S C A L E Third Artificial Life Weekend October 20-22, 1995

Participants

Name: Institution Status email address:

Chris Adami Cal Tech Faculty adami@krl.caltech.edu

Peter Andrews UCLA Grad S andrews@biology.ucla.edu

Takaya Arita Nagoya U. Faculty ari@info.human.nagoya-u.ac.jp

John Batali UCSD Faculty batali@cogsci.ucsd.edu

Mark Bedau Reed Faculty mab@reed.edu

Rik Belew UCSD Faculty rik@cs.ucsd.edu

Titus Brown Reed UG brown@krl.caltech.edu

John Carnahan UCLA Grad S carnahan@biology.ucla.edu

Johan Chu Cal Tech Grad S jchu@cco.caltech.edu

Gary Cottrell UCSD Faculty gary@cs.ucsd.edu

Deborah Forster UCSD Grad S forster@cogsci.ucsd.edu

Chris Fry UCSD Grad S cfry@cogsci.ucsd.edu

Nick Gessler UCLA Grad S gessler@anthro.sscnet.ucla.edu

Bill Grundy UCSD Grad S bgrundy@cs.ucsd.edu

Mike Hamilton UCR Faculty mhamilton@applelink.apple.com

Tom Kammeyer UCSD Grad S tkammeye@cs.ucsd.edu

Brian Keeley UCSD Grad S bkeeley@ucsd.edu

Adam King UCLA Grad S king@cs.ucla.edu

Mark Land UCSD Grad S mland@cs.ucsd.edu

Steve Lansing USC Faculty slansin@mizar.usc.edu

Charles Ofria Cal Tech Grad S charles@krl.caltech.edu

George Marnellos UCSD Post Doc gmarnell@cs.ucsd.edu

Filippo Menczer UCSD Grad fil@cs.ucsd.edu

Eric Mjolness UCSD Faculty emj@cs.ucsd.edu

Joao Munoz UCLA Grad S munoz@biology.ucla.edu

John Northan UCLA Grad S jnorthan@cs.ucla.edu

Chris Rosin UCSD Grad S crosin@cs.ucsd.edu

Chuck Taylor UCLA Faculty taylor@cs.ucla.edu

Eric Welton UCSD Grad S ewelton@cogsci.ucla.edu

Tony Lewis UCLA Faculty

Ken Hayworth UCLA UG

Tina Hallen-Cottrell UCSD

Kyle Cottrell UCSD

Format will be series of 10 min talks, basically without question period. We will keep track of the questions date them tonight, then take up the questions tomorrow. {CT distinction between clarifying and discussion questions}

9-12 3 hours = 18 people

13-3 lunch and hiking break - blife hunt?

3-5:30 2.5 hrs = 15 people

applaud when time is up. 8 min watch beep, then applause after 10 min

Misc Notes:

Get account from UCSD from Rik -- to use Encyclopedia Brittanica

Rik says follow-up from James 1, BJ Williams is introducing EB online to kids.

now grades 3-5 mixed class (reading level a problem, but are handling it)

study whales, then read EB and highlight the words they don't know

use these for vocabulary words. Use propedia to see species relationships

helps to then use propedia for other things

Larry Landweber has been helping to get 2 cable companies in Madison. Promotes competition and probably keeps rates down. Within the year they will be providing Internet over cable. What Rik is waiting for.

Rik's contribution to this effort will be to help provide some content to the Internet that people will care about. Will have map of Madison, and people can become involved with planning their own school bussing routes.

Rik's own grant proposals are to extend his own Internet searching agents to distributed molecular biology data bases -- LEE approach

Mark Bedau agrees that animal rights approach to alife is probably appropriate (had not thought of elevator law). He says that mainstream philosophers do not generally regard this field as well-developed. Among the most noteworthy in this area are Peter Singer, Paul Taulor, and Ba xxx. See his article in ECAL I, where he also lists the main philosophical problems he believes are raised by Alife.

Send to Chris Adami:

CSEOL stuff -- e.g. Jerry Joyce

Nov 3-4 meeting. He will probably attend

Spoke to Adami about his definition of life in terms of its peculiar statistical mechanical properties.

