraint that each intermediate form not only sorts correctly but does so at least as well as bubblesort. --JoSH]
mmt@client2.DRETOR (Martin Taylor) (03/09/91)
I think that much of the argument as to whether nanobots might be likely to mutate into grey goo hinges on a difference of opinion about the underlying structure of a successful nanobot. One designed to do "exactly what we want (we hope) rather than relying on chance" is likely to be built, shall we say, symbolically. and rely on rules with truth values near 1 or 0. The design space is large, and viable designs few and far between. JoSH's arguments apply pretty well to such machines (even though theoretically the chance of an error leading to a new viable design can never be reduced to zero). But it is unlikely that we will know "exactly what we want" the machine to do, and even if we did, we would probably want it to do something quite similar if it was confronted with circumstances very like those we envisaged in our designing. A machine with these desirable abilities would be in a design space that (at least locally) was rather dense with viable machines, and the probability of a mutation leading to a viable design could be appreciably different from zero. If a mutated machine propagated its design better than the original did, then it has at least made the first step toward grey goo. I'm not sure that the argument is as clear-cut as either the worriers or JoSH make it out to be, but I am sure that it is better to err on the side of prudence, and think very carefully about all the trade-offs between behavioural flexibility (topological neighbourhoods likely to contain viable points), design rigidity (enpty neighbourhoods but probably ineffective machine), and mutability. I know that there is not a LOGICAL connection here, but there is a probable linkage if designs are not well thought out. For example, with what we know now, behavioural flexibility is likely to be attained through the use of distributed representations for the perceptual-behavioural knowledge and the incorporation of trainability. But a design of this kind which replicated itself would be very likely to produce a working descendant if some mutation altered the form of the network. It would just do something a little different. I think here we have a situation much closer to that of natural evolution than is envisaged by the "clean design" school. -- Martin Taylor (mmt@ben.dciem.dnd.ca ...!uunet!dciem!mmt) (416) 635-2048 To be a fundamentalist takes considerable flexibility of mind.
forbis@milton.u.washington.edu (Gary Forbis) (03/13/91)
The moderator writes: >[Again: You are talking about two distinct phenomena: > (a) [shades of blue] a machine with a specific, designed, function, > performs that function on something slightly different than intended; > (b) [gray goo] due to an error in copying, a plan for a ten-thousand > part machine which does one specific function and runs on one specific > fuel, becomes a plan for a 100 million part machine, able to perform > hundreds of functions, recognize the circumstances under which each > should be performed, run on a wide variety of naturally occuring > energy sources, and survive the chemical attacks of the natural, > highly adaptable microorganisms it will compete with. I'm not sure I see that much difference between the two cases. Be that as it may, I have deleted most of the following paragraph so I can focus on the specific assertions which cause me trouble. > Almost all the copying errors in > a cell are detrimental, i.e. they make it work less well. I'm not so sure about this. I seem to remember that from parent to child there is usually a cross-over error on at least one chromosome. I don't remember how many transcription errors exist. I doubt there would be any life if these errors were that detrimental. Cells get around this problem by redundant code and huge portions of garbage code. The garbage code is probably created by ill-placed cross-overs. > A tiny fraction > improve, or simply change, its function. Almost all of even that > tiny fraction, in a mechanical design, would simply cause it not > to work at all. This may be true with today's designs when thought of at some levels but is not necessarily true. One of the niffty ideas of the 19th century was the mass-produced replaceable part. This was accomplished by tolerance specifications. We are allowed to have slop during the manufacturing process as long as it is kept within limits. Some replacement car parts will work in several places (becuase their design takes advantage of tolerances) where the originals are not interchangable. > Try changing a bubblesort program into a heapsort, > one character at a time, with the constraint that each intermediate > form not only sorts correctly but does so at least as well as bubblesort. This is easy provided comment lines are indicated by a single character. Simply grow the replacement procedure after the existing procedure and a coment prior to the existing procedure which will cause a branch when activated then activate the