Learning+to+eat

Prev: DODOM

=Learning what to eat=

In this example we combine TODO and learning. In most ecological models predators (or foragers) are assumed to know what to eat, but in reality that is not clearly predefined. Rather, we could expect the diet to depend on the availability of food in the environment. Furthermore, we can expect that you can learn what to eat. The most basic form of learning is probably **trial-and-error**. On the other hand individuals could mimick conspecifics (**social learning**). Here we will consider the impact of the environment on these processes and contrast opportunity and optimality. Moreover we will look at how diet cultures can arise.

To do this [|van der Post] and Hogeweg (2006) formulated a model where (Fig 2):
 * Individuals search and select resources to eat
 * These individuals live in an enivronment with many different types of food.
 * They select resources depending on their (learned) preferences and on preference expectation, i.e. if you have not seen your favourite foods for a while, you are more likely to accept less-preferred foods.
 * Unknown (new) resources are always tried (sampled).
 * Thus we have a model with TODO behaviour which occurs relative to the environment //and an internal state.//
 * Trial-and-error learning in **solitary** and **grouping** individuals is then compared.
 * Learning is furthermore compared in **patchy** and **uniform** environments.

This model gave rise to the following results (Fig 3): >> which seems quite sensible, but is not what is generally observed in nature.
 * In the uniform environment we see heterogenous diets in groups
 * **"minimization of competition"**
 * In the patchy environment we find homogenous diets in groups
 * **"group diets"**
 * Thus we obtain opposite results: what is being done (which food is eaten) is partly an effect of what you see (which food you find).
 * In mixed environments (uniform with patches superimposed) we can observe both effects at the same time[[image:Vanderpost_fig3.png width="457" height="319" align="right"]]
 * this gives a nice search image for experimentalists and field workers
 * In the patchy world different groups have different **"diet cultures"**
 * This differentiation is observed in "average-good" foods: the very good foods are eaten by everyone, but the "less good foods" are only eaten by certain groups (cultures).
 * We see that grouping makes diet **and** diet makes grouping (cf CHIMPs) (see sorting according to diet in van der Post and Hogeweg 2004)

//Inheritance of "dietary cultures"//
Now we turn to culture and how we can use models to learn something about culture. First we point out that culture is a fuzzy concept:
 * on the one hand we have **traditions** (i.e. stagnation)
 * on the other hand we have **cumulative change** (i.e. progress: standing on the shoulders of giants)

How then can we get a better view of the concept "culture"? Well lets take inhe ritance through learning as a basis and see how far we get. To do this van der Post and Hogeweg (2008) extended their **trial-and-error** model with primitive population dynamics to conduct transmission experiments in patchy environments (cf Curio). They basically introduced a new, naive individual into each group every year, while removing one of the older ones. The results of the model were:

[[image:Vanderpost2008_fig2.png width="269" height="413" align="right"]]

 * Diet traits were inherited over generations giving rise to **diet traditions** (Fig 2)
 * But over the generations there was also an increase in diet quality: **cumulative change**

__Side-effects versus evolutionary optimization__

 * We observe cultural phenomena as side-effects: grouping + environmental conditions gives culture for free.
 * But of course we still need to get our parameters right.
 * However, these parameters can evolve due to all types of reasons that have nothing to do with culture!

//Opportunity-based versus optimality-based//

 * We observe that social learning can arise as a side-effect, not a strategy.
 * Importantly we get environmental-based memory, e.g. in uniform environment we get diet divergence.
 * We obtain coherence between sets of behavioiur:
 * grouping and social learning and how this translates into traditional or progressive cultural phenomena
 * Moreover we obtain alternative explanations:
 * We get **automatic adaptation,**
 * e.g. BUMBLE clock, compensatory feeding, enviornmental change.
 * but also stagnation by doing what you did before (DODOM, Learning what to eat).
 * Lastly, we get long term effects and **long term information integration** (e.g. dietary traditions).

