On Replication
Sam Putman
relevant to both. Someone who is on the RepRap list might forward it
to them as well.
This rant is reproduced, with a handy table, on my blog:
http://mnemnion.wordpress.com/2009/11/22/on-replication/
cheers,
-Sam Putman.
On Replication
There is a growing interest in self-replicating machines. Beginning
with the RepRap project, and now continuing into RepLab, the open
source FabLab, there is a serious effort to build machines which can
build themselves.
It is a laudable goal. A machine which can make itself can also make
an unimaginable variety of other machines, each unique if desired, and
promises an era of material abundance and freedom from scarcity.
Pursuing that goal, however, has shown the goal itself to be somewhat
unclear. This is an attempt to remedy this situation.
What is replication? The production of a replica, but what is a
replica? The original meaning was that of a duplicate of a work of
art, properly one produced by the original artist.
I propose we define replication precisely, as the transcription of a
template into a physical form. This definition is perhaps more general
than we are used to, but I believe it accurately captures the use of
the word.
Let’s take the test case: an artist producing a replica of her own
painting. The original painting may have taken many months, while the
replica may take only a day or two at most. The artist is using tools,
a brush and palette, and materials, canvas and paint, to make a copy
of a template, the original work.
How about DNA self-replication? We have a toolchain of proteins, from
DNA Transcriptase on, which transcribe a template, the DNA, into a
copy of itself, using materials (ATP, other nucleic acids, caffeine if
there’s some in the organism in question) both extrinsic to the cell
and fabricated within it.
Note that in the first case the template resembles the result, while
in second case, if we consider the result a new cell rather than new
DNA, the result doesn’t resemble the template at all. This is
immaterial to the act of replication; a high-resolution inkjet printer
and a scanner could replicate the artist’s original work, faster and
more accurately than she could, and in this case the template would be
stored as bits within a computer.
The rest of this discussion will focus on replication in the context
of human-tool interaction.
This means replication of an object, from a template, through the
combined efforts of at least one human with at least one tool. We call
this a ‘tool’ rather than a ‘machine’ for greater generality.
The earliest form of replication was a human using a tool to make a
copy of an object. Our paradigm case for this is a scribe copying a
scroll. A human can’t do this without the materials (paper, ink) and
tools (quill, sharpener) required for the act. With those tools, a
human can copy a scroll in a certain amount of time.
A few observations here: the tool enables, the human works. A faster
tool, let us say a pen rather than a chisel on stone, allows for
faster work, but more copies will always take more human time. We will
call this Type 1 replication. It is characterized by modest gains in
efficiency for mass production and poor scaling characteristics.
A second form of replication came about when humans began using tools
to automate aspects of production. By investing time in setup, many
copies can be replicated much faster. The work is still done by
humans, with tools; but by doing the work intelligently, many copies
can be made with comparatively low effort. We call this Type 2
replication.
The paradigm case for Type 2 is the printing press. By investing
considerable effort in building a press, casting type, and laying out
a document, one is rewarded by the ability to print as many copies as
one wants, with speed and reliability which increase over time. Type 2
replication has good scaling characteristics and rewards mass
production. It is characteristic of the Industrial Age.
Type 2 production next evolved in two directions the first of which
we’ll call Type 2a: significant setup time followed by fully automated
production. Newspaper presses work in this fashion, although the most
recent ones are Type 3. Type 2a has significantly better scaling
characteristics than Type 2, and mostly displaces it when developed.
Type 2b replication is where there is significant tool investment, but
negligible setup time and non-automatic production. The matching
paradigm for this is a Xerox machine, though like newspaper presses
these have become increasingly Type 3 over time. An early Xerox would
make a copy of anything that fit on the glass, or even a hundred
copies, but a human had to stand around feeding it sheets and doing
things like collation by hand. This has poor scaling qualities, being
labor intensive, but it favors customization over mass production.
Type 3 replication has negligible setup times and fully automatic
production. Our paradigm case here is a laser printer. Even a desktop
laser printer can print replicas of stored data, each different, for
many hours without needing attention. The factory scale printers can
make entire books without human intervention. Type 3 replication is
characteristic of the Information Age, or Industrial Revolution 2.0.
Type 3 has good scaling characteristics and rewards customization over
mass production. One can make an arbitrarily large number of copies
off any given template, but there’s little to no advantage over using
many different templates.
Many FabLab technologies, such as laser cutters, hobby-scale 3d
printers, and CNC mills, are Type 2b. What’s interesting about 2b
replication is that it’s often a small step away from being Type 3,
needing only, for example, a small conveyor belt. I hope that by
drawing attention to the radical difference in scaling quality between
Type 2b replication and Type 3 replication, I can encourage
open-source hardware developers to strive for Type 3 tools whenever
possible.
