Background   |   Research   |   Publications   |   Software   |   Research-Related   |   Teaching   |   Contact  

The Evolutionary Emergence of Increasingly Intelligent Behaviours via Computational Natural Selection

A new approach to creating intelligence, rooted in Artificial Life and Natural Selection rather than traditional AI.
Alastair Channon

Research Context (Background)

Simulated evolution is well established as a computational optimization methodology. However, this methodology has so far concentrated on evolution towards pre-specified goals: a route which cannot generate anything like the unbounded evolution or the diversity and complexity of structures that we observe in nature. In “On the Origin of Species” Darwin emphasized the difference between the struggle between organisms for limited resources (biotic competition) and the struggle against features such as drought, of the non-living physical environment (abiotic competition). Biotic competition, he argued, has been the cause of sustained evolutionary progress.

Neo-Darwinism, which adds Mendelian heredity and post-Mendelian genetic theory, has clarified the nature and origins of species to the extent that we can carry out evolutionary experiments within artificial systems such as computer simulations and robotics. However, within the fields of artificial evolution (including evolutionary/genetic algorithms/programming/computing and sub-fields of artificial life, adaptive behaviour and digital biota), work to date uses only abiotic competition, with very few exceptions. Most of Darwin's theory would seem to have been ignored. Where biotic competition has been used, serious problems of evolvability can be identified.

The main focus is now on the generation of a system that exhibits unbounded evolution. Bedau and Packard's evolutionary activity statistics provide a basis for testing for this and have already been applied to a number of both artificial and natural selection systems, including my own. From the successes and failures of these tests we are beginning to extend the list of known requirements for unbounded evolution. Beyond this, biologists will want to use such systems to draw generic conclusions about evolution and engineers will want to evolve solutions that we can find uses for.

         

My Research

The emerging field of computational natural selection is an exciting area of artificial evolution that deals with the generation and analysis of open-ended evolutionary systems, primarily with the aims of overcoming the severe scaling problems exhibited by today's evolutionary algorithms, and evolving computational intelligence [10-13]. In the computational natural selection paradigm, the phenotype to fitness mapping is an emergent property of the evolving environment and competition is biotic rather than abiotic.

Geb is the only closed artificial system to have passed the statistical “Artificial Life Test” for unbounded evolutionary dynamics [7] and my refined (harder to pass) form of that test [1]. Earth's biosphere (through its fossil records) is the only other system to have passed this test, in either form, although a number of Artificial Life systems have been evaluated. This is a very significant result: potentially a second example of unbounded evolution, from which we can begin to draw generalized conclusions about open-ended evolution that were not possible given just the real-world example. The creation of a system capable of passing the test had been identified by Bedau, Snyder and Packard as “among the very highest priorities of the field of artificial life”. The significant refinements made to the test (including my method of component activity normalization) have been found by Stout and Spector to be crucial in resisting attempts to achieve a classification of unbounded dynamics in “intuitively unlifelike” systems.

Extending this research is key to both our understanding of evolution and our ability to replicate its full power to generate emergent processes and structures, including complex and intelligent ones. The next logical step is to enable the observation of evolved behaviours as they emerge. This can best be achieved through the evolution of both agents’ morphologies and controllers within a simulated physical environment, which has the added advantage of providing an open range of low-level actions.

As a first step, preliminary work on such a system, but with simpler, fitness-function based evolution, has been carried out: we have presented a ‘modified replication’ [3-5] of Karl Sims’ work on the evolution of articulated artificial creatures in physically realistic 3D environments. Our system was the first to demonstrate comparable results to those of Sims, and did so using standard McCulloch & Pitts’ neurons rather than ad hoc elements, so as not to provide problem-specific a priori knowledge.

Publications

  1. E. Robinson, T. Ellis and A. D. Channon, ``Neuroevolution of Agents Capable of Reactive and Deliberative Behaviours in Novel and Dynamic Environments,'' accepted to appear in Advances in Artificial Life: Proceedings of the Ninth European Conference on Artificial Life (ECAL 2007), Springer Lecture Notes in Computer Science (LNCS) volume ????, in the Lecture Notes in Artificial Intelligence (LNAI) subseries.

