Chapter 8 Afterword (2024)
In the four years since this book was originally published in 2020 we have come across several additional examples of early thought about self-reproducing machines. The most significant of these are the works of John Adolphus Etzler in the 1820s–1840s, Frigyes Karinthy (1916), and Fred Stahl (1960). These significant works and other more minor additional examples are presented chronologically below according to the chapter structure of the original book.151
8.1 Additional Material for Chapter 3
Chap. 3 covers the first extended explorations of the idea of self-reproducing machines in the 1800s. The growth of interest in the topic at that time was spurred by two factors: the climax of the British Industrial Revolution in the first decades of that century, and the publication of Darwin’s On the Origin of Species in 1859. While most of the work discussed in the chapter is from the 1860s onward and was influenced by both of these factors, the extract from Benjamin Disraeli’s novel Coningsby quoted at the start of the chapter demonstrates that, even before Darwin introduced his idea of evolution by natural selection, the increasing complexity and manufacturing capabilities of machines emerging in the Industrial Revolution were in themselves sufficient to elicit the idea of a self-reproducing machine in some future thinkers.
Undoubtedly the most significant body of work to be added to the material presented in Chap. 3 is that of John Adolphus Etzler, whose work over the period from the late 1820s to the late 1840s certainly falls into this category of pre-Darwinian Industrial Revolution thinkers.
8.1.1 John Adolphus Etzler: The Paradise Within the Reach of All Men (1833)
ⓘ This section expands upon the introductory text provided in Chap. 3 |
Johann Adolph Etzler (hereafter referred to by the Anglicised form John Adolphus Etzler) was a nineteenth-century German inventor who would be regarded today as a techno-utopian; he believed that machinery could be designed to provide for all human needs, abolishing the need for work and money, and leaving people free to pursue their own interests and pleasures.
Despite attracting a significant following during his lifetime, particularly in Great Britain and Germany, Etzler’s work has languished in obscurity after his death (the date of which is unknown). Aside from his own writing (e.g. (Etzler, 1833), (Etzler, 1841)), the most accessible account of his life and work until very recently has been a generally unflattering account presented in American historian Steven Stoll’s book The Great Delusion: A Mad Inventor, Death in the Tropics, and the Utopian Origins of Economic Growth (Stoll, 2008) published in 2008. However, a 2021 PhD thesis by James McIntyre of Loughborough University presents many previously-unknown primary sources of information about Etzler’s life and work, and challenges various inaccuracies presented in Stoll’s and other earlier accounts (McIntyre, 2021).
Born in 1791, Etzler emigrated to the United States in the 1820s and later lived in various other countries including Great Britain and Venezuela (McIntyre, 2021). After some early experiments with mechanical systems in Pennsylvania in the late 1820s and involvement in an emigration scheme from Germany to the United States to establish a co-operative community assisted by mechanical devices in the early 1830s, Etzler published his most influential work, The Paradise Within the Reach of All Men, in 1833 (Etzler, 1833).152
In Paradise Etzler presents a utopian vision whereby all humans could be provided with a high standard of living and with no need to work. This would leave them free to spend their time as they wish, in learning, culture, socialising and pleasure. All of this, he claimed, could be accomplished within the space of ten years. Etzler’s idea was, rather than being guided by existing artefacts and their means of manufacture, to instead take a holistic view of people’s basic wants and needs and to consider how these might most simply and systematically be fulfilled using a relatively small number of multifunctional machines:
To imitate minutely all the infinite variety of produces of human industry by machineries, would be an endless, ungrateful, and foolish undertaking … It would nearly require to invent for every little work of man a particular automaton. This is not my purpose. But the most simple contrivances I could think of, and as few as possible, for producing, not the customary articles of human industry; but all things that may either substitute or surpass the known necessaries, comforts, and luxuries of men, are my objects in view … My object is, to furnish, by an extremely simple system, all what may be desirable for human life, without taking for pattern any existing things of industry. By abstracting from all what is in existence and fashion, I am enabled to devise means, without any artificial machinery, for producing every thing that man may want for his nurrishment [sic], dwelling, garments, furnitures, and articles of fancy and amusements.
