Professional Documents
Culture Documents
DRAFT 10/17/2008
Comments
are welcome.
Emergence Explained:
Getting epiphenomena to do real work
Russ Abbott
Department of Computer Science
California State University, Los Angeles
Los Angeles, California
Russ.Abbott@GMail.com
Emergence Explained:......................................................................................................1
Getting epiphenomena to do real work.......................................................................1
Russ Abbott...................................................................................................................1
1 Introduction.....................................................................................................................4
2 Background and foundations..........................................................................................6
2.1 Functionalism...........................................................................................................7
2.2 Reductionism............................................................................................................7
2.3 Epiphenomena..........................................................................................................9
2.4 Supervenience.........................................................................................................10
2.5 Fundamental forces and strong emergence.........................................................11
2.6 Historical accidents and the environment...........................................................13
3 Emergence in the Game of Life....................................................................................15
Our job will be to show how the improb- So, then, why is there any-
able couple can live happily ever after. thing except physics? … Well, I
admit that I don’t know why. I
2.1 Functionalism don’t even know how to think
Functionalism [Fodor 74] holds that about why. I expect to figure
there are so-called ‘special sciences’ (in out why there is anything ex-
fact, all sciences other than physics and cept physics the day before I
perhaps chemistry) that study regularit- figure out why there is any-
ies in nature that are in some sense thing at all … .
autonomous of physics. In [Fodor 98]
Fodor wrote the following reaffirmation So, like Holland, Fodor throws up his
of functionalism. hands with respect to explaining emer-
gence.
The very existence of the spe-
cial sciences testifies to the re- One of the tenets of functionalism is that
liable macrolevel regularities within any domain, it is the regularities
that are realized by mechan- that appear at the level of phenomena
isms whose physical substance with which the domain is concerned that
is quite typically heterogen- are important. It is not significant how
eous. Does anybody really those regularities are realized in terms of
doubt that mountains are lower level phenomena—both because
made of all sorts of stuff? Does that doesn’t matter and because they can
often be implemented in any of a num-
anybody really think that,
ber of ways. A term commonly used in
since they are, generalization
Functionalism is multiple realizability,
about mountains-as-such
which refers to the notion that many reg-
won’t continue to serve geo-
ularities (many functions) can be real-
logy in good stead? Damn near
ized in multiple ways. As Fodor remarks,
everything we know about the
world suggests that unimagin- that’s why references to can
ably complicated to-ings and openers, mousetraps, cam-
fro-ings of bits and pieces at shafts, calculators and the like
the extreme microlevel man- bestrew the pages of function-
age somehow to converge on alist philosophy. To make a bet-
stable macrolevel properties. ter mousetrap is to devise a
new kind of mechanism whose
Although Fodor does not use the term,
behavior is reliable with re-
the phenomena studied by the special
spect to the high-level regular-
sciences are the same sort of phenomena
ity “live mouse in, dead mouse
that we now call multiscale, i.e., emer-
out.”
gent.
Why is there emergence? Fodor contin- 2.2 Reductionism
ues as follows. Taking the other side of the debate is
Steven Weinberg, one of the most articu-
[T]he ‘somehow’ [of the pre-
late defenders of reductionism. Weinberg
ceding extract] really is en-
distinguishes two kinds of reductionism.
tirely mysterious … .
We ought first of all to distin-
the originals. guish between what (to borrow
exist would not describe epiphenomena. tion of values to the elements of L but
It would describe the epiphenomena of different configuration of values to the
any system that satisfies that specifica- elements of H. In other words when a
tion, but that’s as far as one could go. state of affairs assigns values to predic-
From our perspective, there are no epi- ates in L, that fixes the assignments of
phenomena unless they are epiphenom- values to predicates of H.
ena of something. That is not the case Consider the following simple example.
with functionality, which is understood Let the domain be a sequence of n bits.
in the abstract. Let L be the statements: bit 1 is on; bit 2
It is reasonable to say that the function- is on; etc. Let H be statements of the
ality of most executing software is epi- sort: exactly 5 bits are on; an even num-
phenomenal. The computation con- ber of bits are on; no two successive bits
sidered as an abstraction may be defined are on; the bits that are on form the ini-
independently of the implementation. tial values in the Fibonacci sequence;
But the only real action is at the very etc.
lowest level. No matter how abstract H supervenes on L since any configura-
one’s software, one can always stop a tion of values of the statements in L de-
computation by pulling the plug on the termines the values of the statements in
computer. The computation as an ab- H.
stract epiphenomenon exists only be-
cause electrons are actually flowing. However, if we remove one of the state-
ments from L, e.g., we don’t include in L
2.4 Supervenience a statement about bit 3, but we leave the
A closely related term from the philo- statements in H alone, then H does not
sophical literature is supervenience. The supervene on L.
intended use of this term is to relate a To see why, consider the H statement
presumably higher level set of predicates
or properties6 (call this set H for higher) h1: an even number of bits is on.
to a presumably lower level set of pre- For concreteness, let’s assume that there
dicates or properties (call this set L for are exactly 5 bits. Let’s assume first, as
lower). The properties or predicates in H in the first line of Figure 1, that all the
and L are all presumed to be applicable bits except bit 3, the one for which there
to some common domain of discourse. is no L statement, are on. Thus since
H and L are each ways of characterizing there is no L statement about bit 3, all
the state of affairs of the underlying do- the L statements are true even though bit
main. For any particular state of affairs 3 is off. Since 4 of the 5 bits are on, h 1 is
in the domain of discourse, the predic- also true.
ates in H and L will each be either true Now, assume that bit 3 is on as in the
or false (or perhaps not applicable). second line of Figure 1. All the L state-
One says that H supervenes on (or over)
L if it is never7 the case that two states of 7
Some definitions require that not only is
affairs will assign the same configura- it never the case, it never can be the case. It does
make a formal difference whether we base super-
6
Since properties (e.g., the color of an ob- venience on a logical impossibility or on empirical
ject) may be expressed in terms of predicates about facts. We finesse that distinction by adopting the
that property, from here on we will speak only of rule of thumb of fundamental particle physicists: if
predicates. something can happen it will.
ments are still true. But since 5 bits are some extent this is the case. When we
now on, h1 is now false. consider autonomous entities, however,
we shall see that supervenience is not as
Since we have found an H statement that
useful as one might have hoped. This is
has two different values for a single con-
one reason that emergence has been so
figuration of values of the L statements,
difficult to pin down.
H does not supervene over L.
One reason that supervenience is less
Although we have not attempted to trace
useful than one might have hoped may
its history, the notion of supervenience
be related to the difficulty one encoun-
may have originated as an attempt to
ters when using supervenience for infin-
capture the relationship between epi-
ite domains. Consider our bit example
phenomena and their underlying phe-
again, but imagine that we have a count-
nomena. Presumably epiphenomena su-
ably infinite number of bits. Consider
pervene on underlying phenomena: dis-
the H statement
tinct epiphenomena must be associated
with distinct underlying phenomena, h2: the bits that are on are prime.
which is what one wants.8 You can’t get Clearly the H set consisting solely of h2
two different sets of epiphenomena from supervenes over the entire set of L state-
the same underlying phenomena.9 ments. Just as clearly, that H set does not
Note that the reverse is not true. Two dif- supervene over any proper subset of the
ferent states of the underlying phenom- L statements, and certainly not over any
ena may result in the same epiphenom- finite subset of the L statements—one
ena. In our bit example, there are many needs to look at all of the bits to determ-
different ways in which an even number ine whether it is exactly the prime bits
of bits may be on. that are on.
The position known as supervenience So even though we can conclude that H,
physicalism might be understood as which contains a single relatively simple
claiming that any higher-level descrip- statement, supervenes over the infinite
tion of nature supervenes over some set set of statements in L, that information
of primitive descriptions. doesn’t buy us much. On the contrary,
supervenience of this sort is like describ-
It would appear that the relationship
ing a tapestry by enumerating the threads
defined by supervenience will be useful
that make it up. The epiphenomena are
in analyzing multi-scale phenomena. To
lost. We will see another example of this
8
On the other hand, searches for the later.
phrase “epiphenomena supervene” on Google, Ya-
hoo msn.com, and AskJeeves conducted on
9/17/2005 found no references—other than to an 2.5 Fundamental forces and
online draft of this paper. This suggests that none strong emergence
of these services had ever scanned a document that
contained the sentence fragment “… epiphenom-
Returning to Weinberg and Fodor, pre-
ena supervene over … .” sumably both would agree that phenom-
9
It’s not really the epiphenomena that su- ena of the special sciences supervene on
pervene over their underlying phenomena. It’s phenomena in physics. A given set of
statements about the epiphenomena or properties
of the epiphenomena that supervene over state-
phenomena at the level of fundamental
ments about the underlying phenomena or their physics is associated with no more than
properties. As a short-hand, however, we will talk one set of phenomena at the level of any
about epiphenomena as supervening (or not) over
underlying phenomena.
of the special sciences. Or looking top-
able extension of physics (but one that are no magical mystery forces; there is
we have not seen and do not know how no strong emergence.
to produce), and since the appearance of The fact that all special (i.e., higher
new forces that violate the conservation level) science interactions are epiphen-
laws is currently unimaginable, from omenal is not a problem for functional-
now on we ignore the possibility of ism. Nor is it news to Fodor, who speaks
strong emergence. freely of “the to-ings and fro-ings of bits
If one dismisses the possibility of strong and pieces at the extreme microlevel.”
emergence and agrees that the only Rather, the functionalist claim is that the
forces of nature are the fundamental regularities (be they epiphenomenal or
forces as determined by physics, then not) that appear at the level of any spe-
Fodor must also agree (no doubt he cial science are of significance on their
would) that any force-like construct pos- own. The fact that the interactions to
tulated by any of the special sciences which one refers in describing those reg-
must be strictly reducible to the funda- ularities are in fact implemented by
mental forces of physics. As Weinberg lower level interactions does not dimin-
says, there is no life force. ish either the importance or the lawful-
ness of the higher level regularities—
Note that this is a truly stark choice:
even though as we have seen Fodor has
strict reductionism with respect to forces
no idea how those regularities come
or strong emergence. There is no third
about.
way.
