We’re fascinated with robots because they are reflections of ourselves.–Ken Goldberg
My first publication as a futurist has appeared in the inaugural issue of Age of Robots. It’s based on my interview with Will Mitchell in Seeking Delphi™ podcast #14 and is reproduced below.
Volume 1 Issue 1
Science Fiction VS Science Fact
by Mark Sackler
Science Fiction vs. Science Fact: Replicating Machines
By Mark Sackler
Self-replicating machines have been a staple in science fiction since the 1940’s. A. E. Van Vogt, Philip K. Dick and Arthur C. Clarke, along with many others, have used self-building robots as plot devices. But just how realistic an idea are they?
As far back as 1980. NASA conducted an engineering study of concepts for a self-replicating lunar factory. For decades, the study sat and collected dust. But the concept of robotic explorers, builders, and miners that can land and copy themselves, has come back to the fore. Just how viable is this technology? How far away is it? Are there dangers? Two men who have thought long and hard about this are science fiction author Will Mitchell and space roboticist Dr. Alex Ellery.
FICTION: William Mitchell, Author, Creations
Will Mitchell is an aerospace engineer who writes hard science fiction. The key word is “hard.” It signifies technical plausibility. And in his 2013 novel Creations, he constructs a chillingly plausible scenario of technology gone wrong. Step by step, he covers the building, deploying, evolution, and finally, destruction, of self-replicating robots on the moon. Surprisingly, though, he is bullish on the use of self-replicating machines for space exploitation and colonization.
MS: What was the inspiration for Creations?
WM: The idea came to me a long, long time ago. I was reading the novel 2010 by Arthur C. Clarke. I was in my early teens at the time. At one point in the novel, the black monolith around Jupiter is compared to a self-replicating von Nuemann machine. At that point, I’d never even heard of the idea, or considered it as a possibility. This got me wondering: could you actually build a machine that could copy itself? What would you get if you set it running, say, on the moon. What if you had a vast automated city, populated only by robots, growing and spreading and possibly even evolving.
MS: Just how plausible is this concept?
WM: If you look at this issue, wrongly, you can conclude, wrongly, that it’s not possible at all. The first fallacy we should get past is the notion that a robotic machine can only make something simpler than itself. It’s a very easy fallacy to fall into. If you think of a production line making cars, no matter how many cars you pump out, there’s no way those cars could team together and build another production line. But working back in the 1940’s, John von Neumann, one of the greatest minds of the 20th century, came up with his theory of automata. He looked at the whole thing from a mathematical perspective. He did this, first of all, to confirm that the idea of replicating machines was even possible, but secondly to try to figure out what capabilities and functions it would need in order to duplicate itself. He was able to prove, mathematically, that machine replication is possible.
MS: In the book, your draw deep similarities between biological life and machine life. What can you say about that?
WM: One of the things that confirmed this possibility to von Nuemann is this deep similarity between machine life and biological life. He realized that all life forms, everything from bacteria upwards, they really are a kind of machine, albeit working by chemistry rather than robotics. So that parallel between machine life and biological life is central to the whole thing. This is because another major fallacy that you need to get out of the way is the idea that only biological systems can duplicate themselves. It’s easy to fall into the trap that biological matter has a kind of life force associated with it, as if replication or self-duplication is some kind of mystical process. But really, it isn’t, it just follows the laws of physics.
MS: How does this comparison work?
WM: If you imagine something simple, like a bacterium reproducing, it takes in raw materials from its local environment and performs various functions on those raw materials. It reorganizes them and recombines them according to plans coded in its DNA. What you get is a copy of the original, including its own DNA blue print which allows it to make more copies. When von Nuemann tried to figure out the basic plan for a replicating machine, he realized the exact same kind of function would be needed. You’d need a machine that could take in raw materials, process them and arrange them, using some kind of fabricator. The whole thing would be run by an instruction set recorded on whatever type of storage medium you chose. Von Neumann proved mathematically, that something built like that is not only capable of reproducing, but it’s directly analogous to how biological systems reproduce. It doesn’t matter if it is made of organic matter or metal and plastic, or if it uses DNA or data files, as long as it has the functions and components he identified as being necessary.
MS: Why would we even need these in the first place?