John A. Etzler, The Paradise Within the Reach of All Men, 1833 (Etzler, 1833, p. 62)
Etzler recognised that the fulfilment of his vision would require superabundant sources of energy that were “imperishable, indefatigable, working … day and night, without food or wages” (Etzler, 1841, p. 1) and that did not in themselves require further human labor to extract and use. Rather than employing the typical power sources of the day such as steam, coal or animals, he instead focused on the enormous potential of natural forces: wind, wave, tidal and solar energy. In order to produce a continuous output of energy from these intermittent inputs, he also concentrated on the design of energy reservoirs such as water storage towers to mediate the supply of energy to his machines. His vision was therefore of a post-work world powered by renewable energy.153
In Paradise Etzler gave a brief sketch of the kind of machines he envisioned for agriculture and for architecture. More details of these and other machines were provided in his 1841 publication The New World (Etzler, 1841). The power of machines to easily manufacture multiple copies of their products played an important role in Etzler’s vision. In describing how people may spend their time in his post-work society he says:
Is he fond of mechanical occupation? — He may exercise his dispositions and talents to an extent beyond the present conceptions; he may form models and moulds, and see the objects multiplied for use and show to any extent, without any further trouble. Is he gifted with talents for drawing, painting, sculptures, &c.? — He needs but to make one model of every figure, and it may then be multiplied to any desired number, by moulds, etching and printing machines. Is he fond of music? — Where could he find more opportunity than in such a life? He may at once delight and be delighted, by performances of his own and in company with other musicians: instruments and means are at his disposal unknown yet; and his compositions may be repeated and multiplied by mechanical plays and machines.
John A. Etzler, The Paradise Within the Reach of All Men, 1833 (Etzler, 1833, p. 86)
Having set out in Paradise his vision of what life might be like in a society where all work was performed by machines, Etzler explained how the requisite number of machines might be built and how the endeavour might be financed.
And what is the expense for producing such great things? — None, except for the first machineries of very simple construction, and for the first moulds of all things to be artificially made; for the machineries themselves as well as the moulds for casting the materials for use, are to be made by the same machineries, and may then be multiplied to any number required, without any labor or expense.
John A. Etzler, The Paradise Within the Reach of All Men, 1833 (Etzler, 1833, p. 82)
As this quote demonstrates, Etzler envisaged that these would be self-reproducing machines; specifically, to use the terminology introduced in Chap. 1, they would be maker-replicators (i.e. machines that could manufacture a wide variety of products, including copies of themselves). He went on to present examples of the finance and revenues involved in creating a society supported by these machines, basing his calculations upon an assumed tenfold annual increase in the number of machines in operation due to their capacity for self-reproduction (Etzler, 1833, p. 101).
Etzler argued in Paradise that the most effective way to get such an endeavour off the ground would be to form an association of members, each of whom must buy at least one share in the enterprise. The association would find a suitable location, with a warm climate and an abundance of natural resources, for developing an initial community according to Etzler’s vision. He spent most of the next decade trying to make this happen, in addition to developing designs and models of various proof-of-concept machines.
In 1844 (the same year that Disraeli was writing Coningsby, quoted in Chap. 3), Etzler—in partnership with the anti-slavery campaigner Conrad Frederick Stollmeyer—established the Tropical Emigration Society in London. He had been offered a tract of land by the Venezuelan government and intended this to be the site of his first settlement. Over the next three years the society attracted thousands of paying members across England and sent over 200 of its members, Etzler included, to Venezuela to establish the settlement. Sadly the enterprise ended in tragedy with the deaths of at least 23 of the group (McIntyre, 2021, p. 198). Stoll lays the blame for the venture’s failure squarely on Etzler (Stoll, 2008), although McIntyre more recently convincingly challenges Stoll’s account (McIntyre, 2021). McIntyre uncovers evidence of Etzler in London and Bogot'{a} in the period after the failure of the Venezuelan venture, but the full picture of his latter whereabouts, activities, and death remains unclear.
Etzler’s conception of self-reproduction by mechanical devices equipped with moulds calls to mind William Paley’s earlier image—in his 1802 book Natural Theology which we discussed in Chap. 2—of a self-reproducing watch comprising “a mechanism, a system of parts, a mould for instance, or a complex adjustment of lathes, files, and other tools, evidently and separately calculated for this purpose” (Paley, 1802, p. 11) (see Sect. 2.2). To some extent one might view such a design as an analogue version of John von Neumann’s seminal architecture for a self-reproducing machine, devised over a century later (von Neumann, 1966) and discussed at length in Chap. 5: the mould being an analogue version of the information storage tape, and the machine as a whole being able to make copies of any kind of object defined by the supplied mould, and of the mould itself. It’s unclear how a machine might contain a mould of itself to achieve self-reproduction, but this difficulty might be alleviated by imagining a collection of these machines, each supplied with different moulds to produce different parts of their offspring and with one of more of the machines charged with assembling the various manufactured parts into a complete offspring. The analogy is still only partial, however, as the analogue storage of information in the mould and in the design of the machine itself offer an impoverished potential for heritable mutation compared to the digital information storage, copying and translation processes proposed by von Neumann.
Of course, in 1833 Etzler was not specifically thinking about the potential for mutation and evolution of his machines. It is currently unknown whether Etzler was alive in 1859 to witness the publication of On the Origin of Species. Darwin’s book triggered a heightened interest in the idea of machine self-reproduction, now envisaged with the additional possibilities of mutation and evolution. As discussed in Chap. 3, the most significant work on this topic in the late 1800s was by Samuel Butler, with other notable contributions from George Eliot and Alfred Marshall.