Points to first be established for attempt to determine if something is alive:

1. spatial scale must be given -- electron vs. galaxy or in between

2. temporal scale must be given -- microsecond vs. million years

(have little info in either extreme)

To determine if something is alive:

what is the natural time scale?

is order (entropy) maintained on its natural time scale?

(important that it do so in the face of mutation)

iron pyrite is very slow (billions of years)

DNA is fast - left alone it would quickly decay

Life = how a low entropy system is able to maintain this low entropy through time scales that far exceed the natural time scale for reaching maximum entropy.

[CT Check this with Chris]

CT Buckley here in May?

Fred Roberts using Swarm?

For next year:

Mark Lange

USC philosopher (from Bedau)

Arbib group

Riverside philosopher

Dyer, Hillis, Microsoft person

John Batali

Evolution of signaling system. each creature has a map, send and receive signals. How does popuation come up with a coordinated signal?

If both rewarded, then you can show a maximally effective language will evolve

if only the sender is rewarded, then one can show it will also -- like kin sel

today focus on when they are sending threats

display their motivational state, and interpret other's display

can interpret others and adjust your behavior accordingly -- fight or flight

Model 1 honest signal - is ESS to stay in region of motivation space

Model 2

CT

Msmith history of complex systems

CT view

idea is to explore new vocabulary

Carnahan

Swarm provides

common vocabulary - documentation

easier to replicate

built on common platform [c++, on serial machine now]

Assumptions of current version

Eric Weldon

beginning student -- was working on heavy metal cockroach at Illinois

now in cog sci at ucsd

trying to get synthetic nervous system for the cockroach

looking at general properties of Gas that might help him to ultimately evolve this

(still somewhat inchoate)

Bill Grundy

wants to explore value of pleasure and pain

would seem to be useful only if learning an option [CT an interesting view - why not just to avoid or approach]

inspired by ackely/littman

CT dros that learn or not, more exquisite sensitivity?

Gary Cottrell

neural net models of language

(e.g. word sense disambiguation)

(nets never care if they talk or not -- what is the motivation)

with alife, can presumably evolve creatures who want to talk

systems that talk

Elman's amoebas - get angle and distance to nearest food

evolved to go to food, learn to find food more effectively

Dyer's work -- motivation present, can hear/speak

wants to include hearing and speaking with this, combine both features

one thing might evolve is turn-taking to avoid cacophony

Chris Fry

Bird song model

song sparrows

male can say: (a) get out of my territory; or (b) come have sex with me

how do they come to recognize the different syllables?

teach artificial NN to recognize and produce these from hearing others

different phases,

perturb, recognize, produce, feed back through, compare

gettng better or worse determines if keep perturbation

sounds like net talk

Tom Kammeyer

last year, evolving grammars for sorting networks

now a more general aspect --

stochastic context-free grammar induction using Gas

Takaya Arita

LangE model

storing and pattern completion in the brain

Two points to be made:

1. Concepts can be extracted from sensory input

2. messages can take on shared meaning

Will be attempting to develop this for mobots

Titus Brown

from Avida

number of interactions between objects should affect accuracy of simulation

Mark Bedau

Does 2 sorts of things:

Philosophy -- writing book -- Evolution, Life and Mind

Burning up cycles with Norm Packard on evolutionary waves

will talk about dream of what alife can become

now trying to find out if there is such a field

1. must be unity -- Swarm addresses this

2. relate this to Nature -

picture he wants to give is to focus not on models, and not to focus on general purpose model (e.g. Swarm),

rather focus on fundamental properties of systems that capture the fundamental properties of evolution

e.g. look at an arbitary model, say how complex it is

(What Jim Crutchfield is trying to do)

what properties are part of adaptive system, what is properties of random flux

will give feel for how they are measuring what is adaptation

look at which genes are being used, through a lineage

when you do this, the important adaptations just jump off the page

with him, with Kammeyer (above), someone with ECHO, and a few others

easy to apply to data from Nature,

gives examples

(says Fossil record online from Chicago)

John Northan

looking at ways to evolve interesting things from NN where chromosomes not too long

Kitano

graph L systems to generate NN system

so evolve the generator

Edelman

brain develops by neural group selection, gets rid of those part not used

used by some people at Honeywell, have layers of neurons that map to next higher ICGA III

