The Emergence of Emergence:
A Critique of "Design, Observation, Surprise!"

Richard Gordon
Departments of Radiology and Electrical & Computer Engineering
University of Manitoba, Health Sciences Centre
820 Sherbrook Street, Winnipeg, MB R3A 1R9 Canada
Phone: (204) 789-3828, Fax: (204) 787-2080
E-mail: GordonR@ms.UManitoba.ca

Manuscript started February 16, 2000. Version 09, June 25, 2000.

Abstract: Artificial life research begins from the premise that Alife subsumes real life. A criterion for emergence in Alife has been formulated that, however, excludes real life and postulates the need for a real life Designer and an Observer. This in effect nullifies the premise of Alife and takes us back to the argument for God from design of Bishop Paley in 1802. An alternative is to realize that Alife could include two properties: simulated organisms that both design themselves and are the observers. Self-design can come about via evolution in a population of mating organisms, especially via mutations that are gene or higher order duplications. Duplications permit novelty while retaining previously attained functions. The ability to observe can itself evolve, if its construction process evolves. This may now be possible to simulate, if new paradigms for embryogenesis, such as positional information or differentiation waves, prove accurate, or at least sufficiently robust to construct a wide diversity of observational abilities. The evolution of perception, however, may be limited by the physics available to the Alife organisms, which can come in three forms: simulated physics, real physics accessible to robots, or "Cyberspace physics".

Keywords: artificial life, Alife, emergence, evolution of perception, positional information, differentiation waves, embryonics

Ronald, Sipper & Capcarrère [1999] have proposed a three step rationale for defining a system as emergent:

1. A system is constructed by a designer using "local elementary interactions between components" described by "language L1".
2. The system is observed over finite intervals of space and time ("globally") by someone using a different "language L2".
3. If the observer cannot easily deduce the L2 description of the system from its L1 description, "Surprise!" occurs and the system is emergent.

Without explanation, the authors restrict these criteria to artificial life. But for those of us for whom artificial life is hopefully a step towards understanding real life (Levy [1992]; Emmeche [1994]; Fontana, Wagner & Buss [1995]; Ray [1998]; Gordon [1999, p.210]), these criteria create a problem, in that Ronald, Sipper & Capcarrère [1999] have taken us right back to Bishop William Paley [1802]: design implies a Designer:

"Although I did not think much about the existence of a personal God until a considerably later period of my life, I will give here the vague conclusions to which I have been driven. The old argument from design in Nature, as given by Paley, which formerly seemed to be so conclusive [cf. Clark [1984]], fails, now that the law of natural selection has been discovered. We can no longer argue that, for instance, the beautiful hinge of a bivalve shell must have been made by an intelligent being, like the hinge of a door by man. There seems to be no more design in the variability of organic beings, and in the action of natural selection, than in the course which the wind blows" (Darwin, 1876, in Darwin [1905]).

Ronald, Sipper & Capcarrère [1999] go one step further than Paley [1802], adding that emergence requires an Observer, with "the necessity of there being an observer for emergence to arise at all". A relation is implied that humans are to artificial life as God is to real life. This precludes a unified treatment of life and artificial life.

I would like to suggest an alternative starting point: the designer and the observer are part of the system that we need to simulate. To begin here we must face the question of the emergence of emergence (which, in a way, is the biggest Surprise!). We must simultaneously simulate the emergence of both design and observation.

To accomplish this, I think we will have to add at least three elements lacking from "Design, Observation, Surprise!": evolution, embryology, and physics. A system that develops in the embryological sense constructs itself. Self-construction of an embryo is an interaction between genetics and physics (His [1888]; Gordon [1999]). The embryo becomes an observer in the process. What evolves are the rules of construction, as the individuals change from one generation to another by (occasional) mutations (cf. one component, evolution of the eye: Nilsson & Pelger [1994]; Osorio [1994]).