8.2 Additional Material for Chapter 4
As we saw in Chap. 4, by the turn of the twentieth century the topic of machine self-reproduction and evolution was starting to find its way more regularly into popular works of fiction. Examples offered in Chap. 4 include E. M. Forster’s 1909 short story The Machine Stops, and Karel Čapek’s 1920 play R.U.R.: Rossum’s Universal Robots.
Forster’s story, set in a time when human civilization has become dependent upon a global machine to provide for all needs, makes reference to the machine’s capacity for repairing itself and even evolving new functions, but does not explicitly engage with the idea of machine self-reproduction as such.
Čapek’s play, on the other hand, does explore in more depth the idea of robots being able to build more of themselves in factories without human supervision. In the time since the original publication of our book we have become aware of another work from this period that also directly explores the idea of machine self-reproduction, written by the Hungarian author Frigyes Karinthy and published a few years before R.U.R..154
8.2.1 Frigyes Karinthy: Utazás Faremidóba (Voyage to Faremido) (1916)
ⓘ This section belongs after Sect. 4.1.1 of the original text in Chap. 4 |
Born in Budapest in 1887, Frigyes Karinthy was a prominent “humourist, parodist, writer of utopias, poet [and] philosopher” (Tabori, 1978). In addition to his own writing, he also translated a number of important works into Hungarian, including Jonathan Swift’s Gulliver’s Travels (Swift, 1914). In 1916, a couple of years after that translation, Karinthy published his own novel Utazás Faremidóba (Voyage to Faremido) which took the form of a continuation of Gulliver’s adventures (Karinthy, 1916).
Set in 1914 at the onset of World War I, Voyage to Faremido begins as Gulliver, escaping a sinking ship in a hydroplane, is plucked from the plane by a “huge bird-shaped mechanism” from which an enchanting music is emanating. He awakens to find himself in a strange land, soon discovering that he has been transported to a distant planet populated by machines like the one that rescued him. The machines are called solasis (singular: solasi), and Gulliver discerns that the music they play is their form of communication; he eventually learns the language so that he can communicate with them. Nothing is said about the solasis’ ultimate origin, but it turns out that there is a vast factory where they manufacture more of themselves, and spare parts for existing machines as well. Gulliver describes the activities in the factory as follows:
… it became evident how these amazing creatures or mechanisms came into being: they themselves manufactured their equals from metals and minerals, and they themselves activated the finished solasi through the sources of energy (electric accumulators, steam, gases, etc.) placed within their bodies.
At first glance this method of procreation appeared to be more complex and difficult than the one employed on our globe … but it must be admitted that as far as the end product was concerned, the solasis’ system was more reliable and thorough. The solasi who created or assembled its companion—I must call it a companion because I can hardly call it son or child, in view of the fact that each solasi is the creation not of two but of six or seven individuals, and these are all of the same sex—such a solasi had the opportunity of examining every part thoroughly from the point of view of its practicability and of assembling it without the slightest blemish or functional fault.
Frigyes Karinthy, Voyage to Faremido, 1916 (Karinthy, 1978, p. 38)
As the quote above demonstrates, Karinthy employs the idea of collective reproduction accomplished by a group of machines. As discussed in Sect. 3.1, this idea had already been discussed by Samuel Butler in his 1872 novel Erewhon. The collective reproduction described by Karinthy is of a simple homogeneous variety, where each of the machines involved in the process is of essentially the same kind. The is the same type of reproduction as envisaged by Karel Čapek in R.U.R., published four years after Voyage to Faremido. We discussed these and other examples of collective reproduction in more detail, and compared them to more monolithic designs of self-reproducing machines, in Sect. 7.1.4.
As far as we are aware, Karinthy’s novel is the first significant work to explicitly cover the idea of machine self-reproduction in the early twentieth century—indeed the first since George Eliot’s Impressions of Theophrastus Such in 1879 (Eliot, 1879).
8.2.2 Early Pulp Science Fiction (1920s-1950s)
ⓘ This section expands upon Sect. 4.1.3 of the original text in Chap. 4 |
The 1920s saw the birth of the pulp science fiction genre, offering cheap, regularly published magazines containing a vast variety of short stories and novellas.
In Sect. 4.1.3 we identified S. Fowler Wright’s 1929 story Automata (Wright, 1929) as the earliest example of the genre we had found that features as a central topic the idea of self-reproduction (and self-design) of machines. While Automata remains the earliest example of the genre we are aware of that has an explicit focus on the long-term evolution of machines over many eons, we have recently come across an earlier work that has a more implicit suggestion of machine self-reproduction: the American writer Edmond Hamilton’s 1926 short story The Metal Giants (Hamilton, 1926).