WM: The potential use of this technology is in space exploration. This is where it really will come into its own. We live in a very deep gravity well, and pushing things up that well is very, very expensive. That first hundred miles up are phenomenally difficult and expensive to cross. Once you’re up there, it’s easier. Landing on the moon and taking off again takes a fraction of the fuel it took to get there from earth. The same is true of Mars and its especially true of the asteroids. But they are all very hazardous locations, so ideally you would want to automate any industrial capacity as much as possible. If you want to establish a wider more permanent presence in space without having to ferry every bit of equipment and fuel from earth, you need to develop something self-sustaining. Something that does not need hardware or material to be supplied from earth. A year or so ago, the White House office of science and technology policy started looking for ideas to perform what they called, massless space exploration. The aim is to find the smallest amount of hardware that can be launched into space and then unpack itself to form a self-sustaining mining and manufacturing infrastructure.
MS: What is the biggest technical hurdle yet to be overcome?
WM: The biggest issue is called closure. If you make a list of all the parts a machine is made of, and a second list of everything it can make, everything on that first list must be on the second list. Closure is expressed as a percentage. 100% closure means it can replicate itself. Even 99.999% is not good enough. If there is even one component the thing depends on that it’s unable to make, that’s enough to stop the whole thing. A good example is high grade integrated circuits—processor chips. You need complex clean room facilities to make those. If that type of capacity is included in the replicator design, it must be able to make that chip factory as well. You don’t want to have to supply parts from earth, you want to get to as close to 100% closure as possible.
MS: In the book, things spiral out of control because the machines are allowed to evolve on their own. How does that happen?
WM: In my day job, I’ve used simulated evolution and that’s the career path I’ve given my main character in Creations. He’s the one who appreciates how powerful evolution can be as a design tool, and he’s the one who warns everyone else that what they are not letting loose might not be easy to control. Eventually he’s proven right. The problems arise when the machines that are best at reproducing and surviving take over, not necessarily the ones that are most useful to us.
MS: Yet you are very much in favor of this technology. Do you believe it can be controlled?
WM: The original NASA study considered the possibility of allowing the replicators to evolve. There are some very exciting possibilities from that, because you would get something which is like a form of artificial life which could modify itself and spread much quicker than you could design. But there are also some scary implications to that as well, because if you think you are going to keep full control of these things, you’re probably going to be mistaken. But if NASA did something like this, they would design it to remain under the control of the people who set it running. However, Creations is fiction, and in fiction you need jeopardy, or in this case outright carnage, so for plot purposes some things need to go wrong. In the novel, therefore, the machines designers are grossly negligent in how they oversee them. In reality, though, I’m looking forward to seeing these kinds of replicating machines built, as I feel they will fulfill all the promises people make about them. If it’s done responsibly, I’m an optimist rather than a pessimist.
MS: Creations is set in the year 2040. How do you feel about current progress in this direction vis-à-vis that timeline, including the possible deployment on the moon?
WM: Things have come a long way since the NASA study. They thought in terms of traditional manufacturing methods and not 3D printing. 3D printing, or additive manufacturing, seems like a very viable way of obtaining closure. I think there are still some big hurdles to cross. I think it’s going to be a long time before someone unpacks some robotic seed on the moon and lets it reproduce. What the researchers at Carleton University in Ottawa are doing is a step in the right direction. They are going beyond just the fast prototyping to 3D print an entire electric motor using multiple materials. If they can do that and show that it works, it’s a big step forward. The fundamental research is definitely going in the right direction and going faster than I anticipated. But for the end state it’s possible that I was being optimistic with the 2040 year, but who knows, it may come along and surprise me.
FACT: Alex Ellery, Ph.D, Associate Professor, Space Robotics and Space Technology, Carleton University, Ottowa,
At first glance, the electric motor in the lab of Dr. Alex Ellery looks like something that might have been built as a middle school science project. But in fact, it is the first step toward the ultimate creation of a 3D printer that can replicate itself from materials found on the surface of the moon. Dr. Ellery believes his team may be just a few months away from being able to 3D-print an entire electric motor at a single pass—no other assembly required. And a fully self-replicating machine may closer to reality than Will Mitchell thinks.
MS: Alex, how did you get started in this?
AE: I’ve been aware of self-replicating machines since I was a student. I came across the concept of von Neumann machines when I was studying for a master’s degree in astronomy at the University of Sussex. I became fascinated with idea of self-replicating machines that could populate the galaxy. As a result, I decided to do my Ph.D. in engineering. Up until recently, there was never an opportunity to convert my imagination and fantasy of a self-replicating machine into anything practical.
MS: What was it that change recently? What are you doing now?