The ability to observe involves, at the most primitive level, distinctions of self from others or the environment, recognition of food versus nonfood, avoidance of predators or dangerous environments, and recognition of mates of one's own species. Most of these characteristics are already present in single celled bacteria (Dunny & Winans [1999]), and certainly in single celled eucaryotes (Laybourn-Parry [1984]) (cf. Martin & Gordon [2000]). At this level, at least, behavior should be understandable in terms of the physics of molecules, membrane potentials, photobiology, etc. I assume that the ability to observe, behave, learn and interact with other individuals follows from their construction rules (cf. Dellaert & Beer [1996]).

Real organisms alter their genetics not only by point mutation and recombination, but also mutate by producing duplications of whole genes, gene batteries and whole genomes (Ohno [1970, 1999]; Wagner [1998]; Gordon [1999]). Old functions are thereby preserved while entirely new ones become possible via the new construction potentiated by the duplicated portions of the genome. The dimensionality of the genome actually increases (Gordon [1994]).

In the first individual in which a portion of a genome is duplicated, no new function is likely to ensue. However, in subsequent generations, that duplication can undergo point and other mutations, in many directions, via numerous descendent individuals. Some of these mutations may enhance survival, sometimes by introducing entirely new capabilities. In the meantime the original copy can preserve the capabilities inherent in the creatures' ancestors.

This process can be seen in any of the numerous gene families documented in modern molecular biology (cf. Sluder et al. [1999]). Each gene in a family can be traced to a "common ancestor gene", of which it is a mutated copy (Fryxell [1996]), but tends to have a new function. The same is possible with the gene batteries involved in steps of differentiation in an embryo, giving rise to new tissues (Gordon [1999]). (Duplication and other features of real genomes have only begun to be incorporated into genetic algorithms: Burke et al. [1998]; Soule & Foster [1998]).

Physics is an important component of emergence. Consider vision, for example. We cannot expect artificial life organisms to come up with photoreceptors (let alone eyes) if the simulated environment in which they exist does not contain simulated photons. If we were to simulate evolution of the various feeding behaviors of protozoans (Laybourn-Parry [1984]), nothing much could happen in the simulated evolution of their feeding apparatuses without simulations of the hydrodynamics of water. Without the physics of the cytoskeleton, differentiation waves (real or simulated: Gordon [1999]; Brodland & Clausi [1994]) will not propagate, and thus cannot emerge. Thus, gene duplication is just the beginning of emergence. To get new capabilities, physics must also be represented, so that selection has something to "work against". This phenomenon of matching organisms' capabilities to their environment, through evolution, has facetiously been called The Fitness of the Environment (Henderson [1913]). It is the basis of the idea that "life processes are not necessarily tied to the medium of carbon chemistry" (Ray [1998]), i.e., that artificial life could work on alternates to real world physics. But physics must be there, and may be what places an ultimate limit on how much emergence can occur. If an alternative, cyberspace physics proves less rich than real world physics, the artificial life that evolves there might be quite limited compared to real life. The approach of embryonics, embedding the robot and its (in part) self-construction in the real world (Mange & Stauffer [1994]; Mange et al. [1996]; Marchal et al. [1996]; Tempesti, Mange & Stauffer [1998, 1999]) may succeed precisely because the robot has direct access to real world physics:

"The definition of evolvable hardware hinges on whether or not electronic circuits play a fundamental role in the evolutionary process; the hardware is in the loop, so to speak, as opposed to the entire evolutionary process being run as a software simulation....

"For an application to qualify as evolvable hardware, the presence of real electronic circuits, rather than software simulation, is a must. This is not just a chauvinistic attitude toward software on the part of hardware designers: using real hardware fundamentally changes the evolutionary process--and its results.

"For one thing, there is no need to transfer the result of a simulation to hardware, whereas this step is a problem for several software-based efforts. Designing a car, a robot, or an electronic circuit by simulation frequently presents unpleasant surprises when the device is built. Sometimes, these surprises are trifling; at other times, they can be quite nasty, preventing the designers from reaching their intended goal" (Sipper & Ronald [2000]).

