The Species Problem in Artificial Life

1. The Species Problem in Artificial Life George Kampis(1, 3), László Gulyás (1, 2) and Sándor Soós (1, 3) 1Department of History and Philosophy of Science, Eötvös University, P.o.Box 32, H-1512 Budapest 2AITIA International, Inc., Czetz J. u. 48-50, H-1039 Budapest 3 Collegium Budapest, Szentháromság u. 2, H-1014 Budapest George Kampis, Fellow, Collegium Budapest Wayne G Basler Chair, ETSU http://www.colbud.hu/fellows/current.shtmlhttp://www.etsu.edu/biology/

2. www.evotech.hu Background: Evolutionary Technology Rama, an O’Neill cylinder A.C. Clarke: Rendezvous with Rama Kampis, G. 2002: Towards an Evolutionary Technology http://hps.elte.hu/evotech/doc/Towards_an_Evolutionary_Technology.pdf

3. www.evotech.hu • Complex biological (supraindividual) systems, evolution and agent-based simulation • Evolutionary Technology (EvoTech) is a new initiative of a Hungarian research group specialized in state-of-the-art computer modelling of complex and emergent biological phenomena at the supraindividual level. The main aim is to address the most problematic issues with respect to evolutionary modelling as that of growing complexity, sustained evolution, speciation and the co-existence of species, the species problem or the formation and evolution of proper food-webs within an artificial life environment. • The methodological framework is characterized by agent-based (individual-based) simulation [http://en.wikipedia.org/wiki/Agent_based_model]. The work is to implement a sound and coherent theoretical approach to evolutionary processes from biology and biophilosophy. • The work is the outcome of research conducted by participants from several organizationswithin various scientific projects at the national and the european level. Core staff is associated with the Focus Group Philosophy and Praxis of Complex Systems at Collegium Budapest, the international Institute of Advanced Study in Budapest, Hungary. Most recent work is part of the European Project QosCosGrid supported by the 6th RTD Framework Programme of the European Union.

4. Evolutionary Technology: The Idea • Breed machines with an increasing functionality and complexity… • …in an open ended evolutionary process, … • …which is nevertheless directed. • The proposed idea: breed conintually renewing selection forces by making use of phenotype plasticity • The notion of „fat phenotype” interactions • Full Body: The Importance of the Phenotype in Evolution Proc Worshop "Evolution of Complexity" at ALIFE X, Bloomington, http://ecco.vub.ac.be/ECO/Kampis.pdf • Kampis, G. and Gulyas, L. (2004): Sustained Evolution from Changing Interaction, Alife IX Boston,     http://hps.elte.hu/~gk/EvoTech/Sustained.pdf

5. The FATINT System (2004-) • Download: • http://hps.elte.hu/~kampis/EvoTech/EvoTech_V-v5.0.9.zip

6. For your eyes only…

7. Steps of a Methodology • ABM professional (RePast) • Fully embedded (encapsulated, no global operations, no „invisible hand”) • Informed from theoretical biology (non-arbitrary vs. ad hoc models) • Kampis, G and Gulyas, L 2006: Full Body: The Importance of the Phenotype in Evolution Proc Worshop "Evolution of Complexity" at ALIFE X, Bloomington, http://ecco.vub.ac.be/ECO/Kampis.pdf • Birth and death processes • (upcoming) • Similarity based operations • Kampis, G. & Gulyás, L. 2006: Phat Phenotypes for Agents in Niche Construction, in: Artificial Life X: Proceedings of the Tenth International Conference on theSimulation and Synthesis of Living Systems (Bradford Books), MIT Press, Boston. http://hps.elte.hu/evotech/doc/PhatPhenotypes.pdf • Species • Current paper http://hps.elte.hu/evotech/doc/KGS_paper_v2.0_gk.pdf

8. „The Uncanny Valley” of ABM

9. The Uncanny Valley http://en.wikipedia.org/wiki/Uncanny_Valley http://en.wikipedia.org/wiki/Masahiro_Mori

10. „The Uncanny Valley” of ABM

11. Fundamental Questions of Model Building • (1) Differences caused by realizations of operators of population models • (2) Differences caused by various representations of a population • Ad (1) E.g. Urn models of birth-and-death processes can demonstrate the sensitivity of realization e.g. balanced Pólya’s urn vs. „kill on demand”

12. „The Canny Valley”: Species • Under conditions of relaxation several different species definitions tend to yield identical or similar classifications in an ABM population model • Test bed: the FATINT system • Species Problem in philosophy (etc.) • http://plato.stanford.edu/entries/species/

13. Species. Working definition • A species is a set of interbreeding • individuals. • Reproductively isolated from others • (e.g. mechanically, behaviorally, in • terms of mating preference, etc.)

