Artificial (Virtual) Ecosystems

1. General discussion

References & Reading Material: Yaeger, L., "Computational Genetics, Physiology, Metabolism, Neural Systems, Learning, Vision and Behaviour or PolyWorld: Life in a New Context", Artificial Life III, (ed.) Langton, Addison-Wesley 1994, pp263-298 Ray, T., "An Approach to the Synthesis of Life", Artificial Life II, (eds) Langton et al, Addison-Wesley 1991, pp371-408 Pargellis, A.N., "The Spontaneous Generation of Digital Life", Physica D 91, 1996, pp86-96 Images: Attenborough, D., "The Private Life of Plants", BBC Books, 1995 Beck, W.S, Liem, K.F, Simpson, G.G, "Life", 3rd edn, Harper-Collins, 1991

What is Meant by an Artificial (Virtual) Ecosystem?

Any model in which the (virtual) inhabitants (agents):

• Interact with one another
  
  
• Mate with one another
  
  
• Compete against one another for material resources (food, water etc.)
  
  
• Compete against one another for mates

...within the confines of some (virtual) environment with its own set of 'universal laws'.

Reproduction within a virtual ecosystem may be implemented using the genetic algorithm with an implicit fitness function...

• An individual agent is responsible for its own reproduction
  
  
• Reproduction requires resources
  
  
• Resources are acquired from the environment by appropriate behaviour

Sample behaviours:

	Communication Warnings (bird cry, wasp colouration) Mating calls (changing colour) Food signals (bee's dance) Territory markers (pheromones)
	Predation Pack hunting (lions, wolves) Chasing (leopards) Trapping (spider's web, Venus Fly Trap) Ambushing (jumping spiders, ant lions) Aggregate and social behaviours Pack hunting (lions, wolves) Flocking (sheep, birds) Schooling (fish) Removing lice (monkeys) Nest construction (termites, ants, bees, wasps)...	Lovely Venus
		Ant Lion
	Mimicry Camouflage (stick & leaf insects) Batesian mimicry (tephritid fly, snake-mimicking caterpillar) Inter-species competition & co-operation	Not a wasp! (Syrphid Fly)
Orchid Imitates Bee	Floundering	What moth?

In addition to the characteristics of the organism, characteristics may be added to the environment:

• Spatial & physical laws
  
  
  • Collisions, rigid body dynamics, friction, gravity etc.
  • Organization of space (0D, 1D, 2D, 3D, more D, toroidal, infinite plane)
  
  
• Energy availability
  
  
  • From a global source (sunlight)
  • From a localized source (plants, animals)
  
  
• Number and type of organisms

2. Polyworld
created by Larry Yaeger

Polyworld is a particular example of an artificial ecosystem.

• Polyworld space
  
  
  • A plane with impassable boundaries OR
    
  • A flattened torus OR
    
  • A table top
    (See the film "Erik the Viking" (search for edge in the script))
    
  • The above with extra internal barriers

Polyworld screen shot showing table top world, some graphs of various parameters etc.	Polyworld energy Freely growing food (Growth rate and energy value may be controlled.) Dead organisms (An organism dies when its health energy is zero. A dead organism may have non-zero food energy.) Polyworld sex Genetic algorithm mode (Inhabitant fitness is measured at predetermined times according to a pre-determined function. The fittest organisms are bred by the system.) Free-running mode (The organisms select mates and reproduce under their own control.)
Polyworld vision Organisms and other objects are represented visually (for us) as coloured polygonal objects. Organism vision is implemented by rendering a one-dimensional relative view of the world and using this as input to the neural-network brain of an organism. Polyworld thought Organisms have neural network brains employing Hebbian learning methods. An organism's brain determines its behaviour.
Polyworld creature neural architecture including RGB vision neural structure.	Polyworld neuro-science Specific neurons activate a primitive behaviour once a threshold is reached or with a strength proportional to the organism's characteristics and the activation of the relevant neuron: Eating - Replemish energy resources. (Organism's location overlaps food and eat neuron fires.) Mating - Visual rep. as blue colouration (Organism's location overlaps another's and both organisms' mating neurons fire.) Moving - Move forward by an amount proportional to the activation of the moving neuron. Turning - Turn by an amount proportional to the activation of the turning neuron. Fighting - Visual rep. as red colouration. Attacking another organism. (Organisms overlap and one organism's fighting neuron fires.) Controlling field of view - Control of the horizontal cone of vision. Changing colour brightness - Control brightness of some polygons on the front face of an organism. Organisms expend energy on all activity (including neural activity). Energy may be replemished by eating food.

• There are some number of internal neural clusters as specified by the genes
  
  
• Organism's neural states are updated synchronously
  
  

• The frenetic joggers
  
  Run straight ahead at full speed - always wanting to mate, always wanting to eat.
  
  
• The indolent cannibals
  
  Congregate and do not move - kill, mate with and eat each other all on the spot.
  
  
• The edge runners
  
  Run around the edge of a bounded world - eat or mate with all who pass.

Other Interesting Behaviour

• Respond to visual stimuli by speeding up
  
  
• Respond to attack by running away (speeding up)
  
  
• Respond to attack by fighting back
  
  
• Grazing (respond to food by slowing down)
  
  
• Seeking out and circling food
  
  
• Follow other organisms

Tierra was created by Thomas Ray

Tierran space is represented as virtual machine memory Tierran matter is represented as virtual machine code Tierran energy is represented by CPU time

Tierran Organisms

Organisms are represented as sections of virtual machine code in virtual memory.