This leads to an important conclusion. 2.6 Historical accidents and the
Any cause-like effect that results from a environment
force-like phenomenon in the domain of There is a second area of at least implicit
any of the special (i.e., higher level) sci- agreement between Weinberg and Fodor.
ences must be epiphenomenal.10 Since Consider the following from Weinberg.
epiphenomenal forces supervene on fun- [A]part from historical acci-
damental forces, distinct epiphenomenal
dents that by definition cannot
interactions must be manifestations of
be explained, the [human]
distinct fundamental force actions. In
nervous system [has] evolved
other words, anything that happens at the
to what [it is] entirely because
most fundamental level (if there is one)
of the principles of macroscop-
has no more than one manifestation at
ic physics and chemistry,
any higher level.
which in turn are what they are
It is important to note that this perspect- entirely because of principles
ive establishes one of the basic claims of of standard model of element-
reductionism: forces at all levels must be ary particles.
explicable in terms of—i.e., they are epi-
phenomenal of and reducible to—the
fundamental forces of physics.11 There 11
Compare this with the conclusion Hume
reached [Hume] in his considerations of causality
—that when one looks carefully at any allegedly
10
Kim [Kim ‘93] (denigratingly) used the direct causal connection, one will find intermedi-
term epiphenomenal causation to refer to interac- ary links. Since Hume did not presume what we
tions of this sort. We also consider such interac- now consider to be a bottom level of fundamental
tions to be epiphenomenal, but we don’t find them physical forces, he dismissed the notion of causal-
less worthy as a result. ity entirely.
many as one wants. If we look at the Turing Machine isn’t real; they are all
Game of Life as a programming plat- (nothing but) epiphenomena.
form—imagine that we are kids fooling Furthermore, even the interactions
around with a new toy—we might exper- between and among patterns aren’t real
iment with it to see whether we can either. They’re also epiphenomenal—
make other sorts of patterns. If we find and epiphenomenal in the sense de-
some, which we will, we might want to scribed above: the only real action is at
see what happens when patterns crash the most fundamental level, the Game of
into each other—boys will be boys. Life rules. Pattern APIs notwithstanding
After some time and effort, we might the only thing that happens on a Game of
compile a library of Game of Life pat- Life grid is that the Game of Life rules
terns, including the API16 of each pattern, determine which cells are to be on and
which describes what happens when that which cells are to be off. No matter how
pattern collides with other patterns.17 real the patterns look to us, interaction
among them is always epiphenomenal.
Since its introduction three decades ago,
So what are we talking about?
a community of such Game of Life pro-
grammers has developed. That com- What does one do to show that a Game
munity has created such libraries—at of Life emulation of a Turing machine is
least on an informal basis.18 correct? What one must do is to adopt a
operational perspective and treat the pat-
It has even been shown [Rendell] that by
terns and their interactions, i.e., the
suitably arranging Game of Life pat-
design itself, independently of the Game
terns, one can simulate a Turing Ma-
of Life.
chine.
It is the design, i.e., the way in which the
Moreover, and this is a crucial point, the
patterns—be they real or epiphenomenal
emulation of a Turing Machine with
—interact that we want to claim simu-
Game of Life patterns is also an example
lates a Turing Machine. To show that we
of emergence. There is no algorithm.
must do two things.
The Turing Machine appears as a con-
sequence of epiphenomenal interactions 1. Show that the abstract design con-
among epiphenomenal patterns! sisting of patterns and their interac-
tions actually does simulate a Turing
What did we just say? What does it
Machine. That is, we reify the
mean to say that epiphenomenal gliders
design, i.e., treat it as real, and argue
and other epiphenomenal patterns simu-
about its properties.
late a Turing Machine? How can it mean
anything? The patterns aren’t real; the 2. Show that the design can be imple-
mented on a Game-of-Life platform.
16
Application Programming Interface
17
Note, however, that interactions among
Note what this perspective does. It un-
patterns are quite fragile. If two patterns meet in shackles the design from its moorings as
slightly different ways, the results will generally a Game of Life epiphenomenon, reifies
be quite different.
18
it as an independent abstraction, and lets
Many of these libraries are available on
the web. To explore this world, a good place to
it float free. (The protestors in the streets
start and one that seems to be kept up to date is chanting “Free the design” can now
Jason’s Life Page [Summers]. The patterns avail- lower their picket signs and go home.)
able on that page and on the pages to which that
page links are quite amazing.
The design becomes an abstraction, an grid that seems more complex than most
abstract construct on its own. Once we of the others. The Game of Life rules
have such a design as an abstraction we fully explain every light that goes on and
can then reason about its properties, i.e., every light that goes off on Hacka’s
(a) that it accomplishes what we want, pretty face. But somehow that explana-
namely that it simulates a Turing Ma- tion doesn’t seem to capture everything
chine and (b) that it can be reattached to that’s going on. It just seems more com-
its moorings and be implemented on a plex than that. Did we miss something?
Game of Life platform. In other words, To make a long story short, it turns out
emergence is getting epiphenomena to that the tribe was not as isolated as we
do real (functional) work. had thought. In fact they have an Internet
connection. Hacka had learned not only
3.4 Game of Life anthropologists
that she was a Game of Life system but
Let’s forget everything we just said
that the Game of Life can emulate a Tur-
about the Game of Life, and let’s pretend
ing Machine. She had decided to pro-
we are anthropologists. Let’s imagine
gram herself to do just that. Her parents
that a lost tribe of what turn out to be
disapproved. But girls just want to have
Game of Life creatures has been dis-
fun—especially teenage girls.
covered in a remote wilderness. Of
course, when they are first discovered, No wonder we felt uncertain about our
we don’t know that they are Game of results. Even though the Game of Life
Life creatures. All we know is that their rules explained every light that went on
strange grid-like faces are made up of and off on Hacka’s face, it said nothing
cells that blink on and off. about the functionality implemented by
Hacka’s Turing Machine emulation.
We get a grant to study these creatures.
We travel to their far-off village, and we The rules explained everything about
learn their language. They can’t seem to how the system worked; but they said
explain what makes their cells blink on nothing about what the system did. The
and off; we have to figure that out for rules didn’t have a way even to begin to
ourselves. talk about the functionality of the system
—which was logically independent of
After months of study, we come up with
the rules. The rules simply don’t talk
the Game of Life rules as an explanation
about Turing Machines.
for how the grid cells are controlled.
That seems to work. Every single mem- A Turing machine is an autonomous
ber of the tribe operates in a way that is functional abstraction that we (and
consistent with those rules. The rules Hacka) built on top of the rules of the
even explain the unusual patterns that we Game of Life. Our reductive explana-
observe—some of them, glider-like, tra- tion, that a certain set of rules make the
versing the entire grid. Thrilled with our cells go on and off, had no way to cap-
analysis, we return home and publish our ture this sort of additional functionality.19
results.
19
In “Real Patterns” Dennett [Dennett ‘91]
But one thing continues to nag. One of uses the fact that a Turing Machine may be imple-
the teenage girls—she calls herself mented in terms of Game of Life patterns to argue
that the position he takes in The Intentional Stance
Hacka for reasons that we do not under- [Dennett ‘87] falls midway along a spectrum of
stand—has a pattern of activities on her positions ranging from what he calls “Industrial
strength Realism” to eliminative materialism (that
Facts about Turing Machines, i.e., the do the same thing with gliders. We can
theorems of computability theory, are establish a domain of discourse about
derived de novo. They are made up out gliders as abstract entities. Within that
of whole cloth; they are not based on the domain of discourse we can reason about
Game of Life rules. The fact that such gliders, and in particular we can reason
abstract designs can be realized using about which cells a glider will turn on
Game of Life rules as an implementation and when it will turn them on. Our the-
platform tells us nothing about comput- ory will tell us how fast an in which dir-
ability theory that we don’t already ection a glider moves
know. Having developed facts and rules about
gliders as independent abstract elements,
4.2 Downward entailment
we can then use the fact that gliders are
On the other hand, the fact that a Turing
epiphenomena of the Game of Life and
Machine can be implemented using the
by appeal to downward entailment apply
Game of Life rules as primitives does
those facts and rules to the Game of Life
tell us something about the Game of Life
cells that implement gliders.
—namely that the results of computabil-
ity theory can be applied to the Game of This sounds more complex than it really
Life. The property of being Turing com- is. What it really amounts to is that
plete applies to the Game of Life pre- downward entailment justifies what we
cisely because a Turing Machine can be as human beings tend to do anyway: no-
shown to be one of its possible epiphen- tice regularities in the world and then en-
omena. gage the world in terms of those regular-
ities.
In other words, epiphenomena are down-
ward entailing. Properties of epiphenom-
4.3 Reduction proofs
ena are also properties of the phenomena Consider how the unsolvability of the
from which they spring. This is not quite halting problem applies to the Game of
as striking as downward causation24 Life. The fact that a Turing Machine can
would be, but it is a powerful intellectual be implemented on a Game of Life plat-
tool. form means, among other things that the
Earlier, we dismissed the notion that a halting problem for the Game of Life—
glider may be said to “go to a cell and which we can define as determining
turns it on.” The only things that turn on whether a game of Life run ever reaches
Game of Life grid cells are the Game of a stable (unchanging or repeating) con-
Life rules. But because of downward en- figuration—is unsolvable.
tailment, there is hope for talk of this When reasoning about insolvability one
sort. often talks of reducing one problem to
To prove that a Turing Machine emula- another. In this case, because we can im-
tion on a Game of Life platform does plement Turing Machines using the
what we claim it does we disengaged the Game of Life, we know that we can re-
design of the emulation from its imple- duce the halting problem for Turing Ma-
mentation, and we reasoned about the chines to the halting problem for the
emulation as an abstract design. We can Game of Life: if we could solve the
24
See, for example [Emmeche] for a num-
Game of Life halting problem, we could
ber of sophisticated discussions of downward solve the Turing Machine halting prob-
causation.
lem. But we know that the Turing Ma- may be literally incorrect to say that a
chine halting problem is unsolvable. glider “turns on a Game of Life cell” or
Therefore the Game of Life halting prob- that “one billiard ball pushes another one
lem is also unsolvable. This sort of in a particular direction,” because of
downward entailment reduction gives us downward entailment this sort of con-
a lot of intellectual leverage since it’s not ceptual shorthand is not only reasonable
at all clear how difficult it would be to but essential for how we think about the
prove “directly” that the halting problem world. But see the discussion of Newto-
for the Game of Life is unsolvable. nian mechanics in the next section for a
somewhat stronger version of this per-
Thus another consequence of downward
spective.
entailment is that reducibility cuts both
ways. One can conclude that if Downward entailment in the form of the
something is impossible at a higher level process just sketched is, in fact, a reas-
it must be impossible at the lower (im- onable description of how we do sci-
plementation) level as well. But the only ence: we build models, which we then
way to reach that conclusion is to reason apply to the world around us.
about the higher level as an independent We are not saying that there are forces in
abstraction and then to reconnect that ab- the world that operate according to bil-
straction to the lower level. Logically in- liard ball rules or that there are forces in
dependent higher level abstractions, i.e., the Game of Life that operate according
functionality and design, matter on their to glider rules. That would be downward
own. causation, a form of strong emergence,
which we have already ruled out. What
4.4 Downward entailment as sci-
we are saying is that billiard balls,
ence
gliders, Turing Machines, and their inter-
A striking example of downward entail-
actions can be defined in the abstract.
ment is the kind of computation we do
We can reason about them as abstrac-
when computing the effect of one bil-
tions, and then through downward entail-
liard ball on another in a Newtonian uni-
ment we can apply the results of that
verse. It’s a simple calculation involving
reasoning to any implementation of
vectors and the transfer of kinetic en-
those abstractions whenever the imple-
ergy.
mentation preserves the assumptions re-
In truth there is no fundamental force of quired by the abstraction.
physics corresponding to kinetic energy.