Edmond Hamilton: The Metal Giants (1926)
Hamilton’s story features a professor of electro-chemistry named Detmold, who has discovered how to instil an artificial brain with consciousness. The brain was “constructed … of metal, entirely inorganic and lifeless, yet whose atomic structure he claimed was analogous to the atomic structure of a living brain” (Hamilton, 1926, p. 725). Sacked by his university for his outlandish ideas, Detmold continues working on his artificial brain independently. During the following months of development he augments the brain with arms so that it can interact with the world. After a sudden illness resulting in a period of hospitalisation, Detmold returns to his secluded laboratory to find it ransacked and the brain missing. Soon after, he becomes aware of various incidents reported in nearby towns of “metal giants” attacking people and property. Detmold eventually discovers that the brain, which had gained the ability to move by some unknown method, was building these metal giants itself. The giants possessed a degree of autonomous intelligence but were ultimately controlled by the central brain (Hamilton, 1926, pp. 736–737, 861).
So the storyline does not include the self-reproduction of the
artificial brain as such, but it comes very close to the idea—there
is a suggestion that it had the capacity to do this if it
wanted to:
“… while the [metal giants] undoubtedly had been furnished some
portion of intelligence by their master, the metal brain, that
master had been careful not to repeat [Detmold’s] own mistake and make them
powerful enough to revolt against it” (Hamilton, 1926, p. 737).
The self-reproduction evoked here—the
potential of the artificial brain to create offspring as intelligent
as itself—is therefore of a self-designing type, similar to that
more prominently featured in S. Fowler Wright’s Automata three
years later.
In Sect. 4.1.3 we also highlighted various other pulp sci-fi stories published after Automata (in the period from the 1930s to the 1950s) that featured ideas of machine self-reproduction and evolution. We have since found a few more examples to add to this list, as outlined in the following subsections.
Henry Hasse: He Who Shrank (1936)
American author Henry Hasse’s short story He Who Shrank, published in 1936, is a tale of worlds within worlds, where molecules, atoms and electrons at one level of existence comprise the galaxies, systems and planets of a universe below (Hasse, 1936). A scientist creates a potion that makes anyone who takes it continuously shrink, allowing them to travel between these universes. The story relates the experiences of the scientist’s assistant—the narrator—who is given the potion against his will, as he visits successive universes below his starting point. In one universe he discovers a race of “bird people” who have fled their planet and are settling on a moon, building a protective metal shell around it. It turns out that the main planet has been taken over by machines.
He Who Shrank summons an image similar to that described by George Eliot in Impressions of Theophrastus Such (discussed in Sect. 3.3), of a world run by unconscious machines busily constructing vast cities of grotesque metal structures all around the planet. As well as construction, the machines were also making more machines (Hasse, 1936, p. 38). Their activities of construction and self-reproduction left little room on the planet for their original creators.
The narrator surmises that the machines have evolved from technology originally created by the bird people themselves:
I tried to picture their civilization as it had been long ago before this thing had come about. I pictured a civilization in which machinery played a very important part. I pictured the development of this machinery until the time when it relieved them of many tasks. I imagined how they must have designed their machines with more and more intricacy, more and more finesse, until only a few persons were needed in control. And then the great day would come, the supreme day, when mechanical parts would take the place of those few
But it had proven to be a bitter Utopia. They had gone forward blindly and recklessly to achieve it, and unknowingly they had gone a step too far. Somewhere, amid the machines they supposed they had under their control, they were imbued with a spark of intelligence. One of the machines added unto itself—perhaps secretly; built and evolved itself into a terribly efficient unit of inspired intelligence. And guided by that intelligence, other machines were built and came under its control. The rest must have been a matter of course. Revolt and easy victory.
Henry Hasse, He Who Shrank, 1936 (Hasse, 1936, p. 42)
We learn that the machines have achieved space travel, and the narrator wonders whether this planet was not even the bird people’s original home—they had perhaps already been through several rounds of fleeing their machines and the machines then following them from one planet to the next. The narrator wonders whether the bird people will eventually find a way to check their spread, or whether the machines will ultimately occupy every planet in the universe.
He Who Shrank therefore paints a picture of a supercharged version of the takeover by machines envisaged by Samuel Butler, George Eliot and others, where the machines have not only displaced their designers on their home planet but eventually spread across the whole universe. The image summoned in the quote above calls to mind the extract from S. Fowler Wright’s Automata (Chap. 4, p. ), written seven years earlier, about the dangers of humans tinkering with machine self-reproduction when they have insufficient appreciation of the long-term consequences of their actions. The kind of evolution described is a mixture of machine self-reproduction and self-design of new types by a master intelligent machine—like the design of new machines by Hamilton’s artificial brain in The Metal Giants with the additional capacity for self-reproduction and evolution.