In this case, the hardware is merely electronic, but the authors envisage analog hardware too, so we are beginning to take steps towards the kind of physics that is accessible to real life also being accessible to artificial life. Hybrid simulations (i.e., evolving robots) have the advantages that: 1) all real world physics is available in some way to the simulation; 2) the difficult and time consuming simulation of the physics need not be carried out in software. However, we still have to create an interface between the artificial life simulation and the real world physics that makes the physics accessible to the artificial life. Either by simulation or robotics, the physics of the artificial life organism itself must therefore be included.

Let us then reformulate the rationale of Ronald, Sipper & Capcarrère [1999] in a manner applicable to both real and artificial life:

1. At a given time, a species is defined by construction rules for its individuals, by which the genotype is translated to a phenotype via genetics and physics.
2. The ability to observe is part of an individual's construction.
3. Surprise! occurs when gene duplication is followed by (viable, heritable, survivable) mutations in subsequent generations that construct new capabilities (morphological, behavioral, or observational).

The advantage of this view of emergence is that the individual is both the designer and the observer, "a self-describing, self-constructing system" (Pattee in Rosen, Pattee & Somorgai [1979]; cf. Gordon [1999], p. 210). Design is also a population phenomenon, because it is in mating populations of a species that numerous mutations and culling occur. Simulations of this kind require a simulatable theory of embryo construction, connecting the genotype to the phenotype. Lacking this, to date most artificial life springs forth in adult form. This restriction can now be lifted. With theories such as positional information (Wolpert [1969]; Mange, Sipper & Marchal [1999]) or differentiation waves (Gordon [1999]), we should now be able to proceed with simulation of the emergence of emergence.

Acknowledgments: I would like to thank Steven McGrew and Blake Podaima for their critical comments.