If one had to compute the consequences 4.5 The reality of higher level ab-
of a billiard ball collision in terms of stractions
quantum states and the electromagnetic In “Real Patterns” [Dennett ‘91], Den-
force, which is the one that applies, the nett argues that when compared with the
task would be impossibly complex. But work required to compute the equivalent
the computation is easy to do at the epi- results in terms of primitive forces, one
phenomenal level of billiard balls. We gets a “stupendous” “scale of compres-
know that the computation we do at the sion” when one adopts his notion of an
billiard ball level applies to the real intentional stance [Dennett ‘87]. Al-
world because of downward entailment: though “Real Patterns” doesn’t spell out
billiard balls are epiphenomena of the the link explicitly, Dennett’s position ap-
underlying reality. So even though it pears to be that because of that intellec-
tual advantage, one should treat the on- Newton’s legendary laws have
tologies offered by the intentional stance turned out to be emergent.
as what he calls “mildly real”—although They are not fundamental at
he doesn’t spell out in any detail what all but a consequence of the
regarding something as “mildly real” in- aggregation of quantum mat-
volves. ter into macroscopic fluids and
Our position contrasts with Dennett’s in solids—a collective organiza-
that we claim that nature is often best tional phenomenon. … [New-
understood—the theories of science are ton’s laws] are as exact and
best expressed—in terms of two-level true as anything we know in
(or perhaps multi-level) theories. One physics—yet they vanish into
level is an abstract design; the other nothingness when examined
level is the implementation of that too closely. … [M]any physi-
design. Whether or not one wants to say cists remain in denial. To this
that the abstract design is “mildly real” day they organize conferences
(or real with some other adjective ap- on the subject and routinely
plied) is not our focus.25 speak about Newton’s laws be-
ing an “approximation” for
As we shall see below, our claim will quantum mechanics, valid
also be that the entities (such as billiard when the system is large—
balls) about which higher level abstrac- even though no legitimate ap-
tions are formulated are real in an ob- proximation scheme has ever
jective sense (they have reduced en-
been found.
tropy) but that interactions among those
entities are epiphenomenal—since the A second example to which Laughlin
only forces in nature are the fundamental frequently returns is the solid state of
forces. matter, which, as he points out, exhibits
properties of rigidity and elasticity. The
In a recent book [Laughlin], Laughlin ar-
solid state of matter may be character-
gues for what he calls collective prin-
ized as material that may be understood
ciples of organization, which he finds to
as a three dimensional lattice of compon-
be at least as important as reductionist
ents held together by forces acting
principles. For example in discussing
among those components.
Newton’s laws he concludes from the
fact that (p. 31) Once one has defined an abstract struc-
ture of this sort, one can derive proper-
these [otherwise] overwhelm-
ties of matter having this structure. One
ingly successful laws … make
can do so without knowing anything
profoundly wrong predictions more about either (a) the particular ele-
at [the quantum] scale ments at the lattice nodes or (b) how the
that binding forces are implemented. All one
needs to know are the strengths of the
forces and the shape of the lattice.
25
Furthermore, our position is that the
question of whether or not anyone would find it From our perspective, both Newton’s
advantageous to have beliefs about epiphenomena laws and the solid state of matter are ab-
(which is Dennett’s primary concern) has nothing
to do with the issues we are considering. stract organizational designs, i.e., epi-
phenomena. They are abstractions that
apply to nature in much the same way as designs and the theories built on them
a Turing Machine as an abstraction ap- are new and creative constructs and are
plies to certain cell configurations in the not consequences of the platform on
Game of Life. Laughlin calls the imple- which they are implemented. The Game
mentation of such an abstraction a pro- of Life doesn’t include the concept of a
tectorate. Turing machine, and quantum physics
doesn’t include the concept of a solid.
Laughlin points out that protectorates
tend to have feasibility ranges, which are The point of all this is to support Laugh-
often characterized by size, speed, and lin position: when nature implements an
temperature. A few molecules of H2O abstraction, the epiphenomena described
won’t have the usual properties of ice. by that abstraction become just as real
And ice, like most solids, melts when any other phenomena, and the abstrac-
heated to a point at which the attractive tion that describes them is just as valid a
forces are no longer able to preserve the description of that aspect of nature as
lattice configuration of the elements. any other description of any other aspect
Similarly Newton’s laws fail at the of nature.
quantum level. Our notion that much of nature is best
The existence of such feasibility ranges understood in terms of implementations
does not reduce the importance of either of abstractions suggests that many sci-
the solid matter abstraction or the New- entific theories are best expressed at two
tonian physics abstraction. They just levels: (1) the level of an abstraction it-
limit the conditions under which those self, i.e., how it is specified, how it
abstractions apply, i.e., under which works on the abstract level, and what its
nature is able to implement them. implications are, and (2) the level that
explains (a) under what circumstances
The more general point is that nature im-
(when and where) that abstraction may
plements a great many such abstract
be found implemented and (b) how that
designs. As is the case with computabil-
implementation works.
ity theory, which includes many sophist-
icated results about the Turing machine The use of agent-based modeling in the
abstraction, there are often sophisticated social sciences illustrates this methodo-
theories that characterize the properties logy. Agent-based models are typically
of such naturally occurring abstractions. used to show both (a) how a higher level
These theories may have nothing to do abstraction functions in its context, i.e.,
with how the abstract designs are imple- what its consequences as a theory are,
mented. They are functional theories that and (b) how that abstraction may come
apply to the abstract designs themselves. to be realized in terms of some particular
To apply such theories to a real physical set of lower level interactions.
example (through downward
entailment), all one needs is for the 4.6 Abstractions and abstract
physical example to implement the ab- designs
stract designs. We have intentionally used the terms ab-
straction and abstract design somewhat
Furthermore and perhaps more import- interchangeably. Abstraction is, of
antly, these abstract designs are neither course, a much more general term. Yet
derivable from nor logical consequences from the perspective of scientific explan-
of their implementations. Abstract
ation, what one is looking for is, as This operational perspective seems to be
Weinberg says, a description of why somewhat different from that of tradi-
nature is the way it is. tional functionalism. Functionalism is
concerned with the functions and regu-
To some extent, this includes the ques-
larities that characterize the special sci-
tion of why nature is made up of
ences. It tends not to be concerned with
whatever makes it up. But for the most
how those functions or regularities are
part, this question is generally taken to
realized. In fact the opposite seems to be
be asking why nature works the way it
the case. By emphasizing the possibility
does.
of multiple realizability, functionalism
As summarized by Woodward [Wood- minimizes the importance of any partic-
ward], scientific explanations are in- ular realization and dismisses the im-
tended to explain portance of understanding operational
why things happen, where the design issues.
“things” in question can be
4.7 Phase transitions
either particular events or
Those of us in Computer Science know
something more general—e.g.,
that implementations of abstract designs
regularities or repeatable
are often not perfect. Most include some
patterns in nature.
compromises, and some have bugs.
Typically we are looking for an opera- What about nature’s implementation of
tional explanation: how does nature abstract designs?
work?26 Thus when we speak of an ab-
Nature’s implementation of abstract
stract design, we are referring to the
designs are not always perfect either—
design of an operational mechanism that
especially, as Laughlin points out, either
can be understood as bringing about
(a) when the conditions under which an
some consequence. From here on, when
implementation is feasible are not in ef-
we talk about abstractions or abstract
fect or (b) when one approaches the
designs, this is what we have in mind:
boundaries of such feasibility regions.
the abstract design of a mechanism that
produces certain kinds of results. There will almost always be borderline
situations in which the implementation
Note that it is not just the functional res-
of an abstract design is on the verge of
ults that matter; the abstract operational
breaking down. These borderline situ-
design matters also. A Turing Machine is
ations frequently manifest as what we
not just a device whose input-output
call phase transitions—regions or points
functionality satisfies the theorems of
(related to a parameter such as size,
computability theory; it is an abstract
speed, temperature, and pressure) where
device that operates in a particular way.
multiple distinct and incompatible ab-
The theorems of computability theory
stractions may to be implemented.
follow from the way a Turing Machine
operates, not the other way around. It is Newton’s laws fail at both the quantum
the abstract operational design that level and at relativistic speeds. If as
comes first. The theory follows. Laughlin suggests, the Newtonian ab-
straction cannot be shown to be an ap-
26
As Woodward explains, though, the
proximation of quantum theory, phase
question of what should be taken as a scientific ex-
planation is the subject of continuing investigation.
that are consistent with the constraints Deep Impact's revelations "are
imposed by the fundamental laws of going to change lot of our
physics, i.e., all the possible abstract ideas about comets," predicts
designs that nature may possibly imple- [Jay] Melosh [of the University
ment, and predict which will come to of Arizona in Tucson].
pass.
These are not the words of scientists in
It is worth considering what the role of the process of deriving results about
experiments would be were it possible to comets from the fundamental laws of
start with the fundamental laws of phys- physics. These are scientists building
ics and reconstruct the universe. We can higher level models.
think of two.
It is just as wrong for Weinberg to denig-
1. To help refine our understanding of rate the notion that there can be inde-
the fundamental laws of physics. pendent principles of chemistry or psy-
2. If the laws of physics require that chology as it would be for him to claim
one of a number of possible situ- that the principles of Mathematics and
ations be the case, to decide which of Computer Science are all reducible to
them actually is the case. and derivable from physics.