14. A more realistic species. Not panmictic, yet stable . In an evolutionary engine, a species, which responds to a selection force, usually shows a high degree of stability.

15. FATINT (v5.0.12) Species Maintained by Sexual Selection Finding the „right” mating partner... is a matter of fitting together. • … be modeled as template matching, • which defines a metric (similarity) space • that can be used in any dimensions if individuals are represented as n-vectors.

16. The Species Problem, What is It? • Various species concepts exist • Philosophical species problem (some 22 definitions) • Which of these is (more) „real”? • What difference/error is caused by adopting one notion rather than another? • How does this relate to the dynamics of the system? • Difficulties of „seeing” in n-space • The problem of ordination (known from ecology, statistics…) • Low dimensional visualization and related measures

17. Various Species Concepts • … • Species are individuals (Hull) • Species are sets • Family resemblance notions (Pigliucci) • Quasispecies (Eigen) • …

18. 3 Selected Concepts • Biological Species Concept (BSC) • On BSC, a species corresponds to a maximal group of potentially interbreeding organisms (a „Mendelian”population) that are reproductively isolated from others • Cladistic Species Concept (CSC) • On the Cladistic Species Concept, a species is a (minimal) lineage of populations delineated by two branching events („points”) on the phylogenetic tree • Phenetic Species Concept (PSC) • On the phenetic concept, a species is a cluster of similar organisms delimited with the aid of some statistical clustering method.

19. Definitions and Realizations • BSC.We define a species on BSC at time T as a connected network of agents with d<Mlimit. • CSC.A species is the historical envelope of past reproduction events in a given lineage between T0 and T. Two individuals in T belong to different species if the two envelopes do not intersect. • PSC.We define a PSC species as a cluster in phenotype space, characterized with a clustering constant dcluster. • CSC is realized using network components analysis with component size ≥ 2 applied to lineages traced back to individuals born no earlier than T0 in a history dump up to T. • PSC partitions are clusters formed using agglomerative hierarchical clustering with complete linkage

20. FATINT behavior. Relaxation Reference: Kampis, G. and Gulyas, L. (2004): Sustained Evolution from Changing Interaction, Alife IX Boston,     http://hps.elte.hu/~gk/EvoTech/Sustained.pdf Average convergence; number of components vs time; sensitivity tests Different kinds of histograms (genes vs alleles); tyical FR plot for large t

21. Kampis, G. and Gulyas, L. (2004): Sustained Evolution from Changing Interaction, Alife IX Boston,     http://hps.elte.hu/~gk/EvoTech/Sustained.pdf FATINT behavior. Speciation t = 8 t = 10 t = 34 t = 58 t = 84 t = 100

22. Treatment T1(none) BSC. One giant component after t= 150, size of giant component is about population size CSC. Circular FR plot of partitions of CSC under T1 in the interval T0=0 to T=644. The picture shows one giant, connected lineage, and several small, isolated components (on the left). PSC. Clusters against time. Convergence to (typically) one PSC cluster under T1 fordcluster≥ 1. The Figure summarizes 60 runs, 10 for 6 random seeds each, from dcluster = 0.9 to 1.4 . Time goes left to right, the clustering constant increases right to left. Vertical axis shows number of clusters.

23. The R index • Pairwise comparisons of clusters were performed by calculating the Rand-index (R), a standard method for comparing classifications. The Rand index has a value between 0 and 1. The value 0 indicates that the two data clusters do not agree on any pair of points and 1 indicates that the data clusters are exactly identical. Calculations were performed in the R statistical program package which was successfully applied to philogenetic comparisons before. • http://www.r-project.org/

24. Calculating R • Given a set of n objects S = {O1, ..., On} and two data clusters of S, to be compared, X = {x1, ..., xR} and Y = {y1, ..., yS}, where the different partitions of X and Y are disjoint and their union is equal to S; we compute the following values: • a is the number of elements in S that are in the same partition in X and in the same partition in Y, • b is the number of elements in S that are not in the same partition in X and not in the same partition in Y, • c is the number of elements in S that are in the same partition in X and not in the same partition in Y, • d is the number of elements in S that are not in the same partition in X but are in the same partition in Y. • Intuitively, one can think of a+b as the number of agreements between X and Y and c+d the number of disagreements between X and Y. • The Rand index, R, is defined as:

25. Treatment T2. Speciation experiments (Pmutation = 0.001) • A new global phenotype dimension was hand-added at T=475. Sample runs T0=0 to T=1,000, 7 random seeds. In 5 of the 7 runs 2 BSC species emerged (Fig. 5.). In three cases of the 5, the second species went extinct between T=650 and T=980. Sample FR plot on BSC at T=550 in treatment T2. FR plot of CSC partitions (nodes with different colors) during the speciation process T0=450 to T=550 in T2. The picture shows one single lineage with a divided connectivity structure that reflects the temporal branching of the lineage.

