Information about http://alife.org/alife8/proceedings/sub3217.pdf
in Artificial Life VIII, Standish, Abbass, Bedau (eds)(MIT Press) 2002…
Tags: abbass, artificial life, atoms, australia, bigpond, consciousness, constituent, eds, emergence, emergent property, inadequate level, mechanisms, organisms, reductionist, riley st, st surry hills, stephen jones, surry hills nsw,Pages: 5
Language: english
Display cached document
Page 1
Page 2
Page 3
Page 4
Page 5
in Artificial Life VIII, Standish, Abbass, Bedau (eds)(MIT Press) 2002. pp 418422 1
Organizing Relations and Emergence
Stephen Jones
387 Riley St, Surry Hills,
NSW, 2010, Australia
s jones1@bigpond.com
Abstract Most authors on emergence seem to want to set up
an understanding of emergence such that a bottom-up,
Emergence is largely used as an explanation: such and micro-physical-properties basis for its explanation can-
such an object -- ranging from atoms through multi- not operate and that only a non-reductionist consider-
cellular organisms to consciousness -- is an emergent
property of some ensemble of parts. But this leaves an ation of the behaviors of objects should obtain. I want
inadequate level of understanding, making the term a to explore this issue here and propose a way in which
representation of something almost mystical. For emer- the explanatory value of emergence can be made con-
gence to be useful an understanding of the mechanisms sistent with a micro-physical causality understanding of
of emergence must be brought out. This paper proposes the world. Thus, I will argue that the proper study of
a system of orders of organizing relations that are the
means by which complex objects "emerge" from the in- emergence is in the study of the organizing relations that
teractions of their constituent parts. operate to link the parts of an ensemble. Artificial Life,
then, is the study of the organizing relations that pro-
duce the emergence of coherent entities in computational
Introduction situations.
The term emergence has a somewhat equivocal press.
For some it is an explanation for the appearance of Kinds of Emergence
wholes or coherent objects in any system of things or For most philosophers there are two kinds of emergence:
understandings; while for others it is either meaning- ontological and representational (Searle 1992). Ontolog-
less or, conversely, an invocation of something mystical. ical emergence covers explanations for how objects and
The term is applied to the appearance of novel, coher- organisms can exist in the world given thermodynam-
ent objects that are not predictable from the isolated ics and a causally closed microphysics. Representational
properties of the system's parts, functioning as a short- emergence covers the development of theories about the
hand for the development of new levels in a hierarchy of things which we are able to observe in the world. Car-
organization of complex and adaptive entities. Novelty iani (1991) includes computational emergence, in which
and coherence are the primary properties of emergent "complex global forms can arise from local computa-
objects. Two questions arise here: tional interactions" (p.776) thus modelling similar pro-
cesses to those which, in material systems, might pro-
1. How can a new thing come into being if it cannot be
duce actual emergent objects. It appears, for example, in
predicted by the properties of the stuff that constitutes work on cellular automata and in Holland's work on com-
it? and
plex adaptive systems (CAS). Not only can cellular au-
2. What confers coherence on something made up of tomata replicate themselves, but they produce "gliders",
many kinds of parts so that it has obvious existence or coherent groupings of cells that emerge from a few ex-
as a whole thing? plicitly specified rules. Holland probes the properties of
CAS and argues that aggregation and self-maintenance
Ultimately the question of emergence is the problem of are specifically relevant to the study of emergence. Ag-
how any kind of organized system can come into being. gregation is a function of organizational hierarchy, and
How can a collection of things self-organize? Where does self-maintenance involves the continued coherence of an
the ordered arrangement of the parts, the thing's organi- aggregation despite the flow of resources through it and
zation, come from? How can it self-regulate or maintain the perpetual turnover of its constituent parts (Holland
that organization in the face of the entropy in its envi- 1995).
rons? These questions apply particularly to biological Here I will consider the emergence of physical objects
entities. and systems from the matter of their constituents, and
2 in Artificial Life VIII, Standish, Abbass, Bedau (eds) (MIT Press) 2002. pp 418422
its simulation in computational emergence, rather than b: unpredictability -- The unpredictability of the
dwell on representational emergence. One of the reasons properties of something is at the very basis of calling
emergence is adopted as an explanation is that it seems it emergent (Broad 1925).