Most experiments are done for neither of Most sciences are like Mathematics and
these reasons. Most experiments are Computer Science. They really do stand
done to help establish whether some on their own—which is why (as Ander-
higher level regularity actually holds. son says) the constructionist hypothesis
is wrong and why (as Fodor says) the
Recently in the news were the results of special sciences are autonomous..
the impact of a projectile with the comet
Tempel 1. As reported by Cowan [Cow- 5 Entities
an], the experiment So far, we have discussed what one
revealed several surprises. The might characterize as emergence in the
data … are at odds with a lead- large. There is also emergence on a
ing model for the structure of smaller and more local scale. That sort
comets called the dirty-snow- of emergence is related to what we intu-
ball model. … itively think of as entities. This section
discusses entities and how they relate to
[T]he data from the Deep Im- emergence.
pact mission indicate that al-
though Tempel 1 contains We think in terms of entities, i.e., things
some ices, its primary con- or objects. It seems like the most natural
stituent may be dust particles thing in the world. Yet the question of
finer than talcum power. … how one might characterize what should
and should not be considered an entity
What's more, the comet isn't a has long been a subject of philosophical
mere hodgepodge of different study. A brief review of the recent literat-
materials and structures. "The ure (for example, [Boyd], [Laylock],
damn thing is layered like a [Miller], [Rosen], [Varzi Fall ‘04]) sug-
frozen onion," says Deep Im- gests that no consensus about how to un-
pact scientist Joseph Veverka derstand the notion of “an entity” has yet
of Cornell University. … been reached.
One might adopt a very general position. If the preceding does not formally ex-
For example, Laylock quotes Lowe clude instants, events, and durations, we
[Lowe] as follows. will explicitly exclude them too. Entities
for us will be required to persist in time,
‘Thing’, in its most general
but they will not be aspects of time, i.e.,
sense, is interchangeable with
instants or durations, or events, whatever
‘entity’ or ‘being’ and is
an event is.
applicable to any item whose
existence is acknowledged by An entity for us will be either atomic
a system of ontology, whether (not in the sense of being a chemical ele-
that item be particular, ment but in the more generic sense of
universal, abstract, or having no constituents—if indeed there
concrete. In this sense, not are atomic physical elements in nature),
only material bodies but also or, if an entity has constituents, it will be
properties, relations, events, an epiphenomenon of its constituents.
numbers, sets, and Thus for us non-atomic entities will rep-
propositions are—if they are resent one of the most common forms of
acknowledged as existing—to emergence.
be accounted ‘things’. Our purpose in this section is not to
For our purposes, this is too broad. In settle the grand philosophical question of
this paper we want to exclude properties, what one should mean by the terms
relations, events, numbers, sets and pro- thing, object or entity but to sketch out
positions from our notion of entity. We what it means to be an entity in our
don’t want to think of, say, the American sense. Of course we hope that the frame-
Civil War or happiness as an entity in the work we develop will offer a useful way
same way that we think of an atom is an of thinking about some of the uses to
entity. which we commonly put the terms
thing, object, and entity.
On the other hand, we don’t want to lim-
it ourselves to strictly material objects. 5.1 Entities, entropy, designs,
We want to include countries, teams, and functionality
corporations, and families, for example, The standard model of physics includes
as well as what may seem like quasi- fundamental particles such as electrons,
physical entities such as people and hur- photons, quarks, etc. These are entities
ricanes, whose physical makeup under- which have no constituents. Beyond
goes continual change. these, one has atomic nuclei, atoms, and
For our purposes, entities, by fiat, will molecules, all of which we want to in-
always have some material aspect. That clude in our notion of entity.
is, an entity will at any time consist of For functionalism, entities are every-
physical elements arranged in a particu- where: mice and cans are good exam-
lar way. With this decision we are ex- ples. The higher level sciences speak of
cluding from our notion of entity strictly all sorts of entities, including biological
mental constructs such as sets, numbers, entities (e.g., you and me) and social,
concepts, propositions, relationships, political, and economic entities, such as
designs, abstractions, etc. families, states, and corporations.
the masses of the protons and neutrons with no extra effort expended anywhere
that make up an alpha particle when not else.
bound together as an alpha particle.30 There really is a free lunch. Atomic nuc-
The same entity-mass relationship holds lei form, atoms form, solar systems and
for all physical and chemical entities. galaxies form—all without depleting any
The mass of an atom or molecule is energy reservoirs. We are so used to this
(negligibly) less than the sum of the fact that we hardly notice it. But if one
masses of its components taken separ- stands back and observes that at-equilib-
ately. The mass of the solar system is rium entities exemplify emergence at its
(negligibly) less than the mass of the sun most basic—an atom is emergent from,
and the planets when taken separately. it is an epiphenomenon of, and it super-
venes over its components—we may
This fact implies that the entropy of
conclude that spontaneous emergence is
these entities is lower than the entropy of
fundamental to how nature works.
the components taken separately. In oth-
er words, an entity at an energy equilib- Even so, one might suppose that beyond
rium is distinguishable by the fact that it combining in these basic ways (as atom-
has lower mass and lower entropy than ic nuclei, atoms, and astronomical ag-
its components taken separately. gregations held together by gravity), at-
equilibrium entities are not very interest-
These entities are trivially self-perpetuat-
ing. Standing back again makes it clear
ing in that they are in what is often
that this is not the case. Given what we
called an energy well and require energy
have learned during the past half century
to pull their components apart. This
(and what we still don’t know)—espe-
gives us a nice metric of entityness for
cially about condensed matter physics
at-equilibrium entities: the amount of en-
and including, as we said earlier, the
ergy required to pull it apart.
startling fact that the same matter is cap-
5.3 Entities and emergence are able of implementing multiple abstrac-
fundamental tions with radically different properties
The mechanisms (gravity, the strong —at-equilibrium entities are far from
nuclear force, and the electromagnetic boring.
force) that expel entropy from at-equilib-
rium entities and that hold these entities 5.4 Dissipative structures
In [Prigogine] (and elsewhere) Prigogine
together are the fundamental forces of
discussed what he called a dissipative
nature.
structure. We see dissipative structures
One can say that these mechanisms in as the essential stepping stone from at-
some sense run for free. To the extent equilibrium entities to autonomous enti-
that we understand how they work at all, ties.
we attribute their operation to virtual
Intuitively, a dissipative structure typic-
particles that pop into and out of exist-
ally manifests when energy is pumped
ence and that do the work of the force—
into a bounded region. D
30
It turns out that the atomic nucleus with
the least mass per nucleon is iron. Energy from fu-
issipative structures typically involve
sion is possible for elements lighter than iron; en- structured activities internal to the re-
ergy from fission is possible for elements heavier gion. A standard example consists of the
than iron. (See [Nave] for a discussion of these
matters.)
Bénard convection cycles that form in a
liquid when one surface is heated and But any bounded region has only a lim-
the opposite surface is kept cool. (See ited number of ways in which it can re-
Figure 4.) spond to an inflow of energy. We suggest
that it would be difficult if not im-
A number of interesting phenomena may
possible to formalize a principled dis-
be understood as dissipative structures.
tinction between the Bénard convection
Consider the distribution of water over
cycles that arise in a liquid when energy
the earth. Water is transported from
is pumped into it and the structured
place to place via processes that include
activities within a digital watch.31 The
evaporation, atmospheric weather sys-
primary difference seems to be that a di-
tem movements, precipitation, ground-
gital watch has a much more constrained
water flows, ocean current flows, etc.
static structure and can respond in far
Taken as a global system, these cycles
fewer ways.
may be understood as a dissipative struc-
ture that is driven primarily by solar en- Recall that we previously characterized
ergy, which is pumped into the earth’s Newtonian mechanics and the solid
atmosphere and surface structures. All of phase of matter as abstractions that mat-
this is played out against a static frame- ter implements under various conditions.
work defined and held in place by the We can do the same thing for dissipative
earth’s surface and its gravitational field. structures and say that a dissipative
structure appears within a bounded re-
We note that our definition of a dissipat-
gion when the materials within that re-
ive structure is quite broad. It includes
gion implement an energy-driven ab-
virtually any energy-consuming device
stract design.
that operates according to some design.
Consider a digital watch. It converts an An apparent difference between the ab-
inflow of energy into an ongoing series stract designs that dissipative structures
of structured internal activities. Does a implement and the abstract designs dis-
digital watch define a dissipative struc- cussed earlier is that the abstract designs
ture? One may argue that the design of a of dissipative structures seem to appear
digital watch limits the ways in which it unbidden—we don’t expect them—
can respond to an energy inflow. There- whereas the abstract designs discussed
fore the structured activity that arises as earlier are commonplace. The issue for
energy is pumped into it should not be the more commonplace abstract designs
characterized as a dissipative structure. is how to conceptualize them, not why
they appeared at all, whereas the abstract
design that appear as dissipative struc-
tures seem to demand an answer to the
question: why did they appear at all? In
fact, both kinds of abstract design are
part of nature. The difference is that
some are familiar; others aren’t.32
31
One of the other common examples of a
dissipative structure is the Belousov-Zhabotinsky
(BZ) reaction, which in some ways is a chemical
watch.
32
Reliance on a surprise factor as charac-
teristic of emergence is, in our opinion, a common
error. One often hears that emergent phenomena
them. They are often able to save energy ies are almost always at-equilibrium en-
and to chose in some sense when to use tities through which energy flows. In
it. other words, the nature of human-pro-
duced autonomous entities is that their
5.7 A naturally occurring autono- persistence as entities tends to be inde-
mous entity that is neither pendent of their use of the energy that
biological nor social flows through them. This tends not to be
We suggest that a hurricane qualifies as the case with naturally occurring
an autonomous entity. (See Figure 3.) In autonomous entities.
simple terms (paraphrased from
One of the senses of the word natural is
[NASA]), the internal design of a hur-
to have properties characteristic of ele-
ricane involves a greater than normal
ments found in nature. We suggest that
pressure differential between the ocean
the distinction between entities that rely
surface and the upper atmosphere. That
on an at-equilibrium frame and those
pressure differential causes moist surface
that more actively construct their frame-
air to rise. When the moisture-laden air
work is one of the central intuitive dif-
reaches the upper atmosphere, which is
ferences between what we call artificial
cooler, it condenses, releasing heat. The
and what we call natural. A hurricane
heat warms the air and reduces the pres-
would thus be considered a naturally oc-
sure, thereby maintaining the pressure
curring autonomous entity which is
differential—a marvelous design for a
neither biological nor social.
self-perpetuating process.
