Name: Luke Rendell
Institution: University of St. Andrews
Research: Using the NGS to run a computer tournament on social learning strategies
We know that many organisms can learn for themselves and also acquire information from others (social learning) but we have little idea of what kind of strategies for combining these two information sources we expect natural selection to favour.
To try and answer this, Luke Rendell from the Centre for Social Learning and Cognitive Evolution at the University of St Andrews ran a computer-based tournament on the NGS. All-comers were invited to compete for a 10,000 Euro prize by devising a strategy to guide how agents learn and prosper in a simulated evolutionary environment.
A clear account of how and why social learning is really so valuable, and what exactly it is about the way we do it that sets us apart, has eluded evolutionary theory for decades. Understanding how we should expect natural selection to shape learning strategies and biases is an essential part of the science of human behaviour. It can help us as a society when faced with problems we create for ourselves with our own behaviour, from unhealthy lifestyles to violent fundamentalism.
The entries submitted to the tournament consisted of a learning strategy, defining how an individual agent would make decisions about when and how to learn, based on the information they currently had. These strategies then competed in evolutionary computer simulations, to see which would win out in a virtual world.
Luke and colleagues constructed an evolutionary agent-based model. To succeed in this model, agents had to acquire and exploit knowledge of the environment, modelled as a multi-armed bandit (slot machine), in order to maximise their payoffs. Entrants to the tournament had to devise a strategy, specifying how agents would divide their time between learning for themselves, learning by copying others, or exploiting knowledge they had previously learned. Effective strategies would collect more payoff, and leave more copies of themselves in subsequent generations.
Luke coded the simulation model in the commercial software MATLAB and then ported it to the open-source Octave software which was made available on a number of NGS clusters. The tournament attracted over 100 entries, so running it meant they had to run over 100,000 individual simulations. This took just over 60,000 CPU hours to complete.
Luke had very few options available as 60,000 computer hours would have been a deal breaker if he had been restricted to in-house computing resources. It would have meant the project could not have been completed successfully. Thus, while the advantage was that they could run many simulations in parallel, the principal advantage was more basic than that – doing it on the NGS meant it could be done at all.
Luke explained “We are incredibly grateful to the NGS for the helpful and flexible way that resources were made available to us. Without it, our research would simply not have happened. We are looking forward to working with the NGS again to run follow-up tournaments”.
Funding body - supported by the EU Framework 6 CULTAPTATION project (European Commission contract FP6–2004-NESTPATH-043434)
Associated published papers - Why Copy Others? Insights from the Social Learning Strategies Tournament. Rendell, L., Boyd, R., Cownden, D., Enquist, M., Eriksson, K., Feldman, M. W., Fogarty, L., Ghirlanda, S., Lillicrap, T. & Laland, K. N. 9-Apr-2010 In : Science. 328, 5975, p. 208-213.
Download a summary slide of this case study.
