Genotype Phenotype

Involved

• Will Braynen
• Simon Angus (s.angus AT unsw.edu.au)
• Mollie Poynton mollie.poynton@nurs.utah.edu

Concept

Currently, when we do modelling in economics and other social sciences, we are often creating agents purely as strategies -- we have a population of strategies, withought regard to the characteristics of the carrier. This is an old debate in biology between Phenotype (observable, measurable traits) and Genotype (normally strategy information such as motivation or game-play (in economics)). Should we worry about the current confounding of the two? .. Perhaps not, but if firms have a motivation/strategy for their behaviour, and (say) lose a production line due to a power-failure, or storm, then essentially their observable characteristics have changed, but their motivations/strategy remains the same (they may have a strategy for dealing with this event, but the point remains).

Modelling Thoughts

So we need to build a richer description of agents that are embodied in some way. Some have done this in the biological evolutionary literature by using external 'tagging' to indicate type characteristics, but these often have nothing to do with genotype, they are normally just modelling tools. (One could argue this for the firm model above -- that the firm loses a part of its 'body' is not coded for by strategy.) In economics, we need to ask what is a phenotype for a firm? What is it above and beyond its strategic decisions? Or for an individual, what determines its fitness/sucess, independant of its strategic play?

This ties into ideas of levels of selection -- what we are really arguing is that strategies do not play strategies, but individuals who have a strategy play against other individuals with their own strategy. The battle is at the observable/phenotypic scale, not the strategic... or is it?

Are we just talking about path-dependance in an agent's fitness/relative-success? .. for instance, the classical definition of phenotypic variation is given in the wikipedia link above:

genotype + environment + random-variation → phenotype

or for our purposes we would have:

strategy + environment + random-variation → phenotype

recall, these are leading to how the agent now behaves, not how it fairs in the world (which can also be affected by 'environment' and 'random-variation' at the discretion of the modeller...)

Thinking about this some more, finite state automata might be a good example of an agent model for such a difference. Since the behaviour of an agent might be 'All-D' towards one agent, but towards another (because of the complex state arrangement of the FSA) it might play 'CDDDCCCDDD...etc.' and give a rich behavioural basis, and this, from the single genotype (strategy encodded by the FSA). But is this enough?

Refs

1. Krakauer, D. Domains of Interaction in Evolution: Transmission, Construction and Selection 2004