As an example of a naturally occurring
In effect, a hurricane is a heat engine in
at-equilibrium entity that becomes
which condensation, which replaces
autonomous, consider an atom that is be-
combustion as the source of heat, occurs
ing excited by a photon stream. Because
in the upper atmosphere.36 Thus, al-
of its design it captures the energy of the
though physically very large, a hurricane
photons, which it releases at some later
has a relatively simple design, which
time in what may be a slightly different
causes it to consume energy and which
form. This is the basis of the laser.
allows it to perpetuate itself as an area of
reduced entropy.
5.9 Autonomous entities and
phase transitions
5.8 Natural and artificial
Many autonomous entities exhibit the
autonomous entities
equivalent of phases—and phase trans-
Most of our energy consuming machines
itions. Such phases differ from phases in
also qualify as autonomous entities. The
at-equilibrium entities in that they reflect
primary difference between human pro-
different ways in which the autonomous
duced autonomous entities and naturally
entity makes use of the energy that is
occurring ones is that the naturally oc-
flowing through it. Examples include
curring autonomous entities use at least
gaits (walking, running, etc.), heart beats
some of the energy they consume to per-
(regular and fibrillation), and possibly
petuate themselves as entities. In con-
psychological conditions such as mania,
trast, human-produced autonomous entit-
depression and psychosis.
36
A characterization of hurricanes as “ver-
tical heat engines” may be found in Wikipedia.
The primary concern about global warm-
URL as of 9/1/2005: ing is not that the temperature will rise
http://en.wikipedia.org/wiki/Hurricane
part of a larger autonomous ship entity. Furthermore, the internal design of most
The ship along with its maintenance pro- autonomous entities enable them (a) to
cess is an entity because it is a self-per- store energy, (b) to move it about intern-
petuating region of reduced entropy. It is ally, and (c) to tap it as needed.
a relatively simple example of a social
autonomous entity. Both materials and 5.13 Autonomous entities tend
people cycle through it, but the process not to supervene over their
perpetuates itself by using energy from static components
the society in which it is embedded. As we said earlier, an at-equilibrium en-
tity consists of a fixed collection of com-
So our answer to the question of whether ponent elements over which it super-
“the same ship” is in the harbor from venes. In contrast, autonomous entities
year to year is “No” if we are thinking for the most part tend not to consist of a
about the material ship and “Yes” if we fixed collection of matter. Our hurricane
are thinking about the larger ship-plus- is a good example. A hurricane may be
maintenance entity. relatively stable as a reduced entropy re-
By our definition, the larger ship-plus- gion—even though its boundaries may
maintenance entity would be considered be somewhat vague. But however its
natural rather than artificial because it as boundaries are defined, the material
a social process and is not at-equilib- within its boundaries tends to vary from
rium; it uses some of the energy it con- moment to moment as the hurricane’s
sumes to perpetuate itself. We would winds move air and water about.
consider most social entities to be natur- Similarly, most biological entities re-
al in this sense even though they are con- cycle their physical components, and
structed and maintained by people. most social entities (e.g., families) and
economic entities (e.g., corporations) re-
5.12 Autonomous entities may act
main intact as the people who fill vari-
in the world
ous roles cycle through them. Theseus’s
As we know, hurricanes can cause signi-
ship—when understood as including its
ficant damage. So far we haven’t talked
maintenance process as discussed above
about how that might happen.
—is another example of an autonomous
Since energy flows through autonomous entity that recycles its physical compon-
entities, part of that flowing through in- ents.
volves flowing out. In other words,
Because of this recycling property, most
autonomous entities may include as part
autonomous entities don’t supervene
of their designs means for projecting
over any collection of matter that gives
force into the world by directing outward
us any intellectual leverage.
flows of energy.38
It is easiest to see this when we consider
gliders in the Game of Life, about which
38
This solves a problem that concerned this is true as well. In the Appendix we
Leibniz with respect to monads: how do they inter-
act. Leibniz’s answer was that they don’t. Our
show how to formalize the notion of a
autonomous entities interact with each other and Game of Life pattern. In simplest terms
with the rest of the world though energy flows we define what we call a live cell group
over which they have the ability to exert some
control. Of course our autonomous entities can ex-
to be a connected group of live (i.e.,
ert that control because they have internal designs; “on”) cells. We define a pattern as a con-
Leibniz’s monads didn’t.
Objects differ from agents in that they directly from inorganic materials. Simil-
don’t have what might be considered an arly, simple self-perpetuating organisms
internal source of energy. Agents do. But may not include any stable internal re-
our analogy breaks down entirely if an cord—like DNA—of their design (as
object is allowed to create a thread when hurricanes again illustrate). One would-
one of its methods is executed. (Most n’t expect to see evolution among such
multi-threaded programming languages organisms—at least not evolution that
allow the arbitrary creation of threads.) depends on modifications of such design
For an object to create a thread would be descriptions..
equivalent to an entity in nature creating
an unlimited internal source of energy 6 The evolution of complex-
for itself once it came in contact with ity
any external energy at all.
As we said, a real theory of thermody- 6.1 Stigmergy
namic computing is needed. Once one has autonomous entities (or
agents) that persist in their environment,
5.16 Minimal autonomous entities the ways in which complexity can devel-
In [Kauffman] Kauffman asks what the op grows explosively. Prior to agents, to
basic characteristics are of what he (also) get something new, one had to build it as
calls autonomous agents. He suggests a layer on top of some existing substrate.
that the ability to perform a thermody- As we have seen, nature has found a
namic (Carnot engine) work cycle is fun- number of amazing abstractions along
damental. with some often surprising ways to im-
plement them. Nonetheless, this con-
In what may turn out to be the same an- struction mechanism is relatively pon-
swer we suggest looking for the minimal derous. Layered hierarchies of abstrac-
biological organism that perpetuates it- tions are powerful, but they are not what
self by consuming energy. Bacteria seem one might characterize as lightweight or
to be too complex. Viruses41 and prions responsive to change. Agents change all
don’t consume energy.42 Is there anything that.
in between? We suggest that such a min-
imal autonomous entity may help us un- Half a century ago, Pierre-Paul Grasse
derstand the yet-to-be-discovered trans- invented [Grasse] the term stigmergy to
ition from the inanimate to the animate. help describe how social insect societies
function. The basic insight is that when
Since self-perpetuation does not imply the behavior of an entity depends to at
reproduction (as hurricanes illustrate), least some extent on the state of its en-
simple self-perpetuating organisms may vironment, it is possible to modify that
not be able to reproduce. That means entity’s behavior by changing the state of
that if they are to exist, it must be relat- the environment. Grasse used the term
ively easy for them to come into being “stigmergy” for this sort of indirect com-
41
Viruses are an interesting contrast to our munication and control. This sort of in-
lactose example, however. In both cases, an at-
equilibrium element in the environment triggers a
terplay between agents and their envir-
process in an autonomous entity. In the case of onment often produces epiphenomenal
lactose, the process is advantageous to the entity; effects that are useful to the agents. Of-
in the case of viruses, it is not advantageous to the
entity.
ten those effects may be understood in
42 terms of formal abstractions. Sometimes
Hurricanes aren’t biological.
it is easier to understand them less form- vironmental effects for each other as
ally. do sexes within species.
Two of the most widely cited examples • The “devo” side is even more stig-
of stigmergic interaction are ant foraging mergic. Genes, the switches that con-
and bird flocking. In ant foraging, ants trol gene expression, and the proteins
that have found a food source leave that genes produce when expressed
pheromone markers that other ants use to all have environmental effects on
make their way to that food source. In each other.
bird flocking, each bird determines how
it will move at least in part by noting the • Interestingly enough, the existence
positions and velocities of its neighbor- of gene switches was discovered in
ing birds. the investigation of another stigmer-
gic phenomenon. Certain bacteria
The resulting epiphenomena are that generate an enzyme to digest lactose,
food is gathered and flocks form. Pre- but they do it only when lactose is
sumably these epiphenomena could be present. How do the bacteria “know”
formalized in terms of abstract effects when to generate the enzyme?
that obeyed a formal set of rules—in the
same way that the rules for gliders and It turns out to be simple. The gene for
Turing Machines can abstracted away the enzyme exists in the bacteria, but
from their implementation by Game of its expression is normally blocked by a
Life rules. But often the effort required protein that is attached to the DNA se-
to generate such abstract theories doesn’t quence just before the enzyme gene.
seem worth the effort—as long as the This is called a gene expression
results are what one wants. switch.
Here are some additional examples of When lactose is in the environment, it
stigmergy. infuses into the body of the bacteria
and binds to the protein that blocks the
• When buyers and sellers interact in a expression of the gene. This causes the
market, one gets market epiphenom- protein to detach from the DNA
ena. Economics attempts to formal- thereby “turning on” the gene and al-
ize how those interactions may be lowing it to be expressed.
abstracted into theories.
The lactose enzyme switch is a lovely
• We often find that laws, rules, and illustration of stigmergic design. As we
regulations have both intended and described the mechanism above, it
unintended consequences. In this seems that lactose itself turns on the
case the laws, rules, and regulations switch that causes the lactose-digesting
serve as the environment within enzyme to be produced. If one were
which agents act. As the environment thinking about the design of such a
changes, so does the behavior of the system, one might imagine that the
agents. lactose had been designed so that it
would bind to that switch. But of
• Both sides of the evo-devo (evolu-
course, lactose wasn’t “designed” to
tion-development) synthesis [Car- do that. It existed prior to the switch.
roll] exhibit stigmergic emergence. The bacteria evolved a switch that
On the “evo” side, species create en- lactose would bind to. So the lactose
must be understood as being part of the Just as there in nothing in the rules of
environment to which the bacteria ad- the Game of Life about gliders, there is
apted by evolving a switch to which nothing in a computer’s instruction ex-
lactose would bind. How clever; how ecution cycle about exchanging the
simple; how stigmergic! values of x and y—or about any other
algorithm that software implements.
