Relevant notes and citations provided to TED by Eric Berlow.
Sarah Harper-Smith et al., "Communicating ecology through food webs: Visualizing and quantifying the effects of stocking alpine lakes with trout," in Dynamic Food Webs: Multispecies Assemblages, Ecosystem Development and Environmental Change, Peter C. de Ruiter et al. (Editors), Elsevier, 2005
Artist Jennifer Parks drew this illustration with food web data synthesized by Sarah Harper-Smith. The key source of the data and expert knowledge for the food web was Roland Knapp:
Roland A. Knapp et al. "Resistance and resilience of alpine lake fauna to fish introductions," Ecological Monographs, August 2001
By far the most common inquiry I get about this talk is, “What was the tool you used to create these visualizations?” The network viz tool I used here is called Network3D, which was developed by my friend, colleague and collaborator, Rich Williams. He originally created this in collaboration with the group at FoodWebs.org. At the time I gave this talk, Rich was the founding director of Microsoft Research’s Computational Ecology and Environmental Science Group in Cambridge, UK. Network3D is a prototype desktop application. Given all the inquiries, Rich and I have recently teamed up with two other TED Fellows — artist/designer David Gurman and computer scientist Kaustuv DeBiswas — to create a more powerful and more accessible cloud-based network visualization tool called Mappr. It is currently in beta mode with a first public release planned for spring 2015.
Roland A. Knapp, "Effects of nonnative fish and habitat characteristics on lentic herpetofauna in Yosemite National Park, USA," Biological Conservation, 2005
Eric Berlow et al., "A network extension of species occupancy models in a patchy environment applied to the Yosemite toad (Anaxyrus canorus)," PLOS ONE, August 12, 2013
The "meadow network" in Yosemite was used as a basis for this paper on predicting Yosemite toad breeding hotspots in the park.
Ulrich Brose et al., "Scaling up keystone effects from simple to complex ecological networks," Ecology Letters, 2005
Eric Berlow et al., "Simple prediction of interaction strengths in complex food webs," PNAS, January 6, 2009
In these two papers we simulated species population dynamics in complex food webs and discovered that you can predict a lot about the abundance of any one species from very simple, local information about its network neighborhood. When we tried to simplify the system and just simulate the dynamics of a few species, the results were much more variable than when that same set of species was embedded in a complex network. This "localization of influence" appears to result from the fact that propagation of change through the web dampens quickly with distance from the source. This is sometimes referred to as "correlation decay." These two papers by Baruch Barzel and Albert-László Barabási suggest that this phenomenon may apply to a wide variety of complex dynamical systems.
Many people have confused this part of the talk with a systems-dynamics analysis of leverage points. I never simulated dynamics of this system because the only information we have is the causal structure. There is no information about the dynamic functional relationships. Since this is a very common challenge, my hope was to explore whether we could glean some coarse insights about network dynamics from structure alone by applying new advances in network science (see the above footnotes for work on correlation decay and localization of influence).
I compiled the network structure data directly from the image I show in the talk, the one published in The New York Times, and from a more complete PowerPoint presentation created by PA Consulting Group for the US military. This report drew heavily on General David Petraeus’ 2006 Counterinsurgency Field Manual. I wanted to explore whether our research on "localization of influence" in complex systems could be applied here. Since there were clear objectives (increase popular support for the Afghan government), I distilled the potentially most important sphere of influence to nodes that were less than or equal to three degrees away.
The two broad categories of actionable, nonviolent issues that fell in this local network neighborhood are clearly still big challenges, but that is all we have to go with from the diagram itself. It was not my attempt to pretend I know anything more about the system than was explicitly laid out in this diagram by others.
A critical point I want to make here is that simplicity doesn’t just come from taking more and more details into account — for example, in order to make a more detailed dynamical model. We found in our research that, to our surprise, as you add more nodes to a persistent complex dynamical system (in our case, a food web), prediction gets simpler as you can ignore more and more. But interestingly, you cannot do that just by simplifying the system to what you think is important or a priority. The simplicity of prediction is a property that actually emerges from the full complex system itself. In other words, we find simplicity from the architecture of complexity itself.