A Type 2b tool is a creative enabler, while a Type 3 tool is both a
creative enabler and an economically disruptive force, because it can
compete effectively with 2a mass-production technologies. It also
frees human labor for other pursuits, which is an important goal for
many.
So this is replication; what of self-replication? This is the second
concept I hope to make clear with this essay. Self-replication is not
different from replication, except that, in the context of human-tool
interaction, that which is being replicated is a part of the tool
itself. I refer to this as recursion, because that’s what it is.
Note that there are several Type 1 replicators with a high degree of
possible recursion. The lathe is a classic example: one can hand-turn
many of the components for a lathe on itself, and even bootstrap a
lathe by building the spindle and using it to make other components. A
kiln is another example, since a kiln can be used to fire refractory
bricks to make another kiln. In fact I might argue that a kiln
exhibits the highest recursion of any Type 1 replicator, in that a
kiln can be nothing but stacked bricks and a fire, and the bricks can
all come from a kiln.
Note, however, that this still has poor scaling qualities, though it
favors custom production. The Dave Gingery approach of contributing
massive human inputs to a bootstrapped machine shop is noble, but it
will never be an economic force. It is simply too easy to identify the
main shapes needed, turn them over to Type 2 or 2a production, and
swamp the market.
The RepRap project faces exactly this dilemma. RepRap knocked the
bottom out of 3-d printing, but it did this mostly by making the
MakerBot possible. The RepRap is a 2b replicator, which requires human
tending during the entirety of the recursion process. This makes it
easier and cheaper to simply make a chassis on a faster 2b tool,
namely an Epilog laser cutter that NYC Resistor happens to already
own, and sell that. If the RepRap were a Type 3, with an automatic
conveyor, it might be a different story, since a RepRap could then
print its chassis parts at a rate of a few complete sets a week with
minimal human inputs. For that matter, a MakerBot with a conveyor belt
would be a Type 3 replicator, albiet one exhibiting less recursion
than a Darwin or a Mendel.
Let me conclude by suggesting that the RepLab project needs to place
priority on Type 3 replication, not on a maximum level of recursion
for the whole system. Recursion, I think I’ve shown, is no great trick
if it takes a human hand-holding the machine through the whole
process, since this is how machine tools have always been built, on
other machine tools.
The power of Industrial Revolution 2.0 lies in developing machines
which exhibit Type 3 replication. If the outputs of those machines are
sufficiently general than the toolset as a suite can exhibit
economically significant recursion. An example would be an open-source
toolchain for automated printing and assembly of PCB boards. Such a
suite would be able to print its own control hardware in addition to
any other circuit that can be designed and mounted with a
pick-and-place, which is most of them.
The RepLab project should focus on tools exhibiting Type 3 replication
for a variety of economically significant goods. Recursion is an
emergent result which can lend exponential momentum to the deployment
of RepLab technology. By approaching the project goals in this order,
we have a rational approach to an open source hardware platform for
Industrial Revolution 2.0.
Bryan Bishop
> There is a growing interest in self-replicating machines. Beginning
> with the RepRap project, and now continuing into RepLab, the open
> source FabLab, there is a serious effort to build machines which can
> build themselves.
What about all of Freitas' work? I think it should count for a lot.
Adrian is doing good work and all, but reprap.org is hardly the origin
of the idea of a self-replicating machine. I suggest you read:
http://www.molecularassembler.com/KSRM.htm
http://www.islandone.org/MMSG/aasm/
> It is a laudable goal. A machine which can make itself can also make
> an unimaginable variety of other machines, each unique if desired, and
> promises an era of material abundance and freedom from scarcity.
> Pursuing that goal, however, has shown the goal itself to be somewhat
> unclear. This is an attempt to remedy this situation.
Wishy-washy...
> What is replication? The production of a replica, but what is a
> replica? The original meaning was that of a duplicate of a work of
> art, properly one produced by the original artist.
>
> I propose we define replication precisely, as the transcription of a
I think you should look at Freitas' definitions instead of inventing your own.
> How about DNA self-replication? We have a toolchain of proteins, from
> DNA Transcriptase on, which transcribe a template, the DNA, into a
> copy of itself, using materials (ATP, other nucleic acids, caffeine if
> there’s some in the organism in question) both extrinsic to the cell
> and fabricated within it.
DNA self-replication is a special case of self-replication because the
amino acids and nucleic acids that are used in the construction of the
DNA are the same as the format used to encode the data, and also to
construct the phenotype (the actual expressed objects).
> This means replication of an object, from a template, through the
> combined efforts of at least one human with at least one tool. We call
> this a ‘tool’ rather than a ‘machine’ for greater generality.