branch. There are other ways of doing this but this is the easiest. Burroughs B300 assembler has labels which can be addressed by relative position, that is, +a will take you to the next a label and -a will take you to the last a label. I wouldn't be surprised to learn more recent languages have this feature. I have had many programs malfunction in wonderous ways yet be syntatically correct. I leave it to you to see how a single bit error or card misread could cause this problem. Machine language is usually quit densely coded. A single bit error might turn the IBM PC machine code for jmp into jnz or mov c,d into mov d,d. The MS-DOS loader does not contain error detection/correction. It is not clear to me that any particular bit error will even be executed let alone cause a hard failure. I am a systems analyst/applications programmer. On occassion I will encounter program bugs in production programs which have existed for a decade or more. Some system changes will cause existing programs to malfunction because the data values and abstractions move beyond the original scope. As systems age they become more complex and their behavior becomes harder to predict. I would hate to be the person who has to claim any specific system cannot mutate through random processes. I hope it is clear that I am not a comfortable as you are on this issue. > --JoSH] --gary forbis [Far be it from me to comfort you against your will... I'll just point out a couple of things: the human genome is in fact full of "comments" in which errors can occur and have no effect, and it is also quite redundant. It is easy to design our nanobots without either feature. The other observation is that "evolving" a whole new sorting program under cover of comment and "switching" to it at the last minute has no feedback to guide the evolutionary process--which means that it has the same chance of happening as the program changing wholesale, in a single random event, to the given end state. If this were the way evolution actually worked, I would be a fundamentalist. --JoSH]
markb@agora.rain.com (Mark Biggar) (03/13/91)
I'm not very worried about a nano-machine mutating into grey goo. I'm much more worried about a nano-machine designed to destroy dioxins in the water supply mutating into a machine that goes after some very simular chemical in my cells that I need to stay alive. Small organic toxins can be very simular to other necessary chemicals, but just different enough to foul up the works. In fact that is usualy why they are toxic. Note that all you need is a possibly simple mutation in the sensory part of the nano-machine to get this problem. -- Mark Biggar markb@agora.rain.com [I think this is a much more well-founded concern than some others we've heard. It points out the *extreme* dangers in trying to do large-scale environmental engineering with nanotechnology. --JoSH]
toms@fcs260c2.ncifcrf.gov (Tom Schneider) (03/13/91)
In article <Mar.8.16.32.40.1991.12453@athos.rutgers.edu> bill@braille.uwo.ca (W.B. Carss) writes: > How do we control what would essentially be mechanical errors? The same way that we reduce errors in communications systems and computers: error checking and correcting codes. See: @article{Schneider.ccmm, author = "T. D. Schneider", title = "Theory of Molecular Machines. {I. Channel} Capacity of Molecular Machines", journal = "J. Theor. Biol.", volume = "148", number = "1", pages = "83-123", year = 1991} @article{Schneider.edmm, author = "T. D. Schneider", title = "Theory of Molecular Machines. {II. Energy} Dissipation from Molecular Machines", journal = "J. Theor. Biol.", volume = "148", number = "1", pages = "125-137", year = 1991} >Bill Carss >bill@braille.uwo.ca (Kirk Reiser may be reading these in, ask him please.) Tom Schneider National Cancer Institute Laboratory of Mathematical Biology Frederick, Maryland 21702-1201 toms@ncifcrf.gov
peb@uunet.uu.net (Paul Baclaski) (03/13/91)
In article <Mar.8.16.35.15.1991.12515@athos.rutgers.edu>, mmt@client2.DRETOR (Martin Taylor) writes: >...A machine with these desirable abilities > would be in a design space that (at least locally) was rather dense > with viable machines, and the probability of a mutation leading to > a viable design could be appreciably different from zero. Not necessarily. I can see two types of errors that can be controlled using standard engineering: 1. Development errors. This can include errors from reading the genotype, identifying parts and installing parts. The number of degrees of freedom incorrect parts have is directly proportional to the density of the "construction space". Errors such as these can be mitigated through self testing and through validation-- using multiple nanomachines that test each other such that reproduction is not possible if a machine is not validated. Such validation would certainly slow reproduction, but it creates a nice fail safe link such that two or more machines must fail for a failure to be successful. Further, since self-test is difficult, tests using multiple machines would have more flexibility. 