26. Treatment T2. Speciation experiments (Pmutation = 0.001) • A new global phenotype dimension was hand-added at T=475. Sample runs T0=0 to T=1,000, 7 random seeds. In 5 of the 7 runs 2 BSC species emerged (Fig. 5.). In three cases of the 5, the second species went extinct between T=650 and T=980. SPECIES COMPARISONS IN T2. THE STRUCTURE OF THE SPECIATION EVENT.

27. Treatment T3Autonomous speciation experiments • Sample runs with pnewslot=0.001 in T0=0 to T=1071. The nonzero value of this parameter introduces random phenotype change events, which then induce potential speciation events. The high value applied permits several speciation events and speciation bursts within a single run. Sample run in T3, the number of BSC species shown at T0=0 to T=1071. Random speciation burst at T=500, later on extinctions. THE ABOVE SAMPLE RUN AT SPECIATION BURST. COMPARISONS IN T3 USING DIFFERENT RANDOM SEEDS

28. Discussion • expressed briefly, under relaxation conditions species exist in a very strong sense: up to the point of the extensional coincidence of intenasionally differenty definitions • BSC to PSC trivial (cf. similarity is the basis of both? Not quite; • In BSC to PSC comparisons, the R index depends on the ‘compactness’ of the species: the smaller the diameter in n-space, the more likely BSC and PSC to coincide. • CSC and BSC automatically coincide at T and short intervals around • good coincidence of CSC with BSC at longer intervals indicates the stability of a given composition of species with respect to their ‘defining types’ • CSC classification highly depends on the choice of T, as CSC counts ‘backwards’ in time. Cf, for instance, where only one of two earlier lineages survives at T.

29. The „Darwin Condition” • Darwinfamously foresaw the reconciliation of the individual variation based notion of species (which was introduced by him) with a taxonomic (preexisting) notion and an evolutionary species concept (also introduced by him) • Under relaxation (long intervals w/o speciation) we find a similar situation here. Hence we can call it the „Darwin Condition” • Features of species under the Darwin Condition • Compactness: all individuals clearly belong to species, which are widely separated and lack transitory characters that would temporarily link them. In particular, lineages that originate at characters inside the same phenetic cluster typically end up in the same reproductive cluster (cross-measure closure property). • Being ‘Typed’: members of a lineage (or a reproductive cluster) share certain characters, and any organism with the same characters predictably belongs to the same lineage. In particular, randomly chosen members of a species can represent the species both at a given time and in a historical time frame (prototype property). • Taxonomic realism

30. Summary: many species concepts or one? • It all depends on the dynamics… • Under a broad range of conditions, we find ourselves in a „benevolent universe” • Choice of different ordination measure does not invalidate results • Peephole into how „species construct themselves” in an emergent process • Modeling: there is a „free dinner” • Philosophy: „much ado about nothing” (?)

31. The EvoTech/FATINT team

32. W. de Back PhD, ColBud Nigel Gilbert CS Advisor Péter Érdi CS Advisor Mark Bedau CS Advisor Imre Kondor CS Advisor Sándor Soós Coordinator PhD, ColBud István Karsai Associate Director, ETSU IQB Katalin Mund Associate PhD Student ELTE László Gulyás Researcher (p.t.) PhD, ColBuD György Kampis Group leader Professor ColBud students

33. W. de Back PhD, ColBud Nigel Gilbert CS Advisor Péter Érdi CS Advisor Mark Bedau CS Advisor Imre Kondor CS Advisor Sándor Soós Coordinator PhD, ColBud István Karsai Associate Director, ETSU IQB Katalin Mund Associate PhD Student ELTE László Gulyás Researcher (p.t.) PhD, ColBuD György Kampis Group leader Professor ColBud students

34. W. de Back PhD, ColBud Nigel Gilbert CS Advisor Péter Érdi CS Advisor Mark Bedau CS Advisor Imre Kondor CS Advisor Sándor Soós Coordinator PhD, ColBud István Karsai Associate Director, ETSU IQB Katalin Mund Associate PhD Student ELTE László Gulyás Researcher (p.t.) PhD, ColBuD György Kampis Group leader Professor ColBud students

35. Special thanks to..

36. Thank you!