Organisms compete for CPU time running a virtual operating system to execute their code.

There exist 32 instructions of which an organism may consist.

(eg. MOV, CALL, RET, POP, PUSH, COPY etc...)

Addressing is by template.
(e.g. JMP [template] - execution jumps to the end of the nearest space in either direction in memory which matches the template.)

There are no numeric operands.

Some extra notes on the Tierran machine code are online.

• Organisms have exclusive write permission in their allocated block of memory.
  
  
• All organisms may read or execute any memory cell.
  
  
• An organism may have write access to at most two blocks of memory:
  
  
  • Block 1: the mother cell in which the organism exists.
    
    
  • Block 2: a daughter cell obtained by executing the MAL (memory allocation) instruction and used for reproduction or growth.
    
    
• When Tierran's divide access is lost to the daughter cell but a new one may be allocated.
  
  
• The number of instructions executed by a given organism on its virtual CPU may be altered to favour different sized organisms.

Limiting the population size

• A reaper queue is maintained which culls organisms when they reach its top.
  
  
• New organisms begin at the bottom of the queue.
  
  
• Organisms may move down this queue by successfully executing some instruction.
  
  
• Organisms may move up this queue by unsuccessfully executing some instruction.
  
  
• Vigorous organisms may prolong their life but in general the organism's chances of being culled increase with age.
  

Mutation

• Cosmic rays - at some background rate, the operating system randomly flips bits in the memory.
  
  
• Flawed replication - while copying instructions during replication, a bit may be randomly flipped.
  
  
• Flawed execution - the execution of Tierran instructions is flawed at some rate.
  
  e.g.'s: Result of incrementing a register might be to add two. Bit-flipping may flip the wrong bit or not work at all.

Getting It Started

A seed, self-reproducing Tierran ancestor has been coded by hand (by Ray).

A simulation begins with one ancestor and a random 'soup' of 60,000 machine code instructions.

The ancestor:

• calculates its own size by subtracting the location of its beginning (marked by template) from the location of its end (also marked by template)
  
  
• allocates a block of memory of the same size as itself
  
  
• copies itself into the daughter cell
  
  
• divides, then returns to step 2
  

Tierran Results

Parasites Execute or seize control of the 'copy code' of other organisms in order to copy their own code. This reduces the amount of space the parasite requires and allows it reproductive speed through a need to avoid replicating the replication code, whilst wasting the reproductive efforts of the victim.

Simulation has also developed immunity to parasites, circumvention of this immunity, hyper-parasites, social hyper-parasites, hyper-hyper-parasites etc.

The means by which the above behaviours occur stem from an Tierran's ability to seize the instruction pointers, register values and CPU data from other organisms, change them or use them themselves.

This enables the parasite to improve its own reproductive success whilst limiting the success of the victim.

4. Spatial Organization in Artificial (Virtual) Ecosystems

References:

Oliphant, M.,
"Evolving Cooperation in the Non-Iterated Prisoner's Dilemma: The Importance of Spatial Organization", proceedings of Artificial Life IV, Brooks and Maes (eds), MIT Press, 1994, pp349-352

Angeline, P.J.,
"An Alternative interpretation of the Iterated Prisoner's Dilemma and the Evolution of Non-Mutual Co-operation", proceedings of Artificial Life IV, Brooks and Maes (eds), MIT Press, 1994, pp353-358

• Predation
  
  
• Communication
  
  
• Reproduction
  

...how?

The Prisoner's Dilemma

• The non-iterative version of the Prisoner's Dilemma has been used to study the effects of spatial location on agent interactions
  
  
• Two agents play against one another
  
  
• Each agent must choose one of two moves, co-operation or defection
  
  
• Non co-operative game: players may not communicate before the game
  
  
• Non zero-sum game: winnings of one player are not necessarily the losses of the other
  
  
• The payoff awarded to a player depends on the move made by the opponent as follows:
  
  

  	Co-operate	Defect
  Co-operate	R/R	S/T
  Defect	T/S	P/P

  
  Where: T > R > P > S and: 2R > T + S
  
  Example: T=5, R=3, P=1, S=0
  
  Then if...
  
  
  • both players co-operate they each receive 5 points
    
  • both players defect they each receive 1 point
    
  • one player defects and the other co-operates, the co-operator gets 0 points, the defector 5
    

Strategies for the Prisoner's Dilemma

• Mutual co-operation is of the most mutual benefit
  
  
• Defection is the best solution for a self-interested player
  
  
• Defectors always have the advantage over co-operators, regardless of their opponents...
  
  

  Fc	= percentage of co-operators played against
  Fd	= percentage of defectors played against
  E[Pc]	= expected average payoff for a co-operator
  E[Pd]	= expected average payoff for a defector

  
  E[Pc] = R*Fc + S*Fd
  E[Pd] = T*Fc + P*Fd
  
  But: T > R and P > S, E[Pc] > E[Pd] for any Fc and Fd
  
  
• An ordinary evolutionary simulation evolving the most successful players converges to the stable state of all players being defectors.
  

  This assumes that all players are equally likely to play against any other player in a population.
  

If a spatially-organized set of players is established:

• Players are more likely to play other players in their neighbourhood
  
  
• Players are more likely to mate with players in their neighbourhood
  
  
• Offspring are deposited in the neighbourhood of their parents
  

• Fc and Fd will vary across the space
  
  
• A co-operator will more likely play against another (and they will mutually benefit)
  
  
• A defector will more likely play against another (and they will mutually suffer)
  
  
• The population no longer converges on uniform defection. Co-operative behaviour may emerge and be maintained