2. Genotype transcribing errors. This corresponds to the "dense design space" in the quoted message above. It is tempting that this be a dense space with continuously varying genes-- this is a design for evolvability. On the other hand, the genotype "Turing Machine Tape" does not need to have this characteristic and it can have checksums to validate that it never mutates successfully. Developmental errors must be avoided by using design discipline to test created machines. Genotype errors can be made arbitrarily small by checksums. Genotypes that are designed for evolvability (continuously varying genes that map to continuums in the phenotype) are to be avoided or at least used very carefully (this is what Martin Taylor probably means by "locally dense"). In previous Gray Goo discussions, the conclusion has often been that gray goo will probably not occur accidently--it requires a (malicious) designer. For background, there are two relevant articles in _Artificial Life_, Chris Langston, ed., Addison-Wesley, 1989: The Evolution of Evolvability by Richard Dawkins and Biological and Nanomechanical Systems: Contrasts in Evolutionary Capacity by Eric Drexler. Paul E. Baclaski Autodesk, Inc. peb@autodesk.com
cphoenix@csli.stanford.edu (Chris Phoenix) (03/14/91)
Seems to me that some people have a basic confusion here, perhaps caused in part by the nanomachine/cell parallels that are drawn to show nanotech is possible. People see cells replicating, and cells mutating. People are told that nanomachines will replicate, and so they wonder if they also will mutate. The replication process will be totally different, and this is the key to preventing mutations. When I started to write this I thought I could prove that it was possible to prevent mutations, but now I realize I don't know enough to prove it. I've even managed to unconvince myself. But at any rate, I hope this article will remove some of the red-herring of cellular type mutation. In a cell, chemicals float around in water, bump into each other randomly, and cause changes. For example, producing other chemicals. For example, copying DNA. The process is essentially highly parallel, with no controls except feedback caused by chemicals changing some parameter in the cell. For example, a chemical may "turn on" a section of DNA which produces another chemical which catalyzes a reaction which ... and the desired end product has the ability to deactivate the first chemical, so it's a self-limiting process. Since the cell has so many feedback loops and so many things happening in parallel, if something is changed it has a good chance of leaving the cell alive. If a mutation to the DNA does not kill the cell outright, there are error correction processes; but the error-correction, like everything else, is dependent on chemicals bumping into the DNA at the right time. So it's possible for a change in the DNA to occur, the cell to remain viable, and the change to go uncorrected. This is mutation. (If this is wrong, please correct me; if it's oversimplified, please don't bother.) Picture the following nanomachine, designed to prevent mutation: *everything* will be under the control of one or more computers. If these computers don't like what they see, they can shut down the machine permanently. If they choose not to copy the machine, they won't. And the copying process will also be under their direct control. The nanomachine can't turn itself on--that has to be done from outside. When a nanomachine is copied, it sends the contents of all its computer programs back to the original. If the original verifies that the program is correct, it can turn the copy on. Otherwise, it won't. A machine under total computer control is probably the easiest kind to build, anyway. Now let's consider two kinds of mutation: A change in computer memory, and any other "hardware" change. There is one parallel between cells and nanomachines that I'm willing to leave in: computer memory corresponds to DNA. It contains all the instructions for running and replicating the machine. However, computer memory has one feature that the DNA doesn't: it has a computer. The computer can manipulate the memory far more easily and reliably than the cell can manipulate the DNA. It can store many copies of it, can do calculations on it, and can compare large chunks of it. As far as I know, cells can't do any of these. I know DNA is "redundant", but there are still only two copies of any given chromosome, and the copies are different, and there is no way to compare the chromosomes anyway. A computer can store enough information about its memory, and do enough checking of its memory, that virtually any error in the memory can be detected. I don't know enough theory to do the calculations, but I think it should be relatively easy to ensure that the probability of any undetected memory error in any nanomachine that will ever be created is less than the probability of <insert catastrophic event here>. Can any information-theory people confirm this? If it's true, then mutation of the kind that cells do is impossible. Now, let's consider