Foresight Institute Logo
Image of nano

Study finds self-replicating nanomachines feasible

As reported in Smalltimes, a study done for NASA's Institute for Advanced Concepts by General Dynamics Advanced Information Systems concludes that a useful self-replicating machine could be less complex than a Pentium IV chip, and uncovered no road blocks to extending macroscale systems to microscale and then to nanoscale self-replicating systems. The study also evaluated adherence to the Foresight Guidelines on Molecular Nanotechnology. The final report for the study can be downloaded from NASA as a PDF file.

The Principal Investigator for the study was Tihamer Toth-Fejel, with consultants Robert Freitas and Matt Moses. Their abstract:

General Dynamics Advanced Information Systems has completed preliminary design and modeling studies for the design of a useful Self-Replicating System (SRS). As shown by NASA?s summer study Advanced Automation for Space Missions and other smaller studies, the development of SRSs that constitutes a Universal Constructor (UC) could revolutionize future space missions. Using solar power and in situ resources, a self-replicating lunar factory could build solar cells and other manufactured tools with which to explore and develop the Moon and other extraterrestrial environments with limited exponential growth. But despite the fact that these studies showed the tremendous power of machine self-replication, there have been no large-scale attempts to advance the technology to even the demonstration stage. Before this (2003-2004) NIAC funding cycle, only two small efforts have ever resulted in any non-trivial success in the physical world. Both of the designs lacked significant functionality compared to an autonomous lunar factory, but they proved that machine self-replication is possible. The next step is to make self-replicating machines useful.

This report describes the progress made in that direction, specifically the design of a system of Kinematic Cellular Automata (KCA) cells that are configured as a limited implementation of a Universal Constructor.

Trivial self-replication is not difficult, but a final goal of autonomous and autotrophic self-replication is certain to be extremely difficult (and possibly not desirable). There is a large unexplored area between these two extremes that could still be characterized, but this project advanced our knowledge enough to complete a preliminary design of a physical KCA SRS.

With the advent of nanotechnology, "self-replication is widely viewed as the key to an entirely new industrial era that may one day replace modern microelectronic systems." In addition, self-replicating nanotechnology would reduce the price of complex manufactured goods to that of agricultural products.

Given such potential, it is incredible that half a century after John Von Neumann's seminal work, and 23 years after NASA's summer study, we do not have a single useful SRS. Even worse, we had little idea of the how to quantify the difficulty of useful machine self-replication – until this project attempted to answer that question.

The answer was surprising: The complexity of a useful KCA SRS is less than that of a Pentium IV.

Among the investigators' conclusions:

The expectation at the beginning of this project was that there would be difficulties in designing a KCA SRS … but the important and surprising result was that a small project of this scope could find a fairly clear and successful design with no roadblocks! …

The next logical step would be to build microscale KCA systems made with standard MEMS techniques. …

After that, the final stage of KCA SRS research will be to refine the concept to take advantage of nanoscale parts available at that time.

The Smalltimes story emphasized the attention paid to safety in the design study.

The study also examined machine designs that would meet guidelines established by the California-based nanotech think-tank Foresight Institute to ensure the safety of self-replication techniques. The preliminary study is believed to be among the first U.S.-sponsored studies on self-replication in two decades.

"While self-replication is not necessary for achieving the goal of molecular manufacturing, it's good to see that these NASA-funded system designs are in compliance with the Foresight Guidelines safety recommendations," said Christine Peterson, president of the Foresight Institute.

9 Responses to “Study finds self-replicating nanomachines feasible”

  1. Anonymous Coward Says:

    The study appears fatally flawed.

    I searched the NASA report in vain for the words "cull," "damage," and "apoptosis." In the absence of the overarching concept of damage control typified by such terms I beleive the study is a useless exercise in self-delusion. Your mileage may vary.

  2. Anonymous Coward Says:

    It's a reasonable paper

    Well survey of the field and their conclusions are reasonable.