AE: The most important event was the advent of the RepRap 3D printer, made by Adrian Bowyer at Bath University. It can print some of its own plastic components, but that’s all it can do, so it is only partially self-replicating. I started thinking that if you want to complete the process, the key components are the motors and electronics. So, to completely be able to do this we need to focus on printing the motors and electronics as well. We are currently building a printer where we can print metal and plastic and mill within the same platform. Also at this time I was working on the Lunar Resource Prospector mission, which is an American mission to send a robot to the moon to drill for water and other resources and to cook it and extract oxygen from it. From this I came up with a restricted material set—iron, nickel, cobalt, tungsten, volatiles, etc.—available on the moon, from which we could construct a machine that can manufacture silicone, oils, and plastics. With all those components, we have enough materials to build motors, electronics and structures.
MS: What about closure—and specifically the fabrication of electronics?
AE: One of the key elements in closure is restricting the elements that you need to access. You have a restricted materials list that you use for everything. It’s how I imagine the native Americans used to use Buffalo. They use it for clothing and ropes and eat it as well, they lived within the restrictions of their environment and didn’t waste anything. We use the same principle. We have a restricted materials list. It may not give you optimal results, but it will certainly work. With the electronics, as Will Mitchell pointed out, you can’t build solid state electronics on the moon. But we do have the materials to build vacuum tubes. We can use tungsten, cobalt, we have glass from silica. So instead of solid-state transistors we can use vacuum tubes.
MS: How would the electronics work with these vacuum tubes?
AE: Well the problem with using vacuum tubes to compute is that the device grows extremely large as you add capacity. Early pre-transistor vacuum tube computers were the size of a small house. To obviate against that I’ve come up with a different computational architecture which isn’t based around a CPU. The CPU uses a Turing machine model. You have the program going in, it churns through its calculations, and out comes the answer. Instead, we implement each program we need as a neural network which we print on demand. In this manner, you can have relatively complex programs in a comparatively small circuit. The 3D printer becomes a Turing machine substitute. So, a store of instructions comes in—you might store it perhaps in a magnetic core memory—that goes into the 3D printer which stamps out the neural circuit.
MS: So how far along is this?
AE: I haven’t demonstrated that we can print the vacuum tubes yet, so we are concentrating on the motor first. The biggest problem so far has been 3D printing the magnets. If we can do that and get that done in the next few weeks, and demonstrate the coil design in multiple layers and don’t get any showstoppers I reckon we can get the first fully 3D printed motor working within a few months. I would like to say that if there are no showstoppers we could have it done by the end of the year.
MS: What about full closure, thoroughly replicating and self-assembling?
AE: Well once we have the motor we are not stopping there. We’ll use genetic algorithms to design new concepts to reduce the assembly requirements. We’re building our own 3D printer. Eventually we’ll replace the off-the-shelf motors with our own motors. Eventually we’ll replace our computer programs to control it with neural networks. We’ll use the restricted materials list to make both the motors and the vacuum tubes.
MS: In Will Mitchell’s novel, Creations, the self-replicating machines are allowed to evolve on their own. Is this something you have thought about?
AE: Yes, it is. I wouldn’t allow them to evolve. What I envisage for the lunar scenario is something called the salt contingency. Salt (NaCl) will be required to produce each new generation of replication. The sodium is needed because you need sodium carbonate for the processors and hydrogen chloride for the plastics. If you deny these materials the replication process stops. Both sodium and chlorine are very rare on the moon, so you need to supply that. The second part of the process is to include EDAC—error detection and correction—in the program code which stops it from evolving. It will increase the size of the memory required by about 30 percent, but by building these codes in it provides an extra layer of protection.
MS: So, salt is the one feedstock that would have to be provided from Earth. I suppose if you wanted to do this on an asteroid or on Mars, a different approach might be required?
AE: Yes, on Mars, in particular, there’s plenty of salt so that safeguard wouldn’t work. You’d have to implement something else.
MS: Creations is set in the year 2040. It envisions a colony of fully self-replicating robots mining the moon in that year. Is that over-optimistic? How far away is the complete self-replication of robots using materials, in situ, on the moon?
AE: Most of the work I’m doing is being done on a shoe-string budget. If I was given proper money I could probably get something working much more quickly. I did some estimates on how quickly you might be able to do this. I think we could have a fully self-replicating machine within 12-years if we had the funding required to really make it happen. It could be done a lot sooner than you think.