25 Sep 2008
John H. Holland about CAS
John H. Holland has written a new article in “The Scientist” about CAS. In this article named Biology’s Gift to a Complex World, he talks about his work and how he met the people from the SFI: Murray Gell-Mann, George Cowan, Brian Arthur, Doyne Farmer and Norm Packard. He says about CAS (by the way these are the CAS I had in mind when I founded the CAS-group, site and blog, CAS does not stand for central authentication service):
I’ve applied genetic algorithms to the area I know and like best: complex adaptive systems. These are systems that have multiple levels of co-evolution, like ecosystems, which through their complex interactions, become more than the sum of their parts. One example is the immune system, a conglomerate of cells, cytokines and proteins, containing many levels of interaction and adaptation which provide the power to protect a human from the majority of pathogens encountered. To be able to capture the essence of these complex adaptive systems is to be able to predict their most powerful abilities.
The complex systems that I was interested in were those whose parts not only interacted to create novel properties, but also co-evolved and adapted new rules to weather the fluctuations in their environment. I called them ‘complex adaptive systems’ or CAS
And about the common elements and emergent properties of CAS:
At the outset, it may not seem that complex adaptive systems such as the immune system, the economic function of a city, and tropical rainforests could have much in common. But all CAS involve a diverse array of agents that adapt and learn from their interactions with one another. An individual’s immune system is built upon an array of cells, proteins, and chemical factors that are constantly in flux. This vast system of interacting elements responds to all manner of pathogenic threats, protects the body from too much inflammation and even records to memory its fighting strategy so it can be efficient the next time it encounters the same pathogen. The buyers and sellers that comprise the economic infrastructure of a city and the diverse species in a rainforest act in similar ways. The key is to categorize the similarities across them in order to develop a working theory of all complex adaptive systems
In looking for common elements in these complex adaptive systems, we see that all CAS exhibit lever points – points at which a small effort can produce a desired, directed effect. In the immune system, a vaccine would be an example of a lever point. From an engineering perspective, a vaccine is a solution that, with the relatively small effort of inactivating a virus or bacterium, produces a large effect: protecting an entire population from that pathogen. A virologist might design a vacine by searching for the most immunogenic portion of the pathogen. From a CAS perspective, the solution or lever point could be something much different: perhaps a combination of cytokines or the timing of administration that provides critical advantage. Just as genetic algorithms find solutions that are messier but more effective than a human would design, so a theory of CAS would find solutions that might not be obvious to the virologist. A theory of CAS would eliminate the trial and error. On a broader scale, a theory of CAS would let us transfer well-known results about one CAS to other CAS where data or understanding is lacking
About a theory of CAS:
In order to build that theorem, we are characterizing the agents or parts of a system in a way that is as general as possible, and then using genetic algorithms to simulate the learning between our modeled parts. To construct CAS agents we must first characterize the building blocks that will represent the agent in our theorem, as well as the rules that govern their interaction or combination. With our model of economic systems, we saw how genetic algorithms could recombine rules to form new methods of interacting. Our model of CAS would not only recombine the building blocks of the system, but also the rules by which they interact.
My strategy for finding a theorem that can predicts lever points is to study how novel properties emerge from aggregates of building blocks. For example, the aggregate of immune cells – the proteins and cytokines – protects a human from pathogenic invasion. But this conglomerate of cells also confers immunologic identity between individuals, a so-called emergent property of that system. A good theory of CAS would be able to predict the emergent properties, such as immune identity, which would show us where to look for the lever points of a system.