  3. The Living Fractal Says:

    Re:The study appears fatally flawed.

    I don't understand. Are you saying that damage control, and words therein related, must exist within a feasability study concerned only with technical aspects of just getting the nanomachines built in the first place?

    I see no required connection, myself.

  4. Anonymous Coward Says:

    Re:The study appears fatally flawed.

    If you don't see the connection, then you do not understand the problem space. The problem space is rife with concepts such as damage, interference, error, rework, repair, apoptosis, senescence, quality assurance, quality control, and on an on. If you squint your eyes and look sideways you can manage not to see these real-world considerations, and that is the tack the authors of the study have taken.

  5. qftconnor Says:


    The phenomena which you mention all have natural timescales associated with them: damage rates, error rates, senescence, etc. The self-replication process itself has a characteristic time, namely, the time required for replication to occur (and for the newborn replicator to grow to maturity, if it is not produced in a replication-competent state). So the question is, are any of the failure rates "fast" compared to the replication time? If not, one usually can ignore them, at least to a first approximation.

    My impression from skimming the paper is that that's what the authors did: there was no compelling reason to think failure (whatever its probability distribution) was fast compared to replication, so they didn't worry about it. For a preliminary study – as opposed to a final design – that's not unreasonable. Obviously, subsequent studies should address the issues you raise. But to insist that the original study is without merit ("fatally flawed") because of this omission, without pointing out a single concete example of something likely to fail so fast that it could be expected to kill most replicators before they replicate, is unreasonable, and suggests prior bias.

  6. Anonymous Coward Says:


    Have you ever taken a detailed look at any manufacturing process? The concepts of yield, failure, rework, and discard are inherent to any real-world manufacturing process. Have you ever taken a detailed look at embryology? The concept of apoptosis is inherent to the buidling of any multicellular organism. But go ahead, squint your eyes and picture yourself in a world where such concepts will not impede your fantasies. Knock yourself out.

  7. qftconnor Says:


    Have you ever taken a detailed look at any manufacturing process?

    No, not really. This is similar to but distinct from a manufacturing process: for example, at least at this point, we are not concerned with the economics of the self-replication process. This has many ramifications: for example, rejection rates become less important (you don't necessarily care if the replicator make 99 duds for every functional daughter replicator if you don't care about the cost, providing the initial device has a probabilistic lifetime such that on average it will survive long enough to reproduce).

    Have you ever taken a detailed look at embryology?

    I took two semesters of developmental bio as an undergrad. I was near the top of the class first semester, at the top the second. But I think that the biological analogy, while relevant, can be pushed too far. Replicating systems needn't look like multicellular organisms. Recall the Lackner-Wendt proposal of a few years back, for example. ("Exponential growth of large self-reproducing machine systems," by Klaus S. Lackner and C. H. Wendt, Math. Comput. Modelling Vol. 21, No. 10, pages 55-81, 1995.)

    …go ahead, squint your eyes and picture yourself in a world where such concepts will not impede your fantasies.

    I'm more interested in trying to extract the relevant parameters from the sea of possibilities; that's a key part of modelling. You haven't provided any evidence that the issues you raise are of any quantitative importance in this situation. It would be nice if the authors had addressed some of those issues themselves, but again, that's not necessarily a fatal flaw in preliminary work.

  8. Anonymous Says:

    if anything this “kenematic cellular automation” report just adds to previous knowledge in the field. all those importany manufacturing steps you feel are so absent are in this report

    the new report is just another mechanism by which we might accomplish the same thing

  9. Charles Michael Collins Says:

    This technology is infringing on patent 5,764,518. In fact, identical and quite interesting is the fact that an attempt to “bust” the patent is in the works in a book by Freitas (a principle “advisor” to the NIAC, NASA General Dynamics project) at:

    Freitas neglected to piont out that a working model was presented to the patent office before the patent was allowed. Your corrupt government at work!

    Charles Michael Collins

    the Cornell replicator is as well infringing, you can see here at:

Leave a Reply