Social Learning Strategies Tournament

 

Background

'Social learning' - learning from others - is observed in many species, and is particularly important for humans, as it is underlies our capacity for tradition and culture. Social learning can be a good way for individuals to get information about their environment. However, blindly copying is very unlikely to be useful because information may be wrong, and can become outdated. Therefore, we expect individuals to use social learning on a selective basis by employing 'social learning strategies' - rules about when and whom to copy. But which strategies perform best? Which win out in an evolutionary struggle?

We organised a computer-based tournament to try and generate some interesting answers, as well as stimulate research in this area. We invited individuals or groups to submit strategies which we then pitted against each other in a series of evolutionary computer simulations. We received submissions from a wide range of disciplines, including economics, psychology and behavioural ecology, as well as from people outside academia.

The tournament was overseen by a committee of distinguished scientists with much pertinent experience:

Robert Boyd, University of California, Los Angeles
Magnus Enquist, University of Stockholm
Kimmo Eriksson, Mälardalen University

Marcus Feldman, Stanford University

Kevin Laland, University of St Andrews

The committee has been extensively involved in designing the tournament, and we are also very grateful to Robert Axelrod of the University of Michigan for providing important advice and support with regard to the tournament design.

Download the Tournament Code

If you are interested in this problem you can download the Matlab (R) code for the tournament model and the winning strategy. The code definitely runs on v7.5; it probably runs on some earlier versions but has not been tested so take care in interpreting results if using an earlier version. Instructions for using the code are included in comments within the functions themselves.

The tournament model: tournament.m
The winning strategy: discountmachine.m

Tournament History

Stage I
We were delighted to receive 104 entries from 16 countries. Entries came from a large range of sources including academics from numerous fields (Biology, Physics, Management, Psychology, Anthropology, Interdisciplinary centres, Ethology, Environmental science, Primatology, Sociology, Mathematics, Computer science, Philosophy and Engineering), as well as non-academics, families, and school classes.

After over 100,000 simulations, taking approximately 64,000 hours of CPU time (for which we gratefully acknowledge the the UK National Grid Service), the first stage of the tournament was completed. Details of the Stage I results can be found here.

Stage II
The ten strategies that performed well enough in Stage I to progress onto Stage II were:

* copyWhenYoungThenLearnWhenPayoffsDrop
* discountmachine - our winner!
* dynamicAspirationLevel
* intergeneration
* livingdog
* prospero
* rummer
* valueVariance
* wePreyClan
* whenTheGoingGetsToughGetScrounging

In Stage II, the 'melee' round, the strategies competed simultaneously in a range of conditions - full details of the results and procedures adopted are available in a PDF file here.

Final Results
Tournament Winners

We congratulated Dan Cownden and Tim Lillicrap of
Queen's University, Ontario, Canada, whose strategy discountmachine emerged as the convincing winner
of the tournament. Dan and Tim were presented
with their 10,000 Euro prize at the
EHBEA conference in St Andrews in April 2009.


We also congratulated Ralph Barton and Joshua Brolin,
who are both pupils at Westminster School. Their strategy, whenTheGoingGetsToughGetScrounging,
placed 9th in Stage I. It is a tremendous achievement for these students to have done so well in a competition where the vast majority of entrees were by professional academics. In
recognition of this achievement, Ralph and Joshua were awarded the prize of 1,000 pounds for the best entry from a school.


We would like to thank once again everyone who took part in the strategy, and also the committee members, Rob Boyd, Marc Feldman, Kimmo Eriksson and Magnus Enquist. Full details of the results and procedures adopted are available in a PDF file here. We also have a Facebook page with FAQs and other details.

We have published an analysis of the results in a paper in Science.

Tournament Rules

Full details of the tournament rules and how to enter are given in the following document. The closing date for entries has now passed, so they are for reference only.

Download tournament rules in MS Word format
Download tournament rules in PDF format

If you have any questions please contact the tournament organiser, Luke Rendell.