Other early pulp sci-fi works (1940s–1960s)
Another work from the period that we have come across since the original publication of our book is The Mechanical Mice by Maurice G. Hugi and Eric Frank Russell (1941) (Hugi, 1941)155. This short story features a “Robot Mother” which sends out workers (mechanical mice) to collect raw materials that it needs to reproduce itself. This system is likened to a bee colony, with workers, warriors, a drone and a queen, much like Samuel Butler, in Erewhon, used analogies of bee/flower pollination systems and ant colonies for collective reproduction in machines (see Sect. 7.1.4).
One more story that deserves a quick mention here is Two-Handed Engine156 by C. L. Moore and Henry Kuttner, published in 1955 (C. L. Moore & Kuttner, 1955). The story is set on a future Earth after human society collapsed in the late twentieth century due to over-reliance on machines resulting in a breakdown of social and emotional bonds between humans. Across the world, humankind had been brought to a state of anarchy. The machines, however, fared better in this period:
… some of their species were wiped out entirely and left no machines to breed and reproduce their kind. But most of them minded their raw materials, refined them, poured and cast the needed parts, made their own fuel, repaired their own injuries and maintained their breed upon the face of the earth with an efficiency man never even approached.
C. L. Moore and Henry Kuttner, Two-Handed Engine, 1955 (C. L. Moore & Kuttner, 1955, pp. 281–282)
One unusual aspect of the plot of Two-Handed Engine is that the machines are entrusted with saving human society—by acting as a police force—rather than the more common plot line of machines taking over the world.
We expect there are other relevant stories from this period too, especially from the 1950s onward. By the 1960s the concepts of machine self-reproduction and evolution were becoming a common theme. As noted in Chap. 6, some of the most significant examples from the 1960s include Poul Anderson’s 1962 short story Epilogue (Anderson, 1962), Stanisław Lem’s 1964 novel The Invincible (Lem, 1973), Fred Saberhagen’s Berserker series commencing in 1967 (Saberhagen, 1967b) and John Sladek’s 1968 novel The Reproductive System (Sladek, 1968).
One further example deserving a place in this list is the 1966 novel Sagan om den stora datamaskinen (The Tale of the Big Computer) by the Swedish physicist Hannes Alfvén, written under the pen-name Olof Johannesson (Johannesson, 1966). Although it was published a few years after the period of early pulp sci-fi we concentrate upon in our book (i.e. the 1920s–1950s), we will say a few words about it here because of its unusual status of being written by a scientist who was not only a Nobel laureate but who also very likely met John von Neumann.
The story foresees the increasing role of computers in all aspects of human life, particularly in organising the complex systems of government and society. Eventually it is deemed more reliable to have computers rather than humans in charge of their own maintenance, to which end a self-reproducing supercomputer is designed. The kind of reproduction envisaged by Alfvén is of the maker-replicator type, i.e. a machine that can produce other types of machine as well as reproduce itself, as originally proposed by von Neumann.
It is unclear whether Alfvén was directly influenced by von Neumann’s work, although von Neumann had visited Alfvén’s home department at the Royal Institute of Technology in Stockholm in 1954 (G. Dyson, Vintage Books, p. 305), so it is possible they might have discussed the topic then. The Tale of the Big Computer is noteworthy for its many visions of future technology that have indeed come to pass, including home computers, smartwatches and fitness trackers, working from home, internet shopping, and neuroprosthetics, to name a few.
More recent examples of self-reproducing machines in fiction, from the late 1960s onward, are numerous but are beyond the scope of our early history of the topic.157
8.3 Additional Material for Chapter 6
In Chap. 6 we provided a brief overview of work on the theory and practice of self-reproducing and evolving machines from the 1960s onward. As most of this work has been well covered by other histories of the subject, we restricted ourselves to highlighting some of the most significant work and providing details of existing reviews where further information may be obtained.
In our discussion of developments in software implementations in Sect. 6.2 we talked about the impact of Tom Ray’s Tierra system (T. S. Ray, 1991) in which self-reproducing computer programs compete and evolve. Tierra can be regarded as belonging to a lineage of work on self-reproducing computer programs dating back to the Darwin system developed by V. A. Vyssotsky and colleagues at Bell Labs in 1961 (Vyssotsky, 1972). This lineage has been described elsewhere (e.g. (Banzhaf & McMullin, 2012)) and was therefore not covered in our review. However, in the time since the book’s publication we have become aware of an earlier example of work in this lineage of self-reproducing computer programs, by the American engineer Fred Stahl in 1960, that deserves some additional attention here.158
8.3.1 Fred Stahl: On Artificial Universes (1960)
ⓘ This section expands upon Sect. 6.2 of the original text in Chap. 6 |
In the late 1950s, Fred Stahl—a math major at Wayne State University, Detroit—worked part-time in the university’s Computation Laboratory to fund his studies (Stahl, 2013). As a 16-year-old in 1955 he had already come across John Kemeny’s article in Scientific American describing von Neumann’s work on self-reproducing automata (Kemeny, 1955), and in 1959 he read Lionel Penrose’s article about his physical model of self-reproduction in the same magazine (Penrose, 1959). Combining these sources of inspiration, Stahl envisaged:
… a digital simulation of an extended concept of von Neumann’s notional machine. If I could make my creatures mobile in a digital universe with others of its species then I might have lethal competition. If, as von Neumann had conceived, I included digital mutation in reproduction and if the digital entities could kill and eat each other then I would have survival of the fittest. “Life” in the universe would be Darwinian. With luck I might even observe a little evolution.