Eric Mjolness has paper on how to do Kitano's method a bit more softly

also Nolfi and Parisi

Steve Lansing

Two models and a theory

social bonding and ecology

balinese water temples

link between optimal planting and irrigation system

to optimize water use, each should stagger small patches

to optimize pest control, should have large patches lie fallow

Use hill climbing to find optimal, essentially what Balinese use

several interesting things happen since Alife III paper

1. mean fitness goes up beyond highest originally (beyond Fisher's fundamental theory)

Now working with Barbara Smuts

What is the relation between ecological circumstances and social structure

(about 100 primate species, about 35 social patterns)

question, are there attractors in primate social space?

now working on this in Lisp environment (Skate)

Put in lots of objects, to one kind of environment or another

let each behave to maximize own fitness

finds patterns that emerge, some more commonly than others

Now with Smuts must look at mapping onto real life

Chris Adami

last year circulated 15 problems his lab was working on using Avida

have now finished 4

abundance distribution

area law

speciation

diffusion of information

will talk about directed mutation hypotheses

1. Luria And Delbruck 1943 -- mutation not prompted by environment

2. Carins and Foster, 1993 -- when environment not so lethal, can have dm

will use Avida to explore conditions for increased directed mutation in stressed environments

Charles Ofria

looking at speciation in Avida-- when should the creatures be regarded as different species

avid critters are asexual

line them up in optimal fashion, try crossing over at each point, see if still function

figures that if sig different in algorithms, then recombinants wont function

{CT stresses fundamental importance of recombination for sex}

go to higher level of phylogeny, get genetic distance between critters

[says D is effectively Hamming distance]

Johan Chu

new implementation of Avida-like system

Sanda

parallel implementation of Avida

running up to 5,000,000 cells. uses 512 node Intel Paragon

Propagation of Information

in a homogeneous grid of B, put in one member A with higher fitness

will spread out in a wave. can calculate expected wave speed. fits

Will be growing crystals with mutation

Wants to look at learning to do things, e.g. to add

how much faster if increase pop size

ultimate goal, to look at real space and genetic space

if I change this parameter, system will respond in such a way

----------------------------

Adam King

rats can learn an interval of each 24 hrs, but not each 9 or 10 hrs

can learn 20 second rates and a number of other short intervals -- have trap lines

timing seems to be an intracellular oscillator

Eric Mjolness

was trying to develop neural nets by recursion

ran into two standard problems:

circuits could solve some toy problems, but not as well as hand designed

if tried to reuse for biological modeling, they were not sufficiently realistic

Now thinks:

in visual, hand done relaxation neural nets seem to do better

instead of big environment, try to evolve the subsections to big problem

a structured neural net, because these are structured problems

e.g. find nn to do translational invariant recognition of bit strings

thinks development likely to be important in some way I missed

has several examples of this in relation to Drosophila

George Marnellos

applying tools that Eric described to model in neurobiology

initiation of neuroblast in Dros embryo

neuroblast arises from a sheet of cells in the epithelium

there are clusters of cells that express some gene that is recognizable

then reduces to a single cell in the middle of the cluster that will eventually become the neuroblast

Attempting to explain How does this occur? in terms of Eric's ideas

as matrix of communications among cells, will fit by minimization

Has a demo of how clusters of cells can determine which single cell differentiates

says simulated annealing works better with 6-9 parameters, then GA beyond this

Nick Gessler

artificial cultures

has quote about "personoids" from Stanislaw Lem in early 1970s

tells about a neat web sites where observers vote to get better art works. has a nice gallery

a strong position -- humans are computing objects, and understanding their society can be best understood as societies of computational objects

Chris Rosin

Competitive Coevolution, where advantage to one species results in a disadvantage to others.

most work is related to problem solving

looks at optimal solutions for general games

find better general problem-solving if there are a variety of oppones

prevents forgetting

pedagogical sequences of examples

Filippo Menczer

LEE - using this on WWW - he's been working on this for past year

Today talking instead of more recent work

looking at what features of behavior are required for some environment

Three little expts with LEE to see how changes in environment leads to creation of new collective behavior

(a) patchiness

(b biodiversity

(c) seasonality

Mark Land

How do you get a program to do what you want it to do?

one way is genetic programming

seems to be too problems with this

1) complexity -- can solve up to 5 bit parity with 1,000 nodes, cannot do 6

2) which primitives do you use?