    This paper presents evolved artificial neural controllers that solve tasks requiring deliberative behaviours: tasks that cannot be solved by reactive mechanisms alone and which would traditionally have their solutions formulated in terms of search-based planning. Two very different neural networks are used: one that controls high-level deliberative behaviours, such as the selection of sub-goals, and one that provides reactive and navigational capabilities. Animats controlled by a hybrid of these network architectures are evolved in novel and dynamic environments, on increasingly complex versions of an example problem. The results demonstrate, for the first time ever, incremental neuro-evolutionary learning on such tasks.

  2. A. D. Channon, ``Unbounded evolutionary dynamics in a system of agents that actively process and transform their environment,'' Genetic Programming and Evolvable Machines, vol. 7, no. 3, pp. 253–281, 2006.

    This paper presents significant improvements to the established statistical "ALife Test" for unbounded evolutionary dynamics. Its contribution is in making the revised test (which the system from publication 8 still passes) well-grounded even for long-term unbounded evolution in artificial systems, through the first ever method of computing individual genes' adaptive ('normalized') evolutionary activities. In their paper "Validation of evolutionary activity metrics for long-term evolutionary dynamics", Stout and Spector attempted to "break" the test by achieving unbounded dynamics in "intuitively unlifelike" systems. They concluded that Channon's method of activity normalization is of particular importance to the test's robustness against such attempts.

  3. T. Miconi and A. D. Channon, ``The N-Strikes-Out Algorithm: A Steady-State Algorithm for Coevolution,'' in Proceedings of the 2006 IEEE Congress on Evolutionary Computation (CEC 2006), IEEE World Congress on Computational Intelligence (WCCI 2006), Vancouver, (G.G. Yen et al., eds.), pp. 1639–1646, IEEE Press, 2006.

  4. T. Miconi and A. D. Channon, ``Analysing coevolution among artificial 3D creatures,'' in Proceedings of the 7th International Conference on Artificial Evolution (Evolution Artificielle 2005): Revised Selected Papers, Lille, (E.G. Talbi et al., eds.), pp. 167–178, Springer, 2006. A volume of the LNCS Series.

    This paper presents new accomplishments in the coevolution of neurally controlled agents, and introduces improved methods of coevolutionary analysis. The experiments reported, on the coevolution of physically simulated articulated creatures, are the first to demonstrate realistic co-adapted behaviours using general purpose neurons. The previous need for ad hoc (problem-specific) neurons was a barrier to the long-term evolution of new, emergent behaviours. Novel behaviours are identified using an improved coevolutionary analysis method that is both more informative and an order of magnitude cheaper than the original. Finally, individuals are cross-validated between evolutionary runs, in an improved procedure for evaluating global performance.

  5. T. Miconi and A. D. Channon, ``An Improved System for Artificial Creatures Evolution,'' in Proceedings of the Tenth International Conference on the Simulation and Synthesis of Living Systems (ALife X), Bloomington, Indiana (L.M. Rocha et al., eds.), pp. 255–261, MIT Press, 2006.

  6. T. Miconi and A. D. Channon, ``A virtual creatures model for studies in artificial evolution,'' in Proceedings of the 2005 IEEE Congress on Evolutionary Computation (CEC 2005), Edinburgh (D. Corne et al., eds.), volume 1, pp. 565–572, IEEE Press, 2005.

  7. A. D. Channon, ``Improving and still passing the ALife test: Component-normalised activity statistics classify evolution in Geb as unbounded,'' in Proceedings of Artificial Life VIII, Sydney (R. K. Standish, M. A. Bedau and H. A. Abbass, eds.), (Cambridge, MA), pp. 173–181, MIT Press, 2003.  Instructions for replicating the runs discussed in this paper.

  8. A. D. Channon, ``Passing the ALife test: Activity statistics classify evolution in Geb as unbounded,'' in Advances in Artificial Life: Proceedings of the Sixth European Conference on Artificial Life (ECAL2001), Prague (J. Kelemen and P. Sosik, eds.), (Heidelberg), pp. 417–426, Springer Verlag, 2001. A volume of the LNCS/LNAI Series.

    This paper presents the first ever closed artificial system to pass the established statistical "ALife Test" for unbounded evolutionary dynamics: an achievement identified by Bedau, Snyder and Packard as "among the very highest priorities of the field of artificial life". Earth's biosphere (through fossil-record databases) is the only other system to have passed, although many have been evaluated. This is a very significant result: we can now begin to draw generalized conclusions about open-ended evolution that were previously impossible given just the real-world example. It significantly advances our ability to generate emergent processes and structures, including complex and intelligent ones.