Fred Stahl, 1960—the first artificial universe, 2013 (Stahl, 2013, p. A4)
Working on the lab’s IBM Type 650 computer when it was not otherwise in use, Stahl designed and coded an “artificial universe”, completing the implementation in February 1960. The universe comprised a linear sequence of 1,350 ten-digit words, with the two ends connected to form a circular topology. Multiple programs could live in this environment, being processed in pseudo-parallel by the system’s “virtual CPU.” Stahl explains that “[t]he functions used to define the creature are essentially equivalent to the operation codes of the host computer augmented by an imperfect STORE operation [to introduce mutations] and a special birth operation. Otherwise almost all of the arithmetic, logical, control and input-output (via punch cards) operations are included” (Stahl, 2013, p. B3). The virtual CPU also implemented the death criteria for the programs: if when executing a creature’s code the CPU came across a zero word (all ten digits of the word were zero) or an undefined operation, execution of the program ceased.
Stahl created a handwritten self-replicating program with which to inoculate the universe. When executed, the program would perform the following operations in order: (1) move forward one space in the universe, (2) if it encountered a unit of matter (a non-zero word) immediately in front of it, increment its counter of the number of units of matter encountered, and (3) if it had accumulated enough units of matter to equal its size, iteratively copy itself one word at a time to the space behind it before each subsequent movement until a copy of the whole program had been created (Stahl, 2013, p. B2).
In February 1960 the system was complete and ready to run. The results, however, were disappointing. Initializing the system by distributing random matter (non-zero words) across the space and introducing a single self-replicating program, Stahl reports the the logs showed a second program had been born and that both parent and offspring were crawling and “eating” (i.e. crawling over matter). However, when creature #1 started producing a second offspring this was eaten by creature #2 before it was fully created, and sometime later creature #2 took a lethal bite out of creature #1. At that point only creature #2 survived, but it turned out to be a sterile mutant that ate and crawled but produced no offspring (Stahl, 2013, p. B4).
Stahl discusses some reasons for the disappointing results, including technical issues such as the need for a larger universe, more creatures and a faster computer. In terms of the system’s design, he pointed out that the mutation rate should be optimised, and also that the conservation of matter should be enforced; the latter, however, would require a significant redesign of the system. Related to the lack of conservation of matter, it could also be noted that the concept of having to “eat” matter in order to reproduce was not inherent to the “laws of physics” of the universe but was just coded into the original creature itself; it would be perfectly possible to design a creature in the system that reproduced without first collecting matter.
After discussing the system with the head of the Computation Laboratory and a professor of the university’s Biology Department, the latter did not show any interest in the project and Stahl moved on to other things (Stahl, 2013, p. A5).
Looking back at Stahl’s results now, there is certainly a feeling of “if only” he had tried this or that small modification. Just by lowering the mutation rate, for example, creature #1 might have had a chance of producing multiple exact copies of itself in the universe so that the birth of a sterile mutant wouldn’t necessarily mean the collapse of the whole population.
Still, it is interesting to compare Stahl’s design with what came later. Banzhaf and McMullin (Banzhaf & McMullin, 2012) provide a concise discussion of the lineage of work on artificial universes with self-reproducing computer programs, covering Vyssotsky et al.’s Darwin system from 1961, Dewdney et al.’s Core War game (early 1980s) (Dewdney, 1984), Rasmussen et al.’s Venus Coreworld (1990) (Rasmussen et al., 1990), and Ray’s Tierra (1990).159
We might also add into this lineage work by the Austrian computer scientist Veith Risak, who in 1972 proposed architectures for the self-reproduction of complex computer programs using information compression and functional equivalence to reduce the amount of information required to be transmitted from parent to offspring (Risak, 1972). Risak’s paper included a description of an implementation of a reproduce-by-copy program (the simplest of the schemes for reproduction discussed in his paper and the same approach as used in Darwin and the other systems listed above) in assembly language on a Siemens 4004 computer.