Koza good at picking primitives, but no reason to think that random primitives would do nearly so well

How get around these problems?

hopes autocatalytic sets will help to get around this

when 2 objects interacts they create others

when one operates on one object it makes something, and when other operates on first it too produces meaningful operation

why lisp is nice, objects are both operators and operands

see article by Fontana on algorithmic chemistry

Fontana's does things, but does not solve a problem, Land wants to adopt these to solve some problem

Peter Andrews

Wants to discuss two things

1. working with David Fogel to extend ESS, showing ESS aren't necessarily stable

based on assumptions like infinite populations, average payoff rather than stochastic payoff

did simulation where varied the parameters. They had very strong selection. When made it less strong, then stochastic effect not so great

e.g. when strong selection, go 200 Hawks to 400 hawks to 200 etc. essays 350. but never hits 350.

Peter analogizes, not to strongly, to economic behavior

Batali points out that ESS is a linear approximation and JMS says it is appropriate for low selection only

2. Fungible units in Avida/Tierra

"energy" is currency of biology

someone said maybe information is

perhaps it is time

convertible unit in Avida is time bonus

would like his units to exchange these

Joao Munoz

Units of selection

typically accepted that units of selection

genes)individuals)kin)social group)population)community)ecosystem

says frequency of component parts change in time

other way is that relationships among parts are changing e.g. interactive gene systems

perturbation at lower level might have a big local effect ,but typically seem small as move up

but change at higher level typically has a larger change at the lower levels

looks for explanation of this is related to feedback

other is that all these are part of system,

and in systems the whole is typically more than the sum of its parts

W > S(p) and (p) >p

example? enriched by being at the meeting, getting more than if were alone

to explain changes, need to understand changes at higher level, because the emergent properties

thinks that computer modeling will help to understand this

Rik Belew

distinction between genotypic space and phenotypic space.

typically evaluation in phenotype

selection on phenotype

wants to walk close to a lamarkian line here

stress-related changes in mutation rates might give something here

draws a picture I should consider

typically think of development as maturation as a "ballistic process" where phenotypic change as maturation is independent of environment

learning is different, the environment has a big effect on phenotypic change

interesting to see so much language here

Deborah Forster

not a modeler, trained as a behavioral ecologist

social complexity and cognition -- studies primates

in 1970's it was proposed that social complexity produces higher intelligence somehow, at least in primates

social complexity ------> [black box] -------> intelligence

what is attractive about modeling is that you can take account both of proximal causes and evolution

with baboons, 25 years, maybe 3 generations (old world monkeys)

if social complexity affect evolution, wants to see nuts and bolts about how this occurs

males leave females stay

leave females stay

leave females stay etc

get matrilines

female hierarchy is very stable over generations

the new sibling always gets the rank immediately below the mother

implies youngest always dominant over older

when males come into troop they no longer have the support of others in their group

must establish relation with others means male dominance is always unstable

the longer you stay you go down in dominance in agonistic encounters, but get social bonds, with females and others

in one on one the new male is most dominant, but as stay you gain alternatives to brute force

establishes bond with lowest female by being nice to her infants, and works up

so access to females, etc. typically gets better as he grows older

idea is that get complexity of relationships probably requires intelligence

Harold Kester

Encyclopedia Brittanica - in charge of advanced technology at EB

here on a personal basis - read Levy, Dawkins, Churchland

practicing Buddhist -- reading Bodhi darma

interesting ideas of consciousness

Questions Rik abstracted for discussion on Sunday AM

---------------------

good animal models of communication

fungible unit of exchange

proximate vs ultimate causes

education/commercial use of models

development - maturation comosed with learning

autocatlytic basis

How to use Mike's data

-----------------

Time-based GIS for simulation -- Lansing

How complex adaptive systems useful

multimedia

models are great multimedia

have a model with graphics inserted to application

e.g. interested in Java to do this

discovery course they are working on

Brittanica University

tools for people that would make programs for Britanica online

units on web

Brittanica discovery series of junior college

you are a bee and you must communicate to other

you do your own genetic engineering

Filippo says game of wimsatt

importance of user interface

example of mosaic/Netscape for using web