that there are numerous physical ensembles which do not
predict the behaviors that these systems, when suitably c: coherence, integrity -- Objects that are "held to-
constituted, may demonstrate. Any climb up the levels gether by causal interactions that constitute their or-
of the biological orders produces novel emergent systems, ganic unity . . . act[ing] coherently and resist[ing] in-
e.g. many behaviors of a cell are not predicted by the ternal and external fluctuations". (Collier 1998)
isolated properties of any of its macromolecular compo-
d: self-maintenance -- Contingent stability with re-
nents. In fact many of the behaviors of those macro-
spect to variations in the environment. Self-
molecules don't even come into operation until they are
maintenance is part of the cohesive nature of an emer-
integrated into a cell.
gent system (Collier 1998).
There are two flavours of ontological emergence:
Weak emergence, where an object having emergent e: causal asymmetry -- The emergence of "novel
properties is physically determined by the properties causal properties" is an essential criterion for emer-
of its lower level constituents, yet would not exist gence.
as such without those emergent properties. (Collier
1998). e:i) downward causation -- Properties of the con-
stituent sub-systems of an emergent system only re-
Radical emergence, where the emergent properties of vealed through its emergence.
the whole are metaphysically incompatible with the
properties and relations of its parts, perhaps involving e:ii) non-linearity -- Step-functions, hysteresis and
the appearance of a totally new (e.g. mental) stuff. boundary development in far-from-equilibrium sys-
tems.
In the view canvassed here, the kinds of things that
radical emergence is used to "explain" are accounted for Suitable consideration of these integrative properties
in the roles of the feedback relations discussed in the within levels of an organizational hierarchy, implies that
Taxonomy section. al-though a full description of the processes involved in
any particular emergence may be intractable, they are,
Characteristics of Emergent Objects in principle, precisely explainable in a reductionist pro-
Emergence is best represented as a jump in hierarchi- cedure that acknowledges the organizational hierarchy
cal level of organizational structure of the parts of some of things. For example, in the deconstruction of vital-
system such that they become coherently organized and ism as the principle that brought inanimate matter to
might be characterized as being something with a new life in biology, it was recognized that the organizing re-
name. The problem of emergence becomes the prob- lations that had been attributed to some vitalistic (and
lem of whether it is meaningful to talk about hierar- radically non-reducible) principle were in fact the proper
chical levels of organization in systems, how one might study of biology (Needham 1936). So by what mecha-
describe the boundaries between levels and how a col- nisms do these properties arise in making an ensemble
lection of constituents can actually become a new level. of parts into an emergent whole?
These issues concern every level of science from the jump
from quarks and gluons to nucleons and atoms in micro- Nature of Organizing Relations.
physics, to the difference between collections of organic
The organization of the constituents of a system is a re-
molecules and cells, to the organization of individual hu-
sult of the relations that operate in the physical world,
mans into societies. Emergence thus becomes the ques-
or in the rules of procedure adopted for some compu-
tion of organization and I will proffer a taxonomy of
tational simulation. Emergent objects depend on those
relations by which organization might arise in a collec-
organizing relations that can actually be operating in the
tion of parts giving it integrity, coherence and the status
physical world or in the rules of procedure adopted for
of a whole. But we must first cover some of the proper-
some computational simulation. Explanations for emer-
ties of organized wholes that persuade us that they are
gences require description of the relations that link ob-
emergent.
jects within some level of an integration and between
Emergent systems can be characterised as dynamical
hierarchical levels of order, but we need a definition of
processes showing:
order. Von Foerster suggested that order allows us "to
a: novelty -- Instances of the first time some thing ap- account for apparent relationships between elements of a
pears in the universe, or the emergence of something set which would impose some constraints as to the pos-
new with every instance of a particular organization sible arrangements of the elements of this system. As
of constituents (Bickhard 2000). the organization of the system grows, more and more
in Artificial Life VIII, Standish, Abbass, Bedau (eds)(MIT Press) 2002. pp 418422 3
of these relations should become apparent". (von Fo- sensing to find food, to gauge distance to neighbours,
erster 1960, p.37). He then derives a relation between and other means of gathering environmental informa-
order and entropy such that for a system to be ordered tion. The rules in Reynolds' (1994) boids, each func-
it must carry less entropy than the maximum possible tioning in a feedforward only manner for each boid
entropy of a set of the same elements not in any way so in a local region, allow a graded set of relationships
organized, i.e. a system wherein the elements are not in throughout the larger global flock such that the au-
any relations (other than random spatial relations) with tonomous behaviour of each boid produces an overall
each other. So to produce order in a system it must pos- coherence in the flock.