[|Boyd, R.]/[|Richerson, P.J.] 1988. An evolutionary model of social learning: The effects of spatial and temporal variation. In [| Zentall/Galef, //Social Learning//]//,// 29—48. **D. J. van der Post & P. Hogeweg** (**2004**). Learning What to Eat: Studying Inter-relations between Learning, Grouping and Environmental Conditions in an Artificial World. //LNCS//, **3305**: 492-501.[| Download PDF]. **D. J. van der Post & P. Hogeweg** (**2006**). Resource distributions and diet development by trial-and-error learning. //Behav. Ecol. Sociobiol.// **61**: 65-80.[|DownLoad PDF]. **D. J. van der Post & P. Hogeweg** (**2008**). Diet traditions and cumulative cultural processes as side-effects of grouping. //Anim. Behav.// **75**: 133-144.[| DOI]. [|DownLoad PDF].
 * References**

(CHANGELOG 2014-2015)

- Removed, because no longer part of the course:

TODO versus Evolutionary Optimization
One question we could ask about social learning is //when to learn socially or when to learn via trial-and-error?// In other words, when should we expect social learning to evolve? In this light, for instance, Boyd and Richerson (1988) looked at the evolution of //pseudo-genes// that code for individual tendencies to learn socially or individually:
 * pseudo-genes defining social learning or individual learning
 * costs of individual learning
 * social learning then evolves (is adaptive) when costs of individual learning are high and the information obtained from others is still valid
 * this happens in slow enough changing environments (both in space and time)

In our model however, we only implement trial-and-error learning, but we get social-like learning in patchy environment (i.e. not too fast changing in space!) So to a certain extent there is some convergence in both approaches, but the reasoning is different:
 * in our case we should ask when social learning will **not** evolve
 * which is when **no opportunities** for social arise (i.e. when one is not in a group)

Thus we see that
 * social learning is not a **free evolutionary choice**
 * and a convergence of outcomes in our TODO-based approach, and the pseudo-evolution in the optimality approach

- Removed:



COURSE 2006-2007

Learning what to eat: TODO + Learning

- in most ecological models predators know what to eat, but in reality that is not predefined - foraging depends only on density - so how to learn? - trial and error - mimick conspecifics (social learning) - influence of the environment: oppportunity vs optimality - diet cultures

Model - individuals search and select - select on preferences and preference expectation - unknown resources always tried - solitary or in groups - Behaviour TODO relative to internal state + environment

Results Resource Distributions: - heterogenous diets in groups: cf "minimize competition", quite sensible, but not what we generally see - homogenous diets in groups: "group diets" - opposite effects: what is being done is part an effect of what you see - mixed: both effects at same time: nice search image for experimentalists and field workers

- Patchy: different groups have different "diet cultures" - Grouping makes diet AND diet makes grouping (cf CHIMPS)

TODO vs Evolutionary optimization

when to learn socially or via trial and error? When to expect social learning to evolve: - Boyd and Richerson: pseudo-gene sol vs ind learning + costs of learning - social learning evolves: costs of individual learning high + learn from others is still valid - this happens in "slow" changing environments (space + time)

Our model: - we only implement trial-and-error and get social-like learning in patchy environment (i.e. not too fast changing in space!)

i.e there is convergence in both approaches, but reasoning is different: 1) when will social NOT evolve 2) when NO OPPS for social learning arise (i.e. not in group)

- social learning is not 'free evolutionary choice' - convergence of outcome of TODO + (pseudu)-evolution

"CULTURE" - fuzzy concept - on one hand: traditions (stagnation) - on other hand: cumulative change (progress): stand on shoulders of giants

Can we get better view of concept "culture"? - take inheritance through learning as basis

Conduct Transmission experiments: primitive population dynamics Results: - Inherited diet traits: TRADITIONS - Over generations increase in diet quality: CUMULATVE CHANGE - When what? : depends on Selectivity (N) and movement forward after not finding food (M)

Side-effect vs Evolutionary optimization: - side-effect: group + environmental conditions, culture for free - but still need to put parameters in appropriate way - these parameters can evolve due to all types of reasons that have nothing to do with culture!

Opportunity-based vs optimality-based - social learning as side-effect not strategy

Environment based shared memory - mixed environment diet divergence

Coherence between sets of behaviour - grouping - social learning - N/M type of cultural phenomena

Alternative explanations - automatic adaptation -- BUMBLE clock, compensatory feeding, environmental change - also stagnation by doing what you did before!