TODO

Prev: Opportunity vs Optimality (overview) Next: Self-organization in moving groups

=TODO=

//Introduction//
[|Herbert Simon], the founding father of[| artificial intelligence] said (1969):

"//An [|ant] as a behavioural system is quite simple. The apparent complexity of its behaviour is a reflection of the environment in which it finds itself."//

which we can paraphrase to:

"//An [|human] as a behavioural system is quite simple. The apparent complexity of its behaviour is a reflection of the environment in which it finds itself."//

The basic idea here is that given enough individual-based diversity, individual versatility becomes expressed when individuals meet different circumstances. Thus while an ant may want to walk straight to get back to its [|anthill], its actual meandering route home is a reflection of the obstacles it encounters in the environment. In this light it is also interesting to consider [|apes] that grow up in environments with people where they can do all sorts of human-like things, but if you go into the field one observes them mainly sleeping and eating. This also suggests that apes have "excess capacity" for their natural environment. So how to understand this in an evolutionary sense? On the one hand finding food may be harder than we think. On the other hand, doing easy things //well// may allow you to do hard things //a bit//. In any case, the **environmental structuring** plays an important role.

//Environmental structuring//[[image:http://upload.wikimedia.org/wikipedia/commons/b/b8/DLA_Cluster.JPG width="320" height="214" align="right"]]
One of the prototypes for environmental structuring is **//[|diffusion limited aggregation].//** This is for instance the process of growth of tree-like crystalline structures. Important is that the process defines its own environment:
 * There is diffusion of particles. However these particles becomes fixed on already fixed particles and this generates a branching pattern.
 * As this goes on, getting to the middle gets harder and harder because fixation probability on the periphery is already high.
 * This is happens in: crystal growth, bacterial growth, coral growth (division of polyps depends on food particle).
 * It is a very passive process.
 * At the same time this process makes the environment complex (cf the behaviour of an ant //makes// the environment //complex//).

The themes we address in this section are (i) [|self-organization] and (ii) how simple rules can generate complex behaviour. In terms of model formalisms we have so far looked mainly at CA, which is mainly space based. For synchronous CA, patches are active units, for asynchronous, one can also grow into empty space. Now we take a look at full [|individual-based models] (IBMs):
 * Individuals are simple (in)finite state machines
 * Individuals are located in space
 * Individuals interact with their (possibly complex) environment and with other individuals in this environment.
 * Inidividuals keep the same behavioural rules, but can changes their environment and therewith their "input".
 * Space is not necessarily discretezed, i.e. where individuals sit can be a continous variable
 * Fixed-time step versus [|event-based]: once in a while something happens (cf Gillespie except delays are now locally induced, per individual).

//The TODO principle//
Inspired by Herbet Simons "reflection of the environment", Hogeweg and Hesper conceptualized the TODO principle ([|Hogeweg & Hesper 1985], [|1991]), placing the focus on: > This principle emphasizes that behaviour is steered by local information. > However, this local information is generated by what is done//,// and there can be external memory for behaviour (collective memory for all individuals in a shared environment).
 * //"Do what there is to do!"//
 * //"Do based on what is done!"://

Thus we can observe flexible behaviour arising from rigid rules. Furthermore we might expect to see **automatic adaptation**: behaviour will not be adapted to something that is not there, i.e. no nonsense behaviours will arise. This stands in contrast with common assumptions in behavioural models about optimal behaviour where the environment is often not taken into account. This can lead to rather ridiculous analyses since assumptions are made that individuals do things without cues from the environment.