non-memory changes. This is where I started to wonder if nanomachines could mutate after all. A hardware change should *not* be duplicated when the machine copies itself. It will change the behavior of the original, but will not be transmitted to the copy. This is the theory, anyway. But this is where I started to wonder. It may be possible to change the copying hardware in a way that causes mistakes in the copy, but the change itself is undetectable to the original. In this case, the mistakes might be missed. I think this is unlikely, because the "copying" hardware will probably be a large part of the machine and will probably be used for many other things as well. But consider: a change in the precision of a manipulator arm might cause very few errors, and it's possible that the only error made in copying would be to reduce the precision of the copy's manipulator arm... Although computers can "see" all their internal state, they are dependent on sensors to "see" the outside world. How does a computer know if its dioxin-detector is detecting the right molecule? Well, it has to look at the molecule. With what? With a dioxin-detector... Now suppose the dioxin-detector uses the same arm that is used in duplicating the machine. One mistake in the hardware could be both self-perpetuating and dangerous, and the computer wouldn't have anything to do with it, and couldn't detect it. Now we're getting into nanomachine engineering. The problem, it seems, has come down to this: Is it possible to build a copying mechanism which will have detectable errors whenever it is broken enough to make even slightly imperfect copies? When I put it that way, I get worried...
toms@fcs260c2.ncifcrf.gov (Tom Schneider) (03/16/91)
In article <Mar.13.19.09.22.1991.10983@athos.rutgers.edu> cphoenix@csli.stanford.edu (Chris Phoenix) writes: >Picture the following nanomachine, designed to prevent mutation: >*everything* will be under the control of one or more computers. ... What you are constructing is a way to make error correcting codes. Shannon showed many years ago that it is possible to construct codes that reduce the error rate to as low as you may desire. This stunning result is still not well appreciated by communications engineers. (I have a recent book in which it is incorrectly stated.) Basically it goes like this. If you want a communications line which runs at (say) 10^6 bits per second with one error in 10^5, it can be built. If instead you insist on 1 in 10^10 with the same data rate, sure that can be built. Well! You need 1 in 10^20? sure! And so on! HOWEVER the price you must pay is that you must encode the signal before transmission and decode it afterward. There will be delays in these operations. Actually, you can do this only so long as the data rate is below a certain level called the channel capacity which depends on the power absorbed by the receiver, the thermal noise and the bandwidth. If you go above the channel capacity, you'll get lots of errors that force you back (at least) to the channel capacity. How does this apply to nanotech? What we need to do is make a correlation between the little molecular machines and Shannon's mathematics. I did that in the papers I mentioned previously (JTB 148:83-123,125-137,1991). The translation is a bit bizarre from a biologists viewpoint: so long as the machine capacity is not exceeded, the error rate may be as low as is necessary for survival of the organism the machine is part of. ("desire" has no meaning in evolutionary biology.) Shannon's theorem shows that you can get the mutation rate as low as you might want, but you can't make it zero. (This is based on the assumption that there is white gaussian noise affecting the machine, so if you can get around that, you could beat the capacity.) So what it will come down to in the end is we will have to decide how likely we want errors to be, and then pay the design and material costs to get there. Encoders and decoders are not free. Tom Schneider National Cancer Institute Laboratory of Mathematical Biology Frederick, Maryland 21702-1201 toms@ncifcrf.gov
john@granada.mit.edu (John Olson) (03/16/91)
In article 992, Chris Phoenix writes: >that the DNA doesn't: it has a computer. The computer can manipulate >the memory far more easily and reliably than the cell can manipulate >the DNA. I am not convinced of this. My understanding is that DNA replication is very reliable. Keep in mind the vast information content of DNA, and the rarity of mutations. The August 1988 Scientific American had an article by Radman and Wagner, "The High Fidelity of DNA Duplication," on this topic. They say that DNA is duplicated with an error rate of about 1 error per ten billion (10^10) base pairs. A comparable number for the exabyte tape backup system here would be 1 error per trillion (10^12) bits. That's only a factor of 100, and each base carries two bits worth of data (four states for a base, vs. two states for a bit). How do these error rates compare to, say, the rates for reading RAM or ROM? Someone out there can probably tell us. John Olson.