  9. A. D. Channon, ``Evolutionary Emergence: The Struggle for Existence in Artificial Biota,'' PhD thesis, Department of Electronics and Computer Science, University of Southampton, 2001.

  10. A. D. Channon, ``Three evolvability requirements for open ended evolution,'' in Artificial Life VII Workshop Proceedings (C. C. Maley and E. Boudreau, eds.), (Portland, OR), pp. 39–40, 2000.

  11. A. D. Channon and R. I. Damper, ``Towards the evolutionary emergence of increasingly complex advantageous behaviours,'' International Journal of Systems Science, special issue on Emergent Properties of Complex Systems, vol. 31, no. 7, pp. 843–860, 2000.

  12. A. D. Channon and R. I. Damper, ``The evolutionary emergence of socially intelligent agents,'' in Socially Situated Intelligence: a workshop held at SAB'98, University of Zurich Technical Report (B. Edmonds and K. Dautenhahn, eds.), (Zurich), pp. 41–49, 1998.

  13. A. D. Channon and R. I. Damper, ``Perpetuating evolutionary emergence,'' in From Animals to Animats 5: Proceedings of the Fifth International Conference on Simulation of Adaptive Behavior (SAB98), Zurich (R. Pfeifer, B. Blumberg, J. A. Meyer, and S. Wilson, eds.), (Cambridge, MA), pp. 534–539, MIT Press, 1998.

  14. A. D. Channon and R. I. Damper, ``Evolving novel behaviors via natural selection,'' in Proceedings of Artificial Life VI, Los Angeles (C. Adami, R. Belew, H. Kitano, and C. Taylor, eds.), (Cambridge, MA), pp. 384–388, MIT Press, 1998.

  15. A. D. Channon and R. I. Damper, ``The artificial evolution of real intelligence by natural selection.'' Published on the web site of and poster presented at the Fourth European Conference on Artificial Life (ECAL97), Brighton, 1997.

  16. A. D. Channon, ``The Evolutionary Emergence route to Artificial Intelligence.'' MSc thesis, School of Cognitive and Computing Sciences, University of Sussex, 1996.


Software

Geb Version 7c5 for Unix (Linux/SunOS/IRIX/HP-UX/...) & MS-Windows

Geb is an artificial world containing organisms which evolve by natural selection. The papers above provide the best description of Geb. I recommend reading at least Perpetuating evolutionary emergence before trying to make sense of the source code, and that Passing the ALife test is the next paper you read.

Research-Related Activities

I have delivered invited research talks, and reviewed papers for international journals (Artificial Life, IEEE Transactions Evolutionary Computation, International Journal of Systems Science, Computational and Mathematical Organization Theory, Journal of Memetics – Evolutionary Models of Information Transmission) and international conference series (Genetic and Evolutionary Computation, Congress on Evolutionary Computation, Artificial Evolution, Artificial Life, Parallel Problem Solving from Nature, Soft Computing and Intelligent Systems). I was on the organizing committee for the 2005 IEEE Congress on Evolutionary Computation (CEC 2005) and the 2006 International Conference on Parallel Problem Solving From Nature (PPSN 9).

I am a member of the IEEE, as well as the International Society for Genetic and Evolutionary Computation (ISGEC) and the International Society for Artificial Life (ISAL). I have reviewed research proposals for national funding agencies (EPSRC, Dutch National Science Foundation).

Teaching and Administration

I have designed, developed, delivered and revised high quality research-led undergraduate and postgraduate modules on evolutionary systems, intelligent systems, nature inspired design, artificial intelligence and less specialized units on AI programming, object oriented programming in C++, software engineering and group software engineering.

In 2005-2006 I lead the design, proposal and setup of the new MSc in Intelligent Systems Engineering in the School of Computer Science, University of Birmingham. This programme has proved successful in recruiting high calibre students, and looks set to grow in the forthcoming years.

Other duties have included the roles of programme director, academic advisor and admissions tutor for this new MSc and the MSc in Natural Computation; membership of MSc Board of Examiners; supervision of seventy-three individual undergraduate and MSc projects; PhD student supervision and thesis-group membership; PhD examination (internal examiner for vivas); membership of departmental computing committee, unit assessment boards and faculty curriculum approval committees.

Contact Details

Dr Alastair Channon
School of Computing and Mathematics
Keele University, Staffordshire, ST5 5BG, UK


(for emails <1MB - contact me for >1MB)

Total number of visitors since Tue29Oct1996:

WebCounter