A first example of opportunity vs optimality based behaviour are the robots of Rodney Brooks. Rodney Brooks designed state of the art robots which were clever and could make plans. Such robots would think a long time to make a plan for an optimal path through a room. However, the robots would take so long that in the mean time the room might have changed in which case the robot would have to rethink everything. In other words, if one needs the //best model// (i.e. to plan path), **the best model is the world itself**. It that case it is better to generalize and solve problems on the way (i.e. start walking and change direction if you bump into something). Thus using feedback from the environment is a good planning tool. In this way we can see TODO as an alternative explanation for behavioural patterns which are normally explained in terms of evolutionary optimization.
 * Robots and [|Rodney Brooks]**

Next: Self-organization in moving groups

H. Simon (1969) //The Sciences of the Artificial//. MIT Press, Cambridge, Mass, 1st edition**.** [|P. Hogeweg and B. Hesper (1985) Socioinformatic processes, a MIRROR modelling methodology. J. Theor. Biol. 113, 311-330.] [|P. Hogeweg and B. Hesper (1991) Evolution as pattern processing: TODO as substrate for evolution, From Aninmals to Animats (eds. JA Meyer and SW Wilson) MIT Press Bradford books pp 492-497] Brooks, R.A. (1991) Intelligence without reason. in: Proc. Int Joint conference on Artificial intelligence, pp569-596
 * References**

TODO: do what there is to do.

This idea focusses on how behavioural complexity is structured by the environmental context that the behaviour takes place in. In other words: it is complexity in the environmental contexts that leads to complexity in behaviour given simple behavioural rules. Therefore in order to interpret the significance of behaviour we should understand how the environment structures behaviour. In this way we avoid over-interpreting the functional significance of behavioural complexity in terms of behaviour strategies. Instead we can tease apart complexity that arises "for free" and how behaviour evolves within such a context.

Important in this is how the environment structures opportunities for behaviour, and how interactions between behaviour and the environment can lead to reinforcement of behaviour.

Description of TODO idea, smarty, bee's nest

COURSE 2006-2007

Herbet Simon (1969): founding farther artificial intelligence; said something like ...

"an ant as a behavioural system is quite simple - the apparent complexity of its behaviour is due to the complexity of the environment in which it finds itself."

"a human as a behavioural system is quite simple - the apparent complexity of its behaviour is due to the complexity of the environment in which it finds itself."

IDEA: given enough individual-based diversity, to get versatility out happens primarily when you meet different circumstances i.e. an ant walks straight, thus its meandering walk is a reflection of the environment

e.g. apes (cf van Schaik): apes that grow up in environment with people can do all kinds of things, but go into the field and they don't do much but sleep and eat etc. It would appear that they have excess capacity. So how to understand that in an evolutionary sense? Well finding food may be harder than we think. To do easy things WELL may allow you to do HARD things A BIT. - in any case, the environment plays an important role.

Prototype for environmental structuring: Diffusion Limite Aggregation

- making own environment: fixed diffusion of particles, get fixed on fixed particles generates branching pattern - getting to the middle gets harder and harder because fixation probability on periphery already high - cf crystal growth, bacterial growth, coral growth (division of polyps depends on food particle) - very passive process - but also: behaviour of ant MAKES environment COMPLEX

THEMES: - self organization - simple rules to complex behaviour

FROMALISM - so far with CA (individual based) but space based - synchoronous: patches are active units - asynchronous: can go into empty space

FULL IBM - individual located in space - behaviour depends on local neighbourhood - not discretized where they sit - not so much space limited - fixed time-step vs event: once in a while something happens (cf Gillespie except delays locally induced)

TODO - behaviour steered by local information: Do what there is to do - local information generated by what is done: Do based on what is done: external memory for behavour (collective memory for all individuals in shared environment) - flexible behaviour from rigid rules - AUTOMATIC adaptation: not something relative to something not being there, no nonsense behaviours

Note: many ethology studies assume behavour is optimal for something and environment is not often taken into account. This leads to rather rediculous analyses since assumptions are made that indiviudal do things without cues from the environment

AI - Rodney Brooks 80's 90's - met state of the art intelligent robot: clever and can make plans - robot thinks very long to make plan for path through a room - however takes so long that in mean while room changes and then needs to rethink everything So if need best model (e.g. to plan path): BEST MODEL IS THE WORLD ITSELF - Better to try and generalize and solve problems on the way ... - using feedback from the environment is a good planning tool So: - TODO as ALTERNATIVE explanation for behaviour patterns which are normally explained in terms of evolutionary optimization

TODO1: very simple Deneubourg - self-organization in terms of ants - primary observation: distrurb ant nest and ants reorganize and sort out again. How can stupid ants do that? - algorithm: if see object without many of that object around pick it up, when see many around drop it. - this works! - this leads to distributed behaviour, not COLLECTIVE behaviour (i.e. one ant can do it) - similar to SWARM Intelligence (e.g. cluster analysis).