sess relations among its elements which have the effect The emission of signals may also be a feedforward only
of reducing the indistinguishability of states of the sys- relation, but emission begs the question of to what
tem, thus organizing it. Whitehead defined relations as end? Is it simply a marker, as with MacLennan's
"abstractions from contrasts. A relation can be found (1991) ants, or is it intentional behaviour leading to
in many contrasts; and when it is so found, it is said to mutual interactions such as communication? Other
relate the things contrasted". (Whitehead 1929, p.349) feedforward relations include learning by repetition,
where a contrast is a difference between two perceptions or concentration gradient following in cellular develop-
such as the contrast between red and blue. ment, but these may also be a function of third order
Now, if the system so organized becomes capable of mutualistic feedbacks.
some level of stability such that it develops constrained
regions having boundaries and that for perturbations to Second order: circular causal Feedback relations:
its elements or its boundaries to be damaging they must Feed-back is a function of sensing and enables an en-
be greater than a certain threshold, then that system can tity to regulate its behaviour. It arises when a signal
be said to be integrated (Ashby 1952). This integrity emitted into the environment produces some impact
gives a system its emergent condition as a new order of back onto the originating entity and some aspect of
object and we can then go on to describe various levels that signal provides information to it about its on-
of order in which a set of integrated elements at one level going presence in that environment. Feedback often
become the parts which, in utilizing further kinds of or- involves learning when the individual modulates its
ganizing relations, constitute a new, higher level order. own behaviour, by (1) emission of a signal for later
These organizing relations are complex networks of in- reacquisition, or (2) alteration of a signal or marker
teractions among, for example, the physico-chemical and by an individual for its own purposes. Relations be-
biochemical entities of biology or the neuro-anatomical tween a system's internal self-regulation and its en-
structures of the brain or the individuals of a society. vironment, afforded by feed-back, make the system
possible, emerging from the soup as a distinct and
Taxonomy of Organizing Relations distinguishable entity. All neural networks and clas-
I offer here a taxonomy of the kinds of relations that sifier systems involve feedback learning via some path
could organize the parts of some ensemble into a coher- or another. The kind of feedback imposed by natu-
ent whole. The first and most prosaic are simple environ- ral selection in a genetic algorithm (GA), is another
mental relations which are the basic relations of arrange- example.
ment or position such as shape, momentum, proximity There are a number of different identifiable types of
(Searle 1992). They are not causal relations but the ac- feed-back relations based either on the function of a
cidents of an object. They contribute to the emergence comparison generating an error value (Wiener 1948) or
of an organized object through the opportunities they on the re-entry of processed input (Edelman, 1989).
afford for other more interactive relations to operate. A: Feedback with error values:
Beyond environmental relations I want to draw out
three orders of relations classified on the basis of their 1. Feedback in which a sample of the output is fed
interactivity and thus on their organizational capacity. back into the input as a direct modulation of the
input value. Negative feedback produces an inhi-
First order: Feedforward relations: Where environ- bition of the system and helps a system survive
mental relations have any emergent effect it will be perturbation. Positive feedback may cause a non-
through their enabling of local physical and chemical linearity, producing hysteresis in a system, and
interactions. Feed-forward offers no situation in which may also induce resonance. There are well under-
the object takes into account the degree of impact of stood conditions under which this resonance can
its relations with some other object affording it some occur (e.g. feedback oscillators and resonant filters
kind of "information" by which it can regulate its ac- in electronics) yet it is a perfectly good example of
tion. a weak emergence.
