John H. Miller is Professor of Economics and Social Science at Carnegie Mellon University. In his book A Crude Look at the Whole, he examined how business, systems, and life intertwine regarding their algorithms and ways of functioning. He discusses economic crashes, slime mold behaviors, adaptive system choices, and more.
Here is my interview with Professor John H. Miller about topics in his book:
—
Armen: In your 2004 paper with Scott E. Page, you mention that the Standing Ovation Problem “… belongs to a rich class of problems — decentralized dynamical systems consisting of spatially distributed agents who respond to local information.” This theme is carried over in your book A Crude Look At The Whole, as there are various examples of this presented, most memorably with bees “deciding” where to relocate their swarm to. I am fond of seeing and deconstructing systems in this way. Would you say that any systems which don’t follow this decentralized form have an artificial centralized form in place?
Professor John H. Miller: Complex systems arise when the (often simple) behavior of the individual parts aggregate in surprising ways. System like bee hives (and, hopefully, political primaries) illustrate how such aggregation can lead to effective, decentralized decision making. I think the interesting question about centralized control is whether it is really necessary most of the time. There are plenty of examples, markets and bee hives to name but two, where giving up centralized control and relying on decentralized, bottom-up systems, can lead to very efficient and productive outcomes.
Giving up centralized control is hard, and on occasion decentralized
systems will go terribly wrong (markets crash, bee colonies fail to find
a new hive). Of course, centralized control can go terribly wrong as
well.
Armen: I have always looked at communication as having the root of being agenda-driven, and the computing experiment presented in the book “…suggests that the emergence of communication could be a key path to cooperation in social systems and ultimately to survival.” Would this mean that organisms with all their physiological needs met would have no need for communication?
Professor John H. Miller: Given that most systems involving interacting agents, communication in one form or another is always present. Communication, provides new opportunities for agent interaction, and thus it has the possibility of really extend the fitness of a system. Even if an agent’s base physiological needs are met, the ability to productively communicate can allow that agent to do even better.
In some computational experiments, we found that communication is most
valuable early on when the agents in a system are trying to establish
coordination or cooperation. Once this happens, maintaining the good
outcome requires much less communication. Thus, in these systems
communication is key, though often transient — -when it is working well,
it may not appear to be that important, even though it is.
Armen: You had mentioned that “… both slime molds and bacteria, without a neuron to be had between them, can make productive decisions.” In relation to this, and the breakdown of our brains into decentralized feedback response centers, what are your thoughts on the timetable for deconstructing or mimicking our neuronal networks/pathways on a level of parity with how we process the world?
Professor John H. Miller: The example of bacteria make it clear that brains, as we traditionally think of them, are overrated. Neurons are wonderful when you need to send a message over long distances and it “absolutely, positively has to be there,” but on small scales, random interactions of molecules (think signals) are more than sufficient to allow thoughtful behavior.
Figuring out the entire connectome of the brain might potentially give us some new insights into thinking, though this reductionist approach to science alone will be insufficient — -we know from the study of complex systems that reducing a system to its parts does not help us understand what happens when those parts interact. Reductionism does not equal constructionism, and so it was not a great surprise that even when we decoded the entire human genome, we didn’t have a quantum leap in our understanding of, say, human disease. I suspect the same will be true for the connectome. To really understand how our thinking brain works, we may well need to study simpler, whole systems like chemotaxis in bacteria or hive seeking in bee colonies.
Armen: You may have the best form of humor I have seen in scientific papers and your book, such as describing the bee coming back to the swarm dancing lively like a crazy motivational speaker running through, or your celebrity/academic audience/visibility inverse comparison in your standing ovations paper. It is very subtle, and used sparingly. What forms of humor do you appreciate, and do you find the world funny as it currently is?
Professor John H. Miller: I’m often surprised at the lack of humor in academic work — -it does seem that academics often take themselves and their work far too seriously. Humor can be a very powerful and, occasionally, deeply intellectual phenomenon. Perhaps there is even a complex systems aspect to it — -emergence is when the parts of a system result in a surprising whole, and I could see where depending on the relationship between the parts and the emerging whole, our response can vary between awe and laughter.
As for me, I (perhaps to the dismay of my friends) appreciate all kinds
of humor, and I’m often amused, for better or worse, by the world around me.
Armen: Can Markov chain Monte Carlo and Bayesian statistical methods be used to outdo brute force methods in efficiency in every category, using lowest-energy probabilities to represent what people would eventually figure to be the best way to do or process something? Are there any categories of processes we do that can’t be statistically optimized with this form of computing?
Professor John H. Miller: Markov chain Monte Carlo and Bayesian statistical ideas are, when appropriately applied, incredibly powerful and even “magical” in what they can do. Nonetheless, they are tools, and it is rare to find a tool that works best “in every category.” Indeed, the “No Free Lunch” theorem that comes out of complex systems suggests that there is no best optimization algorithm independent of the problem at hand, and a similar idea would apply here as well.
—
Appreciations to Professor John H. Miller for taking part.