TODO2: Behahavioural differentiation - insect castes: worker bee behaviour depends on age Seeley (1982): 1) % time different ages do different things: division of labour over time 2) types of behaviour in different parts of the nest: center - periphery Might suggest driving force based on environmental cues?

Model: Michiel van Boven (1991 - unpublished?) - model where all behaviours could be done, but triggered only by TODO: if see ... then do .... (i.e. triggered by environment) Results: - young bees born in center do behaviours associated with center (e.g. clean cells, feed brood) - older bees forage - just the spatial structure of the nest is sufficient - behaviour regulated by TODO - enough to stay put for a while where you are

This shows that: Under strict circumstances it is possible that behavoural differentiation is TODO based

Since a short while we have the complete bee genome (REF) and now know: - differentiation forager-nonforager bee: protein kinase over expressed: genetic expression pattern quite different

So model was excercise in showing how powerful TODO can BEE - however, of course when foraging: doing different things can also generate differential gene expression

So what comes first? - first TODO then GENE expression or viceversa - probably combi.

Recently: - cluster analysis of gene expression patterns of different bees shows - cluster of foragers - also gene expression of young like older foragers and they forage: so apparently some flexibility to differentiate sooner - for other behaviours before forage: not much gene expression differences

Therefore: - by doing what is TODO individuals adapt to the enviroment - By doing something: individuals change to do it: LEARNED to do it.

Van Schaik: - orangs being huntted: take baby home - rehabilitation centers: trained to get used to forest: but prefer camp! and try to get back.

Generation of novel invariant features - so if interested in social structure: e.g. primates - but lets not specify: could they be a side-effect on things individuals have to do anyway? Model strategy: - make model without behavour in which one is interested and OBSERVE

Model: CHIMPS (te Boekhorst and Hogeweg 1990) - eat - may take other chimps into account for finding food - also want to mate: males check for receptive females - females need more protein - live in surroundings with: FRUITs and PROTs: estimated how much fruit and prot available and what kind of chunks in terms of chimp hours

Results: - random paths of movement and eat nearest food - study group compositions

Harcourt (REF): often observe all-male groups - and males and 1 female (receptive) - expl: males band together to defend territory, seen as advanced feature (war!)

Model: same pattern and even more exaggerated male grouping -CHIMPS don't have to want to do this to do this - epiphenomena: just happens to be the case - therefore speculation about its purpose, i.e. evolutionary pressures, is nonsensical to think this grouping structure gives fitness WHY epiphenomenon? - nearest food same for everyone: produces grouping - males check for same receptive females - females split up because PROTs sources are more depletable (Note: In ORANGS - food distribution dependent travel bands arise according to similar reasoning)

However: 1 matching observable is a bit minimal! but several observables that have pattern relative to each other makes a stronger case. - walking distance: males walk further. Goodall: males 4km, females 2.7km Model: VERY SIMILAR! This is because males are more often in groups, and deplete food faster and so move on faster.

Moral of story: - CHIMPS are not entities that are STUPID - but: some observables are too basic that from those observables can not infer that chimps are something other than stupid - given that you do certain things other things will emerge Therefore: - the practice of taking an arbitrary feature and then trying to find how it is optimally evolved, cannot be done because many are side-effects! - with models: study how different behaviours relatied to each other to study how some may emerge as side-effects

SO: TODO: behaviour determined by local information - study side-effects OPPORTUNITY-BASED vs OPIMTALITY BASED (largely forgetting about opportunities) - if environment changed: environment shared memory (artificial intelligence=blackboard, ethology=stigmergy). - complex behaviour: other behaviour mapping -> from optimality point of view multi-peaked, reinforces opportunity vs optimality, alternative explanation relative to optimality/ functional, i.e. a CONSEQUENCE, an automatic adaptation: don't need to adaptation process to eat what is not there!

NEXT: TODO + INTERNAL CHANGE : memory in individuals.