References

Brodland, G.W. & D.A. Clausi [1994], Embryonic tissue morphogenesis modeled by FEM. J. Biomech. Eng. 116: 146-155.
Burke, D.S., K.A. De Jong, J.J. Grefenstette, C.L. Ramsey & A.S. Wu [1998], Putting more genetics into genetic algorithms. Evol Comput 6: 387-410.
Clark, R.W. [1984], The Survival of Charles Darwin, A Biography of a Man and an Idea, Random House, New York.
Darwin, F., ed. [1905], The Life and Letters of Charles Darwin, Including an Autobiographical Chapter, F. Appleton and Co., New York.
Dellaert, F. & R.D. Beer [1996], A developmental model for the evolution of complete autonomous agents. In: Maes, P., M.J. Mataric, J.-A. Meyer, J. Pollack, & S.W. Wilson, From Animals to Animats 4: Proceedings of the Fourth International Conference on Simulation of Adaptive Behavior, MIT Press, Cambridge, p. 393-401.
Dunny, G.M. & S.C. Winans, ed. [1999], Cell-Cell Signaling in Bacteria, American Society for Microbiology Press, Washington, D.C.
Emmeche, C. [1994], The Garden in the Machine: The Emerging Science of Artificial Life, Princeton University Press, Princeton.
Fontana, W., G. Wagner & L.W. Buss [1995], Beyond digital naturalism. In: Langton, C.G., Artificial Life: An Overview, MIT Press, Cambridge, p. 211-227.
Fryxell, K.J. [1996], The coevolution of gene family trees. Trends Genet 12: 364-9.
Gordon, R. [1994], Evolution escapes rugged fitness landscapes by gene or genome doubling: the blessing of higher dimensionality. Computers & Chemistry 18: 325-332.
Gordon, R. [1999], The Hierarchical Genome and Differentiation Waves: Novel Unification of Development, Genetics and Evolution, World Scientific & Imperial College Press, Singapore & London.
Henderson, L.J. [1913], The Fitness of the Environment, An Inquiry into the Biological Significance of the Properties of Matter, 1958 reprint ed., Macmillan, New York.
His, W. [1888], On the principles of animal morphology. Roy. Soc. Edinburgh Proc. 15: 287-298.
Laybourn-Parry, J. [1984], A Functional Biology of Free-Living Protozoa, University of California Press, Berkeley.
Levy, S. [1992], Artificial Life, The Quest for a New Creation, Pantheon Books, New York.
Mange, D., M. Goeke, D. Madon, A. Stauffer, G. Tempesti & S. Durand [1996], Embryonics: a new family of coarse-grained field programmable gate arrays with self-repair and self-reproducing properties. Lecture Notes Computer Sci. 1062: 197-220.
Mange, D., M. Sipper & P. Marchal [1999], Embryonic electronics. Biosystems 51: 145-52.
Mange, S. & A. Stauffer [1994], Introduction to embryonics: towards new self-repairing and self-reproducing hardware based on biological-like properties. In: Thalmann, N.M. & D. Thalmann, Artificial Life and Virtual Reality, John Wiley, Chichester, p. 61-72.
Marchal, P., P. Nussbaum, C. Piguet, S. Durand, D. Mange, E. Sanchez, A. Stauffer & G. Tempesti [1996], Embryonics: the birth of synthetic life. Lecture Notes Computer Sci. 1062: 166-196.
Martin, C.C. & R. Gordon [2000], The evolution of perception. In: Trappl, R., Cybernetics and Systems 2000, Proceedings of the Fifteenth European Meeting on Cybernetics and Systems Research, organized by the Austrian Society for Cybernetic Studies, held at the University of Vienna, Austria, 25-28 April 2000, Austrian Society for Cybernetic Studies, Vienna, p. 215-219.
Nilsson, D.E. & S. Pelger [1994], A pessimistic estimate of the time required for an eye to evolve. Proc R Soc Lond B Biol Sci 256: 53-8.
Ohno, S. [1970], Evolution by Gene Duplication, Springer-Verlag, New York.
Ohno, S. [1999], Gene duplication and the uniqueness of vertebrate genomes circa 1970-1999. Semin Cell Dev Biol 10: 517-22.
Osorio, D. [1994], Eye evolution: Darwin's shudder stilled. Trends in Ecology and Evolution 9: 241-242.
Paley, W. [1802], Natural Theology: Or the Evidences of the Existence and Attributes of the Deity, Collected from the Appearances of Nature, R. Faulder, London.
Ray, T.S. [1998], Selecting naturally for differentiation: preliminary evolutionary results. Complexity 3: 25-33.
Ronald, E.M.A., M. Sipper & M.S. Capcarrère [1999], Design, Observation, Surprise! A Test of Emergence. Artif Life 5: 225-239.
Rosen, R., H.H. Pattee & R.L. Somorjai [1979], A symposium in theoretical biology. In: Buckley, P.J. & F.D. Peat, A Question of Physics: Conversations in Physics and Biology, University of Toronto Press, Toronto, p. 84-123.
Sipper, M. & E.M.A. Ronald [2000], A new species of hardware. IEEE Spectrum 37: 59-64.
Sluder, A.E., S.W. Mathews, D. Hough, V.P. Yin & C.V. Maina [1999], The nuclear receptor superfamily has undergone extensive proliferation and diversification in nematodes. Genome Res 9: 103-20.
Soule, T. & J.A. Foster [1998], Removal bias: a new cause of code growth in tree based evolutionary programming. In: Fogel, D., 1998 IEEE International Conference on Evolutionary Computation, May 4-9, 1998, Anchorage, Alaska, IEEE Computer Society Press, Los Alamitos, California, p. 781-786.
Tempesti, G., D. Mange & A. Stauffer [1998], Self-replicating and self-repairing multicellular automata. Artif Life 4: 259-82.
Tempesti, G., D. Mange & A. Stauffer [1999], The Embryonics Project: a machine made of artificial cells. Riv Biol 92: 143-88.
Wagner, A. [1998], The fate of duplicated genes: loss or new function? BioEssays 20: 785-8.
Wolpert, L. [1969], Positional information and the spatial pattern of cellular differentiation. J. Theor. Biol. 25: 1-47.

Dedicated to the memory of the late Susumu Ohno.

[Home Page -Papers]