• Cellular automata operate stigmer-
Those effects are all epiphenomenal.
gically. Each cell serves as an envir-
onment for its neighbors. As we have • The instruction execution cycle itself
seen, epiphenomena may include is epiphenomenal over the flow of
gliders and Turing Machines. electrons through gates—which
knows no more about the instruction
• Even the operation of the Turing Ma-
execution cycle than the instruction
chine as an abstraction may be un-
execution cycle knows about al-
derstood stigmergically. The head of
gorithms.
a Turing Machine (the equivalent of
an autonomous agent) consults the In all of the preceding examples it is rel-
tape, which serves as its environ- atively easy to identify the agent(s), the
ment, to determine how to act. By environment, and the resulting epiphen-
writing on the tape, it leaves markers omena.
in its environment to which it may
return—not unlike the way foraging 6.2 Design and evolution
ants leave pheromone markers in It is not surprising that designs appear in
their environment. When the head re- nature. It is almost tautologous to say
turns to a marker, that marker helps that those things whose designs work in
the head determine how to act at that the environments in which they find
later time. themselves will persist in those environ-
ments. This is a simpler (and more ac-
• In fact, one may understand all com- curate) way of saying that it is the fit—
putations as being stigmergic with entities with designs that fit their envir-
respect to a computer’s instruction onment—that survive.
execution cycle. Consider the fol-
lowing familiar code fragment. 6.3 The accretion of complexity
An entity that suits its environment per-
temp:= sists in that environment. But anything
x; that persists in an environment by that
x := very fact changes that environment for
y; everything else. This phenomenon is
y := commonly referred to as an ever chan-
temp; ging fitness landscape.
What has been less widely noted in the
The epiphenomenal result is that x and
complexity literature is that when
y are exchanged. But this result is not
something is added to an environment it
a consequence of any one statement. It
may enable something else to be added
is an epiphenomenon of the three state-
latter—something that could not have
ments being executed in sequence by a
existed in that environment prior to the
computer’s instruction execution
earlier addition.
cycle.
This is an extension of notions from eco- This notion of building one thing upon
logy, biology, and the social sciences. A another applies to our framework in that
term for this phenomenon from the eco- anything that persists in an environment
logy literature, is succession. (See, for changes that environment for everything
example, [Trani].) Historically succes- else. The Internet provides many inter-
sion has been taken to refer to a fairly ri- esting illustrations.
gid sequence of communities of species,
• Because the Internet exists at all, ac-
generally leading to what is called a cli-
max or (less dramatically) a steady state. cess to a very large pool of people is
available. This enabled the develop-
Our notion is closer to that of bricolage, ment of websites such as eBay.
a notion that originated with the structur-
alism movement of the early 20th century • The establishment of eBay as a per-
[Wiener] and which is now used in both sistent feature of the Internet envir-
biology and the social sciences. Bricol- onment enabled the development of
age means the act or result of tinkering, enterprises whose only sales outlet
improvising, or building something out was eBay. These are enterprises with
of what is at hand. neither brick and mortar nor web
storefronts. The only place they sell
In genetics bricolage refers to the evolu- is on eBay. This is a nice example of
tionary process as one that tinkers with ecological succession.
an existing genome to produce
something new. [Church]. • At the same time—and again be-
cause the Internet provides access to
John Seely Brown, former chief scientist a very large number of people—oth-
for the Xerox Corporation and former er organizations were able to estab-
director of the Xerox Palo Alto Research lish what are known as massively
Center captured its sense in a recent talk. multi-player online games. Each of
these games is a simulated world in
[W]ith bricolage you appropri- which participants interact with the
ate something. That means game environment and with each
you bring it into your space, other. In most of these games, parti-
you tinker with it, and you re- cipants seek to acquire virtual game
purpose it and reposition it. resources, such as magic swords. Of-
When you repurpose ten it takes a fair amount of time, ef-
something, it is yours.43 fort, and skill to acquire such re-
Ciborra [Ciborra] uses bricolage to char- sources.
acterize the way that organizations tailor • The existence of all of these factors
their information systems to their chan- resulted, though a creative leap, in an
ging needs through continual tinkering. eBay market in which players sold
virtual game assets for real money.
43
In passing, Brown claims that this is how This market has become so large that
most new technology develops. there are now websites dedicated ex-
[T]hat is the way we build almost all clusively to trading in virtual game
technology today, even though my law- assets. [Wallace]
yers don't want to hear about it. We
borrow things; we tinker with them; we • BBC News reported [BBC] that
modify them; we join them; we build
stuff. there are companies that hire low-
wage Mexican and Chinese teen- that it may seem gratuitous even to com-
agers to earn virtual assets, which are ment on it. But it is an important phe-
then sold in these markets. How long nomenon, and one that has not received
will it be before a full-fledged eco- the attention it deserves.
nomy develops around these assets? Easy though this phenomenon is to un-
There may be brokers and retailers derstand once one sees it, it is not trivial.
who buy and sell these assets for After all, the second law of thermody-
their own accounts even though they namics tells us that overall entropy in-
do not intend to play the game. (Per- creases and complexity diminishes. Yet
haps they already exist.) Someone we see complexity, both natural and man
may develop a service that tracks the made, continually increasing. For the
prices of these assets. Perhaps fu- most part, this increasing complexity
tures and options markets will devel- consists of the development of new
op along with the inevitable invest- autonomous entities, entities that imple-
ment advisors. ment the abstract designs of dissipative
The point is that once something fits structures.
well enough into its environment to per- This does not contradict the Second
sist it adds itself to the environment for Law. Each autonomous entity maintains
everything else. This creates additional its own internally reduced entropy by us-
possibilities and a world with ever in- ing energy imported from the environ-
creasing complexity. ment to export entropy to the environ-
In each of the examples mentioned ment. Overall entropy increases. Such a
above, one can identify what we have process works only in an environment
been calling an autonomous entity. In that itself receives energy from outside
most cases, these entities are self-per- itself. Within such an environment, com-
petuating in that the amount of money plexity increases.
they extract from the environment (by Progress in science and technology and
selling either products, services, or ad- the bountifulness of the marketplace all
vertising) is more than enough to pay for exemplify this pattern of increasing
the resources needed to keep it in exist- complexity. One might refer to this kind
ence.
of pattern as a meta-epiphenomenon
In other cases, some Internet entities run since it is an epiphenomenon of the pro-
on time and effort contributed by volun- cess that creates epiphenomena.
teers. But the effect is the same. As long This creative process also tends to exhib-
as an entity is self-perpetuating, it be- it a second meta-epiphenomenon. Over-
comes part of the environment and can all energy utilization becomes continu-
serve as the basis for the development of ally more efficient. As new autonomous
additional entities. entities find ways to use previously un-
used or under-used energy flows (or
6.4 Increasing complexity in-
forms of energy flows that had not exis-
creasing efficiency, and his-
ted until some newly created autonom-
torical contingency
ous entity generated them, perhaps as a
The phenomenon whereby new entities
waste product), more of the energy
are built on top of existing entities is
available to the system as a whole is put
now so widespread and commonplace
to use.
The process whereby new autonomous al. Even though the activities of higher
entities come into existence and perpetu- level entities may be described in terms
ate themselves is non-reductive. It is cre- that are independent of the forces that
ative, contingent, and almost entirely a produce them (recall that this is our
sequence of historical accidents. As they definition of epiphenomenal), the funda-
say, history is just one damn thing after mental forces of physics are the only
another—to which we add, and nature is forces in nature. There is no strong
a bricolage. We repeat the observation emergence. All other force-like effects
Anderson made more than three decades are epiphenomenal.
ago. Consequently we find ourselves in the
The ability to reduce position of claiming that the higher level
everything to simple funda- sciences study epiphenomenal interac-
mental laws [does not imply] tions among real if often somewhat eth-
the ability to start from those ereal entities.
laws and reconstruct the uni-
verse. 7.2 Science and emergence
The idea that one can use heat and the
7 Entities, emergence, and expansion of gases that it produces to
science implement a particular function is not a
concept of fundamental physics. Of
7.1 Entities and the sciences course the Carnot engine is a conse-
One reason that the sciences at levels quence of fundamental physics, but it is
higher than physics and chemistry seem not a concept of fundamental physics.
somehow softer than physics and chem- The idea of using a force to implement
istry is that they work with autonomous new functionality is simply not within
entities, entities that for the most part do the realm of fundamental physics.
not supervene over any conveniently Physics, like most science, does not con-
compact collection of matter. Entities in sider new functionality. It examines ex-
physics and chemistry are satisfyingly isting phenomena, and it asks how they
solid—or at least they seemed to be be- are brought about. It does not ask how
fore quantum theory. In contrast, the en- knowledge gained from such an analysis
tities of the higher level sciences are not can be used to implement something
defined in terms of material boundaries. new.
These entities don’t exist as stable
clumps of matter; it’s hard to hold them Here is a representative definition of the
completely in one’s hand—or in the grip term science.
of an instrument. • The observation, identifica-
The entities of the special sciences are tion, description, experi-
objectively real—there is some objective mental investigation, and
measure (their reduced entropy relative theoretical explanation of
to their environment) by which they phenomena. [American Her-
qualify as entities. But as we saw earlier, itage]
the processes through which these entit- Science is thus the study of nature, how
ies interact and by means of which they it is designed, i.e., organized, and how
perpetuate themselves are epiphenomen- its designs work. Science does not have
it was invented. Laplace’s demon may phenomena are often of significant in-
even succeed in a quantum mechanical terest. We also said that formalization
universe in that the quantum wave equa- may not always be in the cards. Soft-
tion is deterministic—even though it ware, which one would imagine to be a
characterizes probability amplitudes and perfect candidate for formalization, now
hence its collapse is not. But if nature in- seems to be a good example of an epi-
cludes asynchronously acting autonom- phenomenon that is unlikely to be form-
ous entities, some of which may them- alized.
selves embody quantum probability It had once been hoped that software de-
transitions, many of which are mutually velopment could evolve to a point at
interdependent, and all of which depend which one need only write down a form-
on their environment, which includes al specification of what one wanted the
other autonomous entities for their oper- software to do. Then some automatic
ation and persistence, Laplace’s demon process would produce software that sat-
will be way beyond its depth. isfied that specification.
One possible simple formal model for That dream now seems quite remote. Be-
such a computational system is a shared sides the difficulty of developing (a) a
tape Turing Machine community: a col- satisfactory specification language and
lection of asynchronously operating Tur- (b) a system that can translate specifica-
ing Machines that share a single tape.45 tions written in such a language into ex-
Some proponents of agent-based model- ecutable code, the real problem is that it
ing argue for that approach on the has turned out to be at least as difficult
grounds that even though some domains and complex to write formal specifica-
may have global characterizations, those tions as it is to write the code that pro-
characterizations are much too complex duces the specified results.
to compute. Our position is that agent- Even if one could write software by
based modeling is appropriate because writing specifications, in many cases—
that’s how nature is. especially cases that involve large and
complex systems, the kinds of cases for
9 Some practical considera- which it really matters—doing so does-
tions n’t seem to result in much intellectual
leverage, if indeed it produces any at all.