bsmart@bsmart.tti.com (Bsmart) (03/16/91)
In article <Mar.13.19.09.22.1991.10983@athos.rutgers.edu>, cphoenix@csli.stanford.edu (Chris Phoenix) writes: > replicating the machine. However, computer memory has one feature > that the DNA doesn't: it has a computer. I think part of the problem is that the point of nanomachines is that they operate at the molecular (or even submolecular) level, and are presumably implemented on a similar scale. Nanocomputers probably won't work like the digital electronic computers we're familiar with today; they'll be mechanical devices that represent logic states by chemical composition, physical arrangement of their components, or some such trick. Now, there's no reason why a mechanical computer can't be digital in its operation (every now and then Marvin Gardner presents some delightful macro-scale mechanical computers in Scientific American; one of my favorites was an enormous hypothetical contraption composed of ropes and pulleys and operated by grunting, sweating teams of slaves) but the fact remains that computers (or any other kind of gadgetry) implemented on a molecular scale will be subject to chemical and radiation interference at the same scale. Just as old-fashioned core memory (does anybody still remember?) represented information by imposing different magnetic charges on iron doughnuts, nanocomputers will most likely represent bits by moving an oxygen atom (or something) from HERE to THERE in their structure, and "execute" their "programs" by folding and unfolding and cleaving and binding -- kinda the way natural proteins do. If a mutagenic chemical or other influence came along and moved the atom elsewhere, the result would be pretty similar to what happens in an electronic computer in response to a voltage surge or a stray cosmic-ray hit. Perhaps the error could be corrected by some kind of detection scheme (we do it with memory errors all the time on the macro scale) but if the error were severe enough, or if it happened to hit the error-correcting part of the nanocomputer, then on come the red lights and it no longer behaves like a computer. Even a computer has to be made out of something, and when you're working with a device composed of only a few hundreds or even a few thousands of atoms, you have to live with some constraints and failure modes that just don't apply on larger scales. My concept of a nanocomputer is that it's basically some exotic and complicated chemical compound -- or maybe a soup containing several such carefully constructed compounds. Perhaps structures on a somewhat larger scale (bigger than molecular, but not bigger than cellular) would fall into the "nanotechnology" realm as well, but even these would be dependent upon very small effects for their operations. Discussion, gentlebeings? [The current tentative designs for assemblers and nano-robots with nanocomputer "brains" are on the order of billions, even up to a trillion, of atoms. Even proteins, which are very special-purpose machines, run into the thousands of amino acids, at a handful of atoms each. If an alpha particle came whiffling through a nanocomputer of the design Eric Drexler has talked about with the rod logic, I would imagine it sould seize up and not work at all, so many random bonds would form between closely-spaced moving parts. This is one more reason expect solid-state, i.e. no moving parts, designs to be used wherever possible--alphas will still give you transient errors (a la DRAM) but wouldn't trash the whole machine. Your nanocomputer would have to reboot, though. Hopefully we can make necessarily moving parts (arms) with tolerances and geometries such that the bonds we put there, reform after ionization. --JoSH]
cphoenix@csli.stanford.edu (Chris Phoenix) (03/25/91)
In article <Mar.15.22.38.00.1991.16995@athos.rutgers.edu> john@granada.mit.edu (John Olson) writes: >In article 992, Chris Phoenix writes: >>that the DNA doesn't: it has a computer. The computer can manipulate >>the memory far more easily and reliably than the cell can manipulate >>the DNA. > >I am not convinced of this. My understanding is that DNA replication is >very reliable. Keep in mind the vast information content of DNA, and the >rarity of mutations. Perhaps I shouldn't have said "reliably". However, my main point was "manipulate". Computers can do things to their memory that cells simply can't do to their DNA. For example, perform arbitrary Turing-computable computations on it, and compare arbitrary different parts of it. This is what allows computers to be designed with "mutation" rates as low as you want. And this was the main thrust of half of my previous article--that we don't have to worry about software mutation in a well-designed nanomachine.