In ALife, rule-following is a feedforward process pro- 2. Feedback in which several processing stages oper-
ducing, e.g., gliders in cellular automata. It includes ate in a system so that its output is filtered or
4 in Artificial Life VIII, Standish, Abbass, Bedau (eds) (MIT Press) 2002. pp 418422
otherwise modulated before it is returned to the single cells. Human interactions: learning, teaching
input stage. These kind of feed-backs can provide and influencing each other, competing or cooperating
emphasis of some particular aspect of the input are the processes that bind individuals into a society.
and are probably involved in conscious attention Communication can lead to a likely increase in the fit-
in humans via the thalamo-cortical loop structure ness of both the signalling and the responding groups.
of the brain (Newman 1997). By emission of a signal individuals can assist each
3. Feedback with comparison to a norm or intended other to survive. (Koza 1991) describes an ant for-
outcome, generating an error value which is re- aging GA in which ants search for food by moving,
turned to the input stage. This is the kind of con- sensing and depositing pheromones. In Co-evolution
trol utilized in successive approximation processes each of a pair of algorithms acts as the environment
like reaching for an object. Each stage of the artic- for the other. One program tries to adapt to the "en-
ulation of the arm provides a fed back error value vironment" created by the other, by testing the per-
narrowing the discrepancy between current and in- formance of the one program relative to the other and
tended position. then vice versa. This is essentially a "biological arms
B: Feedback without error values: race" where each species develops defences and attacks
1. Unguided learning: This is feedback where the out- against the other.
put behaviour of some system becomes part of its Organisms gain their integrity, boundaries and self-
input stream, but without comparison. It occurs maintenance through open-system behaviors such as
when a known goal state is not available and the ingest-ion and sensing, processes that are active in
system has to make its own way, categorizing as all biological systems. When a complex system has
it proceeds. In a constructivist interpretation of adequate organization there will be a massive array
the world, our categorization of input stimuli can- of internal and external feed-back relations. A state
not have developed in comparison with pre-existing of organization is maintained by the exchange of en-
norms but must have been made according to a ergy with the environment, which helps to maintain
series of recurrences of events reinforcing certain the chemical metabolism of the membrane bounded
particular ways of viewing the situation. system, but it also produces potentially deleterious
2. Memory is possibly the most important conse- metabolic waste products. Excretion of these wastes
quence of feedback networks, particularly in bio- through the membrane completes a circular relation-
logical and neural systems. Something very similar ship with the environment. These wastes may be
to short-term memory occurs with the propagation the basis for the development of mutualistic relations
delay and resonance effects that a neural system (Jones 2000; Pachepsky, Taylor, & Jones 2002). In
will thus contain. It is very likely that re-entrant humans, perceptions and productions are the primary
feedback produces that slightly smeared experience elements of exchange with the environment that keep
of the present that we have as part of our con- us integrated as conscious beings capable of interac-
sciousness. tion with other entities and processes in the world of
3. Self-reflection: Our ability to reflect on what is oc- matter and ideas.
curring in comparison to things which have been
experienced previously, categorized and learned as Conclusion
history is a function of re-entrant processing cou- In a sense Cariani (1991) quite correctly dismisses emer-
pled with memory and the error-value generating gence, but in so doing he throws the baby out with the
processes that allow us to evaluate the effectiveness bath-water. The value of emergence is that it alerts us
of some act. to situations in which explanations must include causal
Feedback systems and the kinds of relations that are processes that are not usually recognized from within the
sensing and probing in environments where there are purview of micro-physics. Reductive explanation is, in
other entities of adequate complexity, may lead to mu- principle, possible for those objects that we commonly
tualistic (or third order) relations and bring about think of as emergent. Nevertheless such reduction ob-
communication. Any of these processes will indicate scures the forest not simply by a description of the trees
some sort of primitive intentionality in the system. but by the demand for a description of the quantum pro-
cesses that make the atoms and then the molecules and
Third order: mutualistic feedback relations. Different then the amino-acids and the proteins and then the cells
entities in an environment emitting signals and re- and so on up to the ecology of the forest. Strict reduc-
sponding to each other in natural or ALife systems tionist explanation is thus rather wasted, tedious if not
can lead to cooperative behaviour and co-evolution intractably difficult and not very useful. To completely
(Jones 2000). Interdependence of metabolic regula- explain some complex thing we must carry out a series
tion produces multi-celled organisms from a soup of of explanations, each of which amounts to a reduction
in Artificial Life VIII, Standish, Abbass, Bedau (eds)(MIT Press) 2002. pp 418422 5
of one level into the level of its constituents and their to the study of communication. In Langton et al.