9.1 Emergence and software
As noted earlier, the computation that This illustrates quite nicely that we often
results when software is executed is find ourselves in the position of wanting
emergent. It is an epiphenomenon of the to produce epiphenomena (epiphenom-
operation of the (actual or virtual) ma- ena, which may be very important to us),
chine that executes the software. whose formalization as an abstraction
we find to be either infeasible or not par-
Earlier we defined emergence as syn- ticularly useful.
onymous with epiphenomenon. At that
time we suggested that formalizable epi- 9.2 Bricolage as design
45
Wegner’s work [Wegner] on non-tradi-
The process of building one capability
tional Turing Machine models begins to explore on top of another not only drives the
his own models. Cockshott and Michaelson [Cock- overall increase in complexity, it also
shott] dispute whether Wegner’s models extend the
power of the Turing machine.
provides guidance to designers about
how to do good design work. Any good onment, i.e., our infrastructure, and not
designer—a developer, an architect, a just a piece of closed and isolated func-
programmer, or an engineer—knows that tionality.
it is often best if one can take advantage By infrastructure we mean systems such
of forces and processes already in exist- as the Internet, the telephone system, the
ence as part of one’s design. electric power distribution system, etc.
But even before engineering, we as hu- Each of these systems can be character-
man beings made use of pre-existing ized in isolation in terms of the particu-
capabilities. Agriculture and animal hus- lar functions they perform. But more im-
bandry use both plant reproduction and portant than the functional characteriza-
such animal capabilities as locomotion tion of any of these individual systems is
or material (i.e., skin) production for our the fact that they exist in the environ-
own purposes. The exploitation of exist- ment in such a way that other systems
ing capabilities for our own purposes is can use them as services.
not a new idea. We should apply this perspective to all
An interesting example of this approach new systems that we design: design them
to engineering involves recent develop- as infrastructure services and not just as
ments in robotics. Collins reported bits of functionality. Clearly Microsoft
[Collins] that a good way to make a ro- understands this. Not only does it posi-
bot walk is by exploiting gravity through tion the systems it sells as infrastructure
what he called passive-dynamic motion services, it also maintains tight owner-
—raise the robot’s leg and let gravity ship and control over them. When such
pull it back down—rather than by direct- systems become widely used elements of
ing the robot’s limbs to follow a pre- the economy, the company makes a lot
defined trajectory. of money. The tight control it maintains
and the selfishness with which it controls
This illustrates in a very concrete way
these systems earns it lots of resentment
the use of an existing force in a design.
as well. Society can’t prosper when any
Instead of building a robot whose every
important element of its infrastructure is
motion was explicitly programmed,
controlled primarily for selfish purposes.
Collins built a robot whose motions were
controlled in part by gravity, a pre-exist- The US Department of Defense (DoD) is
ing force. currently reinventing itself [Dick] to be
more infrastructure-centric. This requires
9.3 Infrastructure-centric devel- it to transform what is now a
opment
huge collection of independent
Building new capabilities on top of ex-
isting ones is not only good design, it is “stovepipe” information sys-
highly leveraged design. But now that tems, each supporting only its
we are aware of this strategy a further original procurement specifica-
lesson can be drawn. New systems tion, to a unified assembly of
should be explicitly designed to serve as interoperating systems.
a possible basis for systems yet to come. The evocative term stovepipe is intended
Another way of putting this is that every to distinguish the existing situation—in
time we build a new system, it should be which the DoD finds that it has acquired
built so that it becomes part of our envir- and deployed a large number of func-
increasing number of disaster planning tion at all.) There certainly are models of
services being offered. population size effects in predator-prey
simulations. But by biological arms
9.6 Modeling: the difficulty of races we are talking about not just popu-
looking downward lation sizes but actual evolutionary
The perspective we have described changes.
yields two major implications for model-
Imagine a situation in which a plant spe-
ing. We refer to them as the difficulty of
cies comes under attack from an insect
looking downwards and the difficulty of
species. In natural evolution the plant
looking upwards. In both cases, the
may “figure out” how to grow bark. Can
problem is that it is very difficult to
we build a computer model in which this
model significant creativity—notwith-
solution would emerge? It is very un-
standing the fact that surprises do appear
likely. To do so would require that our
in some of our models. In this section we
model have built into it enough informa-
examine the difficulty of looking down-
tion about plant biochemistry to enable it
ward. In the next we consider the diffi-
to find a way to modify that biochem-
culty of looking upward.
istry to produce bark, which itself is
Strict reductionism, our conclusion that defined implicitly in terms of a surface
all forces and actions are epiphenomenal that the insect cannot penetrate.
over forces and actions at the funda- Evolving bark would require an enorm-
mental level of physics, implies that it is ous amount of information—especially
impossible to find a non-arbitrary base if we don’t want to prejudice the solu-
level for models. One never knows what tion the plant comes up with.
unexpected effects one may be leaving
The next step, of course, is for the insect
out by defining a model in which inter-
to figure out how to bore through bark.
actions occur at some non-fundamental
Can our model come up with something
level.
like that? Unlikely. What about the
Consider a model of computer security. plant’s next step: “figuring out” how to
Suppose that by analyzing the model one produce a compound that is toxic to the
could guarantee that a communication insect? That requires that the model in-
line uses essentially unbreakable encryp- clude information about both plant and
tion technology. Still it is possible for insect biochemistry—and how the plant
someone inside to transmit information can produce a compound that interferes
to someone outside. with the insect’s internal processes. This
How? By sending messages in which the would be followed by the development
content of the message is ignored but the by the insect of an anti-toxin defense.
frequency of transmission carries the in- To simulate this sort of evolutionary pro-
formation, e.g., by using Morse code. cess would require an enormous amount
The problem is that the model didn’t in- of low level detail—again especially if
clude that level of detail. This is the we don’t want to prejudice the solution
problem of looking downward. in advance.
A further illustration of this difficulty is Other than Tierra (see [Ray]) and its suc-
that there are no good models of biolo- cessors, which seem to lack the richness
gical arms races. (There don’t seem to be to get very far off the ground, as far as
any good models of significant co-evolu- we know, there are no good computer
—and operational, i.e., they actually ergy wells are reasonably well under-
function in the real world. stood—although it is astonishing how
many different epiphenomena nature and
• Computer Science has figured out
technology have produced through the
how to represent information in data- use of energy wells.
bases in ways that allow that inform-
ation to hang together meaningfully. We have not even begun to catalog the
ways in which energy-consuming pro-
• Computer Science has faced—and to cesses may be used to construct stable,
a significant extent resolved—the self-perpetuating, autonomous entities.
problem of working on many levels
of abstraction and complexity simul- Earlier we wrote that science does not
taneously. consider it within its realm to ask con-
structivist questions. That is not com-
If insights gained from these and other pletely true. Science asks about how we
intellectual wrestling matches can be ap- got here from the big bang, and science
plied in a wider context, it is only Com- asks about biological evolution. These
puter Science paying back the debt that are both constructivist questions. Since
it owes to the engineers, scientists, math- science is an attempt to understand
ematicians, and philosophers who set the nature, and since constructive processes
stage for and participated in its develop- occur in nature, it is quite consistent with
ment. the overall goals of science to ask how
these constructive processes work. As far
10.2 Constructive science as we can determine, there is no sub-dis-
For most of its history, science has pur- cipline of science that asks, in general,
sued the goal of explaining existing phe- how the new arise from the existing.
nomena in terms of simpler phenomena.
That’s the reductionist agenda. Science has produced some specialized
answers to this question. The biological
The approach we have taken is to ask evolutionary explanation involves ran-
how new phenomena may be construc- dom mutation and crossover of design
ted from and implemented in terms of records. The cosmological explanation
existing phenomena. That’s the creative involves falling into energy wells of
impulse of artists, computer scientists, various sorts. Is there any more to say
engineers—and of nature. It is these new about how nature finds and then explores
phenomena that are often thought of as new possibilities? If as Dennett argues in
emergent. [Dennett ‘96] this process may be fully
When thinking in the constructive direc- explicated as generalized Darwinian
tion, a question arises that is often under- evolution, questions still remain. Is there
appreciated: what allows one to put ex- any useful way to characterize the search
isting things together to get something space that nature is exploring? What
new—and something new that will per- search strategies does nature use to ex-
sist in the world? What binding forces plore that space? Clearly one strategy is
and binding strategies do we (and human inventiveness.
nature) have at our disposal?
Our answer has been that there are two 11 Acknowledgement
sorts of binding strategies: energy wells We are grateful for numerous enjoyable
and energy-consuming processes. En- and insightful discussions with Debora
Shuger during which many of the ideas Brown, J.S., Talk at San Diego State
in this paper were developed and re- University, January 18, 2005. URL as of
fined. 6/2005: http://ctl.sdsu.edu/pict/jsb_lec-
ture18jan05.pdf
We also wish to acknowledge the fol-
lowing websites and services, which we Carroll, S.B., Endless Forms Most
used repeatedly. Beautiful: The New Science of Evo Devo
and the Making of the Animal Kingdom,
• Google (www.google.com);
W. W. Norton, 2005.
• The Stanford Encyclopedia of Philo-
sophy (plato.stanford.edu); Chaitin, G. Algorithmic Information
• OneLook Dictionary Search Theory, reprinted 2003. URL as of Sept.
(onelook.com). 6, 2005:
http://www.cs.auckland.ac.nz/CDMTCS/
References chaitin/cup.pdf.
Abbott, R., “Emergence, Entities, En- CFCS, Committee on the Fundamentals
tropy, and Binding Forces,” The Agent of Computer Science: Challenges and
2004 Conference on: Social Dynamics: Opportunities, National Research Coun-
Interaction, Reflexivity, and Emergence, cil, Computer Science: Reflections on
Argonne National Labs and University the Field, Reflections from the Field,
of Chicago, October 2004. URL as of 2004. URL as of 9/9/2005:
4/2005: http://abbott.calstatela.edu/Pa- http://www.nap.edu/books/0309093015/
persAndTalks/abbott_agent_2004.pdf. html/65.html.