Self-reproducing programs in the Darwin system and in Risak’s work were implemented in native machine code, which had an advantage of speed but also introduced various difficulties. In contrast, Core War and systems after that used their own machine languages designed for the task, interpreted by a virtual computer. Stahl’s system, created before Darwin, also used the virtual computer approach and was therefore the first to do so. Darwin, Risak’s work and Core War also lacked a mechanism for evolution through mutation—this was introduced in Venus and Tierra in 1990, but we see that Stahl’s system already featured a mutation mechanism in 1960. Several important aspects of Stahl’s design were therefore ahead of their time in terms of allowing for programs that could both reproduce and evolve.
8.3.2 A Note on Quines, Hamish Dewar and Jürgen Kraus
ⓘ This section expands upon Sect. 6.2 of the original text in Chap. 6 |
As the focus of this book is on work done prior to the early 1960s, it is beyond our scope to explore the later history in detail; the logic being that other sources already do a good job at covering this. We will however just briefly mention another flavour of work on self-reproducing computer programs that developed after this period: this focused on programs written in high-level languages (rather than assembly language) that were able to print a copy of their own source code—these programs are called quines.160
The first quine is attributed to Hamish Dewar at Edinburgh University, written in the IMP programming language. Dewar recently recounted that although his “memory of those days is fairly dim now”, his quine was written in “the late 1960s” (Dewar, 2021). As for his inspiration for writing it, he reported: “My curiosity was probably sparked by someone posing the question of whether such a program was possible or not in the coffee room one day. There was no influence from biological reproduction or anything like that”.
The subject of quines reached a larger audience in the 1970s with the publication of a short paper by Paul Bratley (a former colleague of Dewar) and Jean Millo in 1972, which featured implementations in four different languages (SNOBOL, LISP, FORTRAN and ALGOL) (Bratley & Millo, 1972). Bratley and Millo pointed out that the general structure of the various self-reproducing programs they had presented comprised the program itself accompanied by a string representation of the same program. Dewar’s original program was of the same structure too. There are notable similarities between this fundamental design principle and von Neumann’s design for a self-reproducing automaton which we discussed at length in Chap. 5. Furthermore, the same basic principle is also employed by nature in the process of reproduction of a biological cell—despite the fact that Dewar had not been influenced by biological analogies when designing his quine. Douglas Hofstadter discussed the connections between quines and cellular reproduction at length in Gödel, Escher, Bach (Hofstadter, 1979, pp. 495–548).
Our main reason for mentioning quines here is to highlight one other piece of early work in this area that is not often included in histories focused on the field of Artificial Life but which particularly deserves a mention in the current context; this is the 1980 MSc thesis of Jürgen Kraus at the University of Dortmund (Kraus, 1980). Kraus described and implemented various schemes for quines in two high level languages (SIMULA and PASCAL) and also for self-reproducing programs in a low level language (SIEMENS assembly language). The latter were of simpler design than that discussed by Risak in his 1972 paper, and more akin to the handwritten self-replicators that would be used to inoculate Tierra and similar systems in later years. At the end of his thesis Kraus also outlined proposals for models to investigate competition among self-reproducing programs in a shared environment and the addition of mutation to allow for evolution of the programs.
Although he did not actually implement these additional ideas, Kraus was nevertheless one of the first computer scientists to envisage evolution in a population of self-reproducing computer programs (i.e. employing mutation in addition to the self-reproduction of programs seen in Darwin and Core War). Kraus’ work was published twenty years after Stahl had implemented these features in 1960 but still a decade earlier than Rasmussen et al.’s Venus and Ray’s Tierra systems which received widespread attention within the Artificial Life community.
As mentioned in Chap. 6, for work from the mid-1960s onward, detailed histories of other research into self-reproducing and evolving digital organisms, collectives and ecosystems—including self-reproducing programs, string systems, cellular automata and simulated agents—can be found in various sources such as (Banzhaf & Yamamoto, 2015b), (T. J. Taylor, 1999), (Dorin et al., 2008), (Reggia, Chou, et al., 1998) and (Sipper, 1998).
8.4 Closing Remarks
The three most significant works we have highlighted in the preceding sections, by Etzler, Karinthy and Stahl, represent contributions of three different flavours: a utopian inventor’s plans for the physical self-reproduction of mechanical devices; a fiction writer’s vision of a world inhabited by superintelligent machines; and a computer programmer’s creation of an artificial universe in which digital organisms could reproduce, mutate and evolve. Each work brings its own contribution to the history.
Ideas in science fiction often precede their development by scientists and engineers (and that is one reason why we have included discussion of sci-fi works in this history). While this was certainly true for many of the gadgets envisaged by Alfvén in The Tale of the Big Computer, Etzler’s inventions and proof-of-concept experiments as a first step to achieving his dreams of a work-free society powered by self-reproducing machines—while largely unsuccessful and a long way short of fully realising his vision—are an example of engineering work preceding science fiction on the topic. Writing at the climax of the British Industrial Revolution, Etzler’s emphasis on renewable energy to power the exponentially expanding activities of his machines was certainly ahead of its time.