organizing relations. In turn the constituents have to (1991), 631658.
be explained from within their own level, working down Needham, J. 1936. Order and Life. New Haven: Yale
to the ultimate micro-physical components in a series of University Press.
steps. Newman, J. 1997. Putting the puzzle together: Towards
Emergence, as explanation, is a shorthand. Useful ex- a general theory of the neural correlates of conscious-
planations of a system need to actively account for the ness. Journal of Consciousness Studies 4(1 & 2).
role of the organizing relations among the parts at each Pachepsky, L.; Taylor, T.; and Jones, S. 2002. Mu-
sub-level. This has the great value of opening up mecha- tualism promotes diversity and stability in a simple
nistic explanation, rendering it relevant in understanding artificial ecosystem. Artificial Life .8(1).
the dynamics of process. As Holland comments: "When Reynolds, C. 1994. An evolved, vision-based model of
we can formulate macrolaws that describe the behaviors obstacle avoidance behaviour. In Langton, C. G., ed.,
of emergent phenomena (for instance, the laws of chemi- Artificial Life III. Reading, Mass.: Addison-Wesley.
cal bonding) we gain greatly in comprehension, whether 327-346.
of a model universe or a real one". (Holland 1998, p.189). Searle, J. 1992. The Rediscovery of the Mind. Cam-
The dynamics of organizing relations enable emer- bridge, Mass.: MIT Press.
gence, affording the emergent object the means by which von Foerster, H. 1960. Environments of self-organizing
it can self-organize, self-regulate and, within the biolog- systems. In Yovits, M., and Cameron, S., eds., Self-
ical levels, be alive. In the simulations of computational Organizing Systems. New York: Pergamon Press.
emergence we need to find means by which emergent Whitehead, A. 1929. Process and Reality: an Essay in
objects at one level may become the constituents of a Cosmology. New York: Macmillan.
further emergence, by generating their own organizing Wiener, N. 1948. Cybernetics, or Control and Commu-
relations, thus climbing multiple levels in the hierarchy nication in the Animal and the Machine. New York:
of organization. Wiley.
References
Ashby, W. 1952. Design for a Brain. New York: Wiley.
Bickhard, M. 2000. Emergence. In Andersen, P. B.;
Emmeche, C.; Finnemann, N. O.; and Christiansen,
P. V., eds., Downward Causation. Aarhus, Denmark:
University of Aarhus Press. 322348.
Broad, C. 1925. The Mind and its Place in Nature.
London: Routledge & Kegan Paul.
Cariani, P. 1991. Emergence and artificial life. In Lang-
ton et al. (1991), 775797.
Collier, J. 1998. The dynamical basis of emergence in
natural hierarchies. In Farre, G., and Oksala, T., eds.,
Emergence, Complexity, Hierarchy and Organisation.
Selected and Edited Papers from the ECHO III Con-
ference, Acta Polytechnica Scandinavica, MA91. Es-
poo: Finnish Academy of Technology.
Holland, J. 1995. Hidden Order: How Adaptation Builds
Complexity. Helix Books.
Holland, J. 1998. Emergence: From Chaos to Order.
Redwood City, California: Addison-Wesley.
Jones, S. 2000. Evolvability in the context of the bio-
sphere. In Maley, C., and Boudreau, E., eds., ALife
VII Workshop Proceedings.
Koza, J. 1991. Genetic evolution and co-evolution of
computer programs. In Langton et al. (1991), 603
629.
Langton, C. G.; Taylor, C.; Farmer, J. D.; and Ras-
mussen, S., eds. 1991. Artificial Life II. Reading,
Mass.: Addison-Wesley.
MacLennan, B. 1991. Synthetic ecology: An approach