cphoenix@csli.stanford.edu (Chris Phoenix) (03/25/91)
In article <Mar.15.23.15.59.1991.17434@athos.rutgers.edu> bsmart@bsmart.tti.com (Bsmart) writes: > >In article <Mar.13.19.09.22.1991.10983@athos.rutgers.edu>, >cphoenix@csli.stanford.edu (Chris Phoenix) writes: >> replicating the machine. However, computer memory has one feature >> that the DNA doesn't: it has a computer. > >I think part of the problem is that the point of nanomachines is that >they operate at the molecular (or even submolecular) level, and are >presumably implemented on a similar scale. Nanocomputers probably won't >work like the digital electronic computers we're familiar with today; ... > Now, there's no reason why a mechanical computer can't be >digital in its operation ... I guess I should have thought more about the computer I was talking about. Yes, I was assuming a digital computer, with distinct memory and processor. I think Drexler's rod-logic computer qualifies. I wasn't saying "nanomachines will not mutate," but rather "It's possible to build a nanomachine which will not mutate as a cell does"--that is, the "genetic material" (computer memory) will not make undetected changes. But maybe I should have thought more about the processor... > ... but the fact remains that computers (or any other kind of gadgetry) >implemented on a molecular scale will be subject to chemical and >radiation interference at the same scale. Notice, I said *undetected* changes. I don't deny errors will happen--what I was trying to point out is that if they happen, you can almost certainly detect them. And you can make that "almost" as close to "always" as you want. >Perhaps the >error could be corrected by some kind of detection scheme (we do it with >memory errors all the time on the macro scale) but if the error were >severe enough, or if it happened to hit the error-correcting part of the >nanocomputer, then on come the red lights and it no longer behaves like >a computer. In my scheme, "on come the red lights" means "the nanomachine turns itself off." This is the best scenario. The one we're trying to avoid is where there's an error and the red lights don't come on. We both need to think about what "the error-correcting part of the nanocomputer" means. I can see two possibilities: 1) a certain area of memory; 2) a certain part of the processor. I don't think 1) needs any special consideration. Consider a correcting scheme in which everything in memory is stored three times. The computer reads all three locations each time it wants a byte, and if they disagree it turns off the machine. Now, which of the three copies is the "error-correcting part?" In other words, there needn't be a "critical" part of memory such that if it's damaged the scheme won't detect the error. 2) may be more worrysome. I would almost class this under the "hardware errors" that I talked about in my first post--the ones that don't involve "mutation" of the "information" but can still cause problems. Here we're getting into areas I don't know about, like circuit testing. The question is whether it's possible for a computer's hardware to fail so that it makes mistakes, but it doesn't "know" it's faulty, and the fault can't be detected from outside. Currently, CPUs are hitting the market that have errors in them--I think I read about one that would crash if it tried to read the last 36 bytes of a segment! How do you test for an error like that? And what if the only code that filled a segment was the machine-copying code? Again, anyone who knows enough, please fill in! I find it really unlikely that such an error would be self-propagating, ie that a hardware bug in the computer could cause copies to be built with exactly the same hardware bug in the computer. (The case of the sloppy arm building another sloppy arm is probably more likely, because there the cause and effect are both mechanical, whereas with the computer error the effect is still mechanical but the cause is a logical error in running a program--the only way such a bug can propagate is by changing the execution of a program.) But I'm starting to handwave here, so I'll stop.