In Chap. 7 (Sect. 7.1.4) we discussed the variety of schemes for achieving self-reproduction proposed in the works we reviewed. To extend that discussion to the works covered in this afterword, a popular scheme has been self-reproduction achieved by a collection of robots manufacturing parts of a whole, where the parts are later assembled into a full copy of the original robots. It seems reasonable to conclude that this is the kind of scheme that Etzler envisaged. It is also the scheme discussed more explicitly by Karinthy, Alfvén and others.
A variation of the theme of collective self-reproduction appears in various works including Hamilton’s The Metal Giants, Hasse’s He Who Shrank and Hugi’s The Mechanical Mice. This variety involves a central superintelligent machine that directs a collection of lesser robots to perform various tasks including assisting in its reproduction. These stories often involve the idea that the central machine can design its own offspring rather than simply replicating its own design or relying upon mutations to power the evolution of its kind. This concept of self-designing machines was also present in some of the works reported earlier in the book, as highlighted in Sect. 7.1.1.
Turning to Stahl’s contribution, his work significantly extends—by thirty years—the history of efforts to build ecosystems of self-reproducing computer programs with the capacity for mutation and evolution, even though he only conducted minimal experimentation with the system. It is also ten years prior to Conrad and Pattee’s early studies of evolution of artificial ecosystems (which were not based upon self-reproducing computer programs) (Conrad & Pattee, 1970). Stahl’s work is—as far as we are aware—only the second example of a computational system designed to accommodate the open-ended evolution of self-reproducing structures, after Nils Aall Barricelli’s contributions in the 1950s, discussed in Sect. 5.2.1. Barricelli’s system was based upon a one-dimensional cellular automaton, making Stahl the first to use an assembly language to instantiate his digital organisms.
Writing a history of thought about self-reproducing and evolving machines—like any other history—is a Sisyphean task; there is always more to be discovered and more to be said. Nevertheless, each addition to the history provides us with a richer picture of where ideas originated and how they developed over years, decades and centuries. While this afterword, in combination with the original book, sets out the current state of our understanding of the subject, we cannot, of course, say for certain that there are no other relevant sources still awaiting discovery. If any such sources come to light, we look forward to reading them and reporting them back to interested readers in the Artificial Life community and beyond.
References
The material in this afterword was originally published in the Artificial Life journal (https://direct.mit.edu/artl) and is Ⓒ MIT Press 2024 (T. Taylor, 2024). It is reproduced here in revised form in accordance with the publication agreement between MIT Press and Tim Taylor.↩︎
To give it its full title, The Paradise Within the Reach of All Men, Without Labor, By Powers of Nature and Machinery. An Address to All Intelligent Men.↩︎
As unusual as this was for the time, Etzler’s green credentials—from an anachronistic present-day perspective—are somewhat diminished by his lack of consideration of issues relating to the continued supply of raw materials or of the effect of the envisioned widespread development of the land and sea on other species.↩︎
There are also several works from the early 1900s involving the creation of new forms of life in a test tube and their subsequent evolution, although as these feature biochemical rather than mechanical life forms they are out of scope of the subject matter of our book. Examples include the 1903 novel The Prots: A Weird Romance by an unknown author using the pseudonym Dudbroke (Dudbroke, 1903), and the 1929 novel The Greatest Adventure by Eric Temple Bell (published under his pen name John Taine) (Taine, 1929).↩︎
The story was published with Maurice G. Hugi as the author. There are various accounts of the extent to which it was written by Hugi or by his friend Eric Frank Russell who was a much more successful British sci-fi author (see (Ingham, 2010, pp. 157–159)).↩︎
The title is a reference to the enigmatic lines in John Milton’s 1637 poem Lycidas: “But that two-handed engine at the door Stands ready to smite once, and smite no more.”↩︎
As stated in Chap. 6, a partial—yet extensive—list of self-reproducing machines in fiction can be found on Wikipedia at https://en.wikipedia.org/wiki/Self-replicating_machines_in_fiction.↩︎
Fred Stahl is no relation to Walter Stahl (Stahl, 2020), who also published work on simulating self-organization and self-reproduction of artificial cells in the 1960s (e.g. (W. R. Stahl, 1965), (W. R. Stahl, 1967)). Walter Stahl’s impressive work, conducted over the mid- to late-1960s, is outside of the time period covered by our history, and has been referenced in various other histories (e.g. (Langton, 1989a)), so we do not cover his contributions here.↩︎
Steven Levy reports that Ray completed Tierra in January 1990 (Levy, 1992, p. 221). He presented it at the Artificial Life II workshop the following month, although the proceedings were not published until 1992.↩︎
The term “quine” was coined by Douglas Hofstadter in his book Gödel, Escher, Bach (1979), a decade after the original conception of this kind of program (Hofstadter, 1979).↩︎