Madsen and Lipo 2007 - Beyond Beanbag Models of Cultural Transmission

I really started looking at the idea of incorporating “meaning” or “structured information” into cultural transmission models back in 2003, when I was the co-founder and VP of Engineering for a software startup (Network Clarity). I was reading everything I could get my hands on about the Semantic Web, description logics, and ways of doing deductive reasoning in a declarative manner (i.e., outside Prolog or pencil-and-paper predicate logic). We were attempting to represent the common-sense knowledge that network engineers had about a properly configured, properly functioning network into rules which we could use to check router and firewall configurations at the semantic, not just syntactic level.

Since I had a day (and night) job, my notes from back then languished until 2007, when I wrote them up into a research note, co-authored with long-time collaborator Carl Lipo. Carl didn’t actually write these notes, so any idiocy or naivete is solely mine, and I’ve added some contemporary comments to put them in context.

Mainly, this a record of what I was thinking about back then with respect to cultural transmission and semantic relationships. Our current collaboration on structured information and the so-called “Semantic Axelrod” model are derived from this note, and it remains a valid proposal for constructing models which combine copying processes, cognitive bias, and traits that bear real semantic meaning and relationships among one another.

Such models continue to be essential today, to allow us to understand how various social learning processes are conditioned not just by space, population structure, or cognitive bias, but by the actual content of the information being transmitted. Discussion of CT models that have semantic content between traits have remained verbal and heuristic, and formal models of transmission continue to locate all of the deep content in CT processes within the copying rules or cognitive biases which shuffle “tokens” within a frequency space. Real evolutionary processes are not token shufflers; the things whose frequency are changed by population processes have real effects and meaning. In modern terms, this proposal is our first entry into an “evolutionary developmental biology” or “evo-devo” of cultural transmission, which we argue is badly needed.

(Mesoudi and O’Brien 2008) offer one of the only treatments of structured information to date, in their paper examining the fitness of hierarchical versus non-structured cultural information. Their work showed that there are good reasons why we should expect cultural information to be transmitted in varying sized, hierarchical pieces, which bear relationships to one another like “recipe step” or “prerequisite,” but their work was not meant to directly incorporate such relationships into the dynamics of CT models themselves.

An ``evo-devo’’ for cultural evolution is greatly needed in order to enrich the range of cases our theories can address. We hope to help kick off such an effort, and offer these historical notes as bona fides of our interest and thinking about how to go about it.

Madsen and Lipo 2007: Beyond “Beanbag” Models of Cultural Transmission: A Proposal for Exploring the Effect of Semantic Relationships on Trait Distributions