opus@triton.unm.edu (UseCondomsFight AIDS) (03/25/91)
In article <Mar.13.19.09.22.1991.10983@athos.rutgers.edu> cphoenix@csli.stanford.edu (Chris Phoenix) writes: > > >Picture the following nanomachine, designed to prevent mutation: >*everything* will be under the control of one or more computers. If >these computers don't like what they see, they can shut down the >machine permanently. If they choose not to copy the machine, they >won't. And the copying process will also be under their direct >control. The nanomachine can't turn itself on--that has to be done >from outside. When a nanomachine is copied, it sends the contents of >all its computer programs back to the original. If the original >verifies that the program is correct, it can turn the copy on. >Otherwise, it won't. A machine under total computer control is >probably the easiest kind to build, anyway. This brings up an interesting point... Most likely a privately owned company will be producing a particular nanobot for a particular application. Now say that said company finds that the memory (whether it be organic, mechan- ical, or electrical is irrelevent) they will be using will work fine for about a week, but starts to show errors after this time period. The company has already made contractual agreements for other companies to produce this memory. (now assume that the nanobot is too complicated to be self-replica- ting, or that the FDA has not yet approved self-replicating nanotechnology) And if the nanobots have to stop operating after a week, but that the nanobot must work for atleast two weeks. Assume that the company has already made public statments about it's new nanobot. Instead of possibly breaking the job that the nanobot does and having to create two versions of the nanobot (greatly increasing the price), the company goes ahead with production. This scenario has happened in other technologies too many times to count. If this fictional nanobot were used to clean the teeth of dentists patients, could it possibly have it's memory corrupted and try to clean the brian? ------------------------------------------------------------------------------- Institute for Combat Arms and Tatics - System programmer MIDCO - Stereosyntatic Neurosurgery - System programmer opus@triton.unm.edu jkray@bootes.unm.edu ------------------------------------------------------------------------------- [Most computers use memory that starts to show errors after about a millisecond. So they "refresh" the memory periodically. Any well understood phenomenon can simply be taken into account in an engineering design. What you have described above would constitute incompetent engineering, and the company would be out of business very soon. Unless propped up by the government... --JoSH]
mike@everexn.com (Mike Higgins) (03/25/91)
In <Mar.15.22.38.00.1991.16995@athos.rutgers.edu> john@granada.mit.edu (John Olson) writes: > . . . My understanding is that DNA replication is >very reliable. Keep in mind the vast information content of DNA, and the >rarity of mutations. > . . . DNA is duplicated with an error rate of about 1 error per ten >billion (10^10) base pairs. > . . . How do these error rates >compare to, say, the rates for reading RAM or ROM? Someone out >there can probably tell us. I'm told that hard disk drives make errors at about 1 per 10^10 to 10^12 bits. So a hard disk is as good or better than DNA, and RAM is MUCH BETTER! Consider: on a typical IBM AT computer, you have one megabyte, or 10^6 bits. These bits are distructively read and re-written (inside the chip) once every 4 miliseconds for the normal dynamic RAM refresh! That's 10^6 bits refreshed 10^9th times a year. You would be very upset if your PC gave you 1 parity error a year, but that represents an error rate of only 1 error every 10^15th! DNA doesn't light a candle to the box on my desk... Mike Higgins mike@everexn.com
landman@eng.sun.com (Howard A. Landman) (03/25/91)