American Heritage, The American Her- Church, G. M., “From systems biology
itage® Dictionary of the English Lan- to synthetic biology,” Molecular
guage, 2000. URL as of 9/7/2005: Systems Biology, March, 29, 2005. URL
http://www.bartleby.com/61/67/S014670 as of 6/2005:
0.html. http://www.nature.com/msb/journal/v1/n
Anderson, P.W., “More is Different,” 1/full/msb4100007.html.
Science, 177 393-396, 1972. Ciborra, C. "From Thinking to
BBC News, “Gamer buys $26,500 virtu- Tinkering: The Grassroots of Strategic
al land,” BBC News, Dec. 17, 2004. Information Systems", The Information
URL as of 2/2005: http://news.b- Society 8, 297-309, 1992.
bc.co.uk/1/hi/technology/4104731.stm. Clarke, A. C., "Hazards of Prophecy: The
Bedau, M.A., “Downward causation and Failure of Imagination,” Profiles of The
the autonomy of weak emergence”. Future, Bantam Books, 1961.
Principia 6 (2002): 5-50. URL as of Cockshott, P. and G. Michaelson, “Are
11/2004: http://www.reed.edu/~mab/pa- There New Models of Computation:
pers/principia.pdf. Reply to Wegner and Eberbach.” URL as
Boyd, Richard, "Scientific Realism", of Oct. 10, 2005: http://www.dc-
The Stanford Encyclopedia of Philo- s.gla.ac.uk/~wpc/reports/wegner-
sophy (Summer 2002 Edition), Edward 25aug.pdf.
N. Zalta (ed.), URL as of 9/01/2005: Colbaugh, R. and Kristen Glass, “Low
http://plato.stanford.edu/archives/sum20 Cognition Agents: A Complex Networks
02/entries/scientific-realism/. Perspective,” 3rd Lake Arrowhead Con-
Prigogine, Ilya and Dilip Kondepudi, Shalizi, C., “Review of Emergence from
Modern Thermodynamics: from Heat Chaos to Order,” The Bactra Review,
Engines to Dissipative Structures, John URL as of 6/2005:
Wiley & Sons, N.Y., 1997. http://cscs.umich.edu/~crshalizi/reviews/
holland-on-emergence/
Ray, T. S. 1991. “An approach to the
synthesis of life,” Artificial Life II, Santa Shalizi, C., “Emergent Properties,”
Fe Institute Studies in the Sciences of Notebooks, URL as of 6/2005: http://csc-
Complexity, vol. XI, Eds. C. Langton, C. s.umich.edu/~crshalizi/notebooks/emer-
Taylor, J. D. Farmer, & S. Rasmussen, gent-properties.html.
Redwood City, CA: Addison-Wesley, Smithsonian Museum, “Chimpanzee
371--408. URL page for Tierra as of Tool Use,” URL as of 6/2005: http://na-
4/2005: tionalzoo.si.edu/Animals/ThinkTank/Too
http://www.his.atr.jp/~ray/tierra/. lUse/ChimpToolUse/default.cfm.
Rendell, Paul, “Turing Universality in Summers, J. “Jason’s Life Page,” URL
the Game of Life,” in Adamatzky, An- as of 6/2005:
drew (ed.), Collision-Based Computing, http://entropymine.com/jason/life/.
Springer, 2002. URL as of 4/2005:
http://rendell.server.org.uk/gol/tmde- Trani, M. et. al., “Patterns and trends of
tails.htm, early successional forest in the eastern
http://www.cs.ualberta.ca/~bulitko/F02/p United States,” Wildlife Society Bullet-
apers/rendell.d3.pdf, and http://www.c- in, 29(2), 413-424, 2001. URL as of
s.ualberta.ca/~bulitko/F02/papers/tm_wo 6/2005:
rds.pdf http://www.srs.fs.usda.gov/pubs/rpc/200
2-01/rpc_02january_31.pdf.
Rosen, Gideon, "Abstract Objects", The
Stanford Encyclopedia of Philosophy University of Delaware, Graduate Col-
(Fall 2001 Edition), Edward N. Za- lege of Marine Studies, Chemosynthesis,
lta (ed.), URL as of 9/1/05: http://pla- URL as of Oct 10, 2005:
to.stanford.edu/archives/fall2001/entries/ http://www.ocean.udel.edu/deepsea/level
abstract-objects/. -2/chemistry/chemo.html
Sachdev, S, “Quantum Phase Trans- Uvarov, E.B., and A. Isaacs, Dictionary
itions,” in The New Physics, (ed G. of Science, September, 1993. URL as of
Fraser), Cambridge University Press, (to 9/7/2005:
appear 2006). URL as of 9/11/2005: http://oaspub.epa.gov/trs/trs_proc_qry.na
http://silver.physics.harvard.edu/newphy vigate_term?p_term_id=29376&p_term
sics_sachdev.pdf. _cd=TERMDIS.
Shalizi, C., Causal Architecture, Com- Varzi, Achille, "Boundary", The Stan-
plexity and Self-Organization in Time ford Encyclopedia of Philosophy (Spring
Series and Cellular Automata, PhD. Dis- 2004 Edition), Edward N. Zalta (ed.),
sertation, Physics Department, Uni- URL as of 9/1/2005: http://plato.stan-
versity of Wisconsin-Madison, 2001. ford.edu/archives/spr2004/entries/bound
URL as of 6/2005: ary/.
http://cscs.umich.edu/~crshalizi/thesis/si Varzi, A., "Mereology", The Stanford
ngle-spaced-thesis.pdf Encyclopedia of Philosophy (Fall 2004
Edition), Edward N. Zalta (ed.), URL as
an LCG are determined only by other For any LCG l 0, consider a sequence of
cells in the same LCG.48 successor LCGs generated in this man-
ner:
Suppose that an LCG contains the only
live cells on a Game of Life grid. Con- l0 -> l1 -> l 2 -> l3 -> … .
sider what the mapping of that LCG by
the Game of Life rules will produce. Extend such a sequence until one of
There are three possibilities. three conditions occurs.
1. The live cells may all die. 1. There are no successor LCGs, i.e.,
Game of Life(li) is empty—all the
2. The successor live cells may consist
live cells in the final LCG die. Call
of a single LCG—as in a glider or
these terminating sequences.
still life.
2. There is a single successor LCG, i.e.,
3. The successor live cells may parti-
Game of Life(li) = {lk}, but that suc-
tion into multiple LCGs—as in the
cessor LCG is in the same state as an
so-called bhepto pattern, which starts
LCG earlier in the sequence, i.e., lk =
as a single LCG and eventually sta-
lj, j < k. Call these repeating se-
bilizes as 4 still life LCGs and two
quences.
glider LCGs.
3. The set Game of Life(li) of successor
In other words, the live cells generated LCGs contains more than one LCG,
when the Game of Life rules are applied i.e., the LCG branches into two or
to an LCG will consist of 0, 1, or mul- more LCGs. Call these branching se-
tiple successor LCGs. quences.
More formally, if l is an LCG, let Game Note that some LCG sequences may
of Life(l) be the set of LCGs that are never terminate. They may simply pro-
formed by applying the Game of Life duce larger and larger LCGs. The so-
rules to the cells in l. For any particular called spacefiller pattern, which actually
l, Game of Life(l) may be empty; it may consists of multiple interacting LCGs,
be contain a single element; or it may one of which fills the entire grid with a
contain multiple elements. If l’ is a single LCG as it expands,49 is an amazing
member of Game of Life(l) write l -> l’. example of such a pattern. I do not know
if there is an LCG that expands without
limit on its own. If any such exist, call
these infinite sequences.
48
In particular, no LCG cells have live neighbors For any LCG l0, if the sequence
that are outside the LCG. Thus no cells outside
the LCG need be considered when determining l0 -> l1 -> l2 -> l3 -> … .
the GoL transitions of the cells in an LCG. A
dead boundary cell may become live at the is finite, terminating in one of the three
next time-step, but it will do so only if three of ways described above, let seq(l0) be that
its neighbors within the LCG are live. Its
neighbors outside the LCG are guaranteed to
sequence along with a description of
be dead. how it terminates. If
50
A simple fuse pattern is a diagonal configuration
of live cells. At each time step, the two end cells
die; the remaining cells remain alive. A simple
51
fuse pattern may be augmented by adding more Many much more practical and efficient
complex features at one end, thereby building a programs have been written to search for patterns
pattern that becomes active when the fuse exhausts in the GoL and related cellular automata. See
itself. Such a pattern can be built with an arbitrar- http://www.ics.uci.edu/~eppstein/ca/search.html
ily long fuse. for a list of such programs.
12.4 Game of Life patterns: com- Even though the set BP of basic patterns
binations of basic patterns is infinite, since each LCG is finite, by
Many of the interesting Game of Life using a technique similar to that used for
patterns arise from interactions between generating BP itself, one can (very tedi-
and among basic patterns. For example, ously) enumerate all the possible BP in-
the first pattern that generated an unlim- teractions.
ited number of live cells, the glider gun, More formally, let Pf(BP) be the set of
is a series of interactions among combin-
all finite subsets of BP. For each member
ations of multiple basic patterns that cyc-
of Pf(BP) consider all possible (still
lically generate gliders.
only a finite number) relative configura-
To characterize these more complex pat- tions of its members on the grid so that
terns it is necessary to keep track of how there will be some interaction among
basic patterns interact. In particular, for them at the next time step. One can then
each element in BP, augment its descrip- record all the possible interactions
tion with information describing among finite subsets of BP.
a) its velocity (rate, possibly zero, and These interactions would be equivalent
direction) across the grid, to the APIs for the basic patterns. We
could call a listing of them BP-API.
b) if it cycles, how it repeats, i.e., which Since BP is itself infinite, BP-API would
states comprise its cycle, and also be infinite. But BP-API would be
c) if it branches, what the offspring ele- effectively searchable. Given a set of
ments are and where they appear rel- elements in BP, one could retrieve all the
ative to final position of the terminat- interactions among those elements. BP-
ing sequence. API would then provide a documented
starting point for using the Game of Life
Two or more distinct members of BP as a programming language.
that at time step i are moving relative to
each other may interact to produce one As in traditional programming lan-
or more members of BP at time step i+1. guages, as more complex interactions are
The result of such a BP “collision” will developed, they too could be docu-
generally depend on the relative posi- mented and made public for others to
tions of the interacting basic patterns. use.