The skdb developers have been thinking about that. Instead of assuming
we have to start our own Big Bang or our own supernovas, we're going
to have to assume that the user has at least a few things available to
them. Some air to breath; some tools in an inventory; maybe money in
some cases, etc. It's truly unfortunate that we can't give them a
direct path from "nothing to something", though on the other hand
starting on the hardest aspect of the problem ("from nothing") wasn't
a good idea, we found.
> A few observations here: the tool enables, the human works. A faster
> tool, let us say a pen rather than a chisel on stone, allows for
> faster work, but more copies will always take more human time. We will
> call this Type 1 replication. It is characterized by modest gains in
> efficiency for mass production and poor scaling characteristics.
>
> A second form of replication came about when humans began using tools
> to automate aspects of production. By investing time in setup, many
> copies can be replicated much faster. The work is still done by
> humans, with tools; but by doing the work intelligently, many copies
> can be made with comparatively low effort. We call this Type 2
> replication.
It looks like you're just trying to say "replication period".
> The paradigm case for Type 2 is the printing press. By investing
> considerable effort in building a press, casting type, and laying out
> a document, one is rewarded by the ability to print as many copies as
> one wants, with speed and reliability which increase over time. Type 2
> replication has good scaling characteristics and rewards mass
> production. It is characteristic of the Industrial Age.
You should check out Freitas' analysis of this. The scaling
characteristics and mass characteristics have already been analyzed
into some useful categories, like for open-ended replication, versus
closed-tail replication, and some other items.
> Many FabLab technologies, such as laser cutters, hobby-scale 3d
> printers, and CNC mills, are Type 2b. What’s interesting about 2b
> replication is that it’s often a small step away from being Type 3,
> needing only, for example, a small conveyor belt. I hope that by
How about you just write some code for the representation of these
types and commit it into skdb? Blah..
> So this is replication; what of self-replication? This is the second
> concept I hope to make clear with this essay. Self-replication is not
Hey why are you writing an essay anyway? :-)
> different from replication, except that, in the context of human-tool
> interaction, that which is being replicated is a part of the tool
> itself. I refer to this as recursion, because that’s what it is.
>
> Note that there are several Type 1 replicators with a high degree of
> possible recursion. The lathe is a classic example: one can hand-turn
> many of the components for a lathe on itself, and even bootstrap a
> lathe by building the spindle and using it to make other components. A
> kiln is another example, since a kiln can be used to fire refractory
> bricks to make another kiln. In fact I might argue that a kiln
> exhibits the highest recursion of any Type 1 replicator, in that a
> kiln can be nothing but stacked bricks and a fire, and the bricks can
> all come from a kiln.
I suggest you minimize bootstrapping to a minimum, and maximize
self-augmentation and self-replication in the end; this way, you're
not "bootstrapping into infinity" by increasingly requiring more and
more tools and so on. Freitas described this with a power series, I
think it was a p-series of the form 1^(-p). Bootstrapping should be
quick and painless-- not an attempt to reconstruct the galaxies just
to get to simple amino acid copying machines.
> Note, however, that this still has poor scaling qualities, though it
> favors custom production. The Dave Gingery approach of contributing
> massive human inputs to a bootstrapped machine shop is noble, but it
He just didn't know what to create first, really. Of course his goal
was different from a self-replicating machine: to create a typical
machine shop. If he was more focused, and he figured out what to
create first, and get the most "bang" for some amount of human input,
with minimal bootstrapping, then he would have been on a better track
IMHO.
> will never be an economic force. It is simply too easy to identify the
I don't know if we care about being an economic force or not.
> Let me conclude by suggesting that the RepLab project needs to place
> priority on Type 3 replication, not on a maximum level of recursion
> for the whole system. Recursion, I think I’ve shown, is no great trick
> if it takes a human hand-holding the machine through the whole
> process, since this is how machine tools have always been built, on
> other machine tools.
I don't think it's actually recursion if it's not doing anything.
"Recursion, in mathematics and computer science, is a method of
defining functions in which the function being defined is applied
within its own definition. The term is also used more generally to
describe a process of repeating objects in a self-similar way." Maybe
you're just using the word incorrectly ?
> The power of Industrial Revolution 2.0 lies in developing machines
Oh please.. let's not do that to ourselves.
> The RepLab project should focus on tools exhibiting Type 3 replication
> for a variety of economically significant goods. Recursion is an
> emergent result which can lend exponential momentum to the deployment
> of RepLab technology. By approaching the project goals in this order,
> we have a rational approach to an open source hardware platform for
> Industrial Revolution 2.0.
Sounds like an argument for you checking out the code base that's been
cooking under our noses. :-)