In article <Mar.13.19.09.22.1991.10983@athos.rutgers.edu> cphoenix@csli.stanford.edu (Chris Phoenix) writes: >Picture the following nanomachine, designed to prevent mutation: I don't think anyone can have any serious argument with the notion that it is physically possible to design machines which can build useful items but have no chance of reproducing themselves. Your average high-school wood shop, if you imagine it being run by a computer, is as good an example as any. QED. However, people seem to be assuming that this completely settles the question. I don't think so. It doesn't address the "argument from practicality". Consider: we know how to build computers which are pretty much impervious to attacks by viruses. Yet not only don't we do so, we often don't even build systems that make use of the built-in protection available in their own microprocessors (e.g., in the Macintosh OS, every program is run in privileged mode). Why? Because it's "faster" or "cheaper". So (for the sake of argument) I claim that when people first start doing nanotech it will be hard enough just to get the damn stuff to work at all, and few if any will be concerned about making absolutely sure that nothing can go wrong. Some of the design problems will be so hard that we will use evolution (the physical equivalent of "genetic algorithms") to solve them. If you don't believe this, consider that there are already people developing artificial antibodies this way. And some of the methods for creating large systems via tiny operators might more profitably make use of reproduction and morphogenesis based on simple local rules, than of global direction by a Master Control Program with all the massive communication and coordination that implies. In the first case (evolution as a design technique), you can't (by definition) turn off evolution. In the second, since the end product is based on a few simple rules (which it is cheaper NOT to place under strict error correction control), a mutation in the rules could lead to a VERY different end product, which has some potential for altering the reproductive viability of the replicators. None of these things are very likely to create a problem. But as nanotech spreads, it will be used more and more often, by a broader variety of people in a more motley collection of settings, and the cumulative probability of SOME disaster happening SOMEWHERE will eventually approach unity. Note that it is wholly inadequate to counterargue that we "can" do this or that to prevent problems. You must argue that people "will" do this or that, even under pressures of schedule, budget, politics, war, etc. This is a much harder argument and I haven't seen anyone attempt it yet. -- Howard A. Landman landman@eng.sun.com -or- sun!landman
lovejoy@alc.com (Alan Lovejoy) (03/27/91)
In article <Mar.24.18.57.41.1991.897@athos.rutgers.edu> landman@eng.sun.com (Howard A. Landman) writes: >Note that it is wholly inadequate to counterargue that we "can" do >this or that to prevent problems. You must argue that people "will" >do this or that, even under pressures of schedule, budget, politics, >war, etc. This is a much harder argument and I haven't seen anyone >attempt it yet. An excellent argument that hits an unarguable bulls-eye at precisely the right target. Before one learns to run, one must first master walking. Before one lears to walk, one must first master crawling. And so on. This argues that we should strive for extremely tight safety procedures and secure environments for nanotech experimentation and practical usage that the FDA or its equivalent will obsessively require until such time as we have mastered a practical (i.e., usable, cheap, fast, simple, reliable and accepted) method (methodology?) for keeping our nanotech fire under control. Safe nanotechnology is like safe sex. Until the safety technology is so user-friendly that no one would consider doing things the unsafe way, we had better assume that people will be tempted to do things the unsafe way. Given the fact that the nanotech genie cannot be kept in its bottle indefinitely--no matter what precautions we take--it behooves us to make viable safety methodology and technology our number one development priority. Another priority issue is this: at what point does the current situation with respect to nanotech research and experimentation--where there are essentially no controls or safeguards--become significantly dangerous? That time is arguably decades away. But perhaps there are those who feel an even greater urgency? (Genetic engineering is admittedly already a potential problem in the case of malicious or intentionally-inimical activities). -- %%%% Alan Lovejoy %%%% | "Do not go gentle into that good night, % Ascent Logic Corp. % | Old age should burn and rave at the close of the day; UUCP: lovejoy@alc.com | Rage, rage at the dying of the light!" -- Dylan Thomas __Disclaimer: I do not speak for Ascent Logic Corp.; they do not speak for me! [There is one very serious danger to this approach: If responsible people inhibit themselves too much in developing nanotechnology, irresponsible people are certain to beat them to Breakthrough, and all hell will be out for noon. There are no simple black-and-white issues here. --JoSH]