On Agents and Architectures
Note: This was originally written in response to an assignment for my course on the Collective Dynamics of Complex Systems at Binghamton University.
Recent conversations have challenged my thinking about the study of complex adaptive systems (CASs) around a central question—where should we focus our attention when studying them?
Is it more important to understand the behavior of the individual acting agents within systems and the complex dynamics that emerge from their interactions?
Or, should we focus on the architecture of systems and seek detailed understanding of their internal structure which constrains and exerts influence on the agents?
This week’s readings have provided valuable historical context and philosophical perspectives that are helping me develop more useful mental models around this question.
Complexity = Hierarchy?
In The Architecture of Complexity (1962), Herbert Simon argues that one of the main reasons we are able to see, describe, and understand most complex systems in the first place is because of their “nearly decomposable” hierarchic structure.1 He even speculates that humans may be fundamentally incapable of understanding non-hierarchical complex systems due to cognitive limitations that prevent us from calculating all relevant interactions.
The methodology for deep systems analysis which embraces a hierarchical and recursive approach to decomposing systems of interest that I’m applying to the study of blockchain systems is directly inspired by Simon. It has proven invaluable in deepening my understanding of the nature of the interactions between their technical, social, and economic components.
The Limits of Hierarchical Thinking
A decade later, in The Organization of Complex Systems (1973), Simon contrasts two visions of science with differing approaches to explanation.2 Laplacian science “seeks to formulate a single set of equations describing behavior at the most microscopic, the most fundamental level, from which all macro-phenomena are to follow and to be deduced.”
In contrast, Mendelian science “seeks to formulate laws that express the invariant relations between successive levels of hierarchic structures.” Since nature is only nearly, not fully decomposable, if we embrace the second view we are left to accept the fact that “many of the most beautiful regularities of nature will only be approximate regularities and will fall short of exactness.”
For me, this hints at the limits of deep structural understanding and speaks to the need for bottom up agent-oriented approaches that are uniquely enabled by modern computation and allow us to reason about complex behavior which emerges from the micro level.
Pros and Cons of Complexity
Finally, in Complexity and Philosophy (2006) Cilliers, Heylighen and Gershenson highlight how the different historical approaches of systems science and cybernetics are gradually being integrated under the banner of complexity science.3 The central paradigm of complexity science “is the multi-agent system: a collection of autonomous components whose local interactions give rise to a global order."
However, the authors crucially note that there hasn’t been much success in productively applying complexity theory to the social sciences. They speculate this might be a result of the fact that “many social theorists were introduced to complexity via the work done by 'hard' complexity scientists.” This work is strongly informed by chaos theory and contains strong elements of reductionism.
I feel like this hints at the exact issue that I, with my interdisciplinary degree in the social sciences, ran into after I took my first introductory course in agent-based modeling through the Santa Fe Institute in 2019. The toy models of birds flocking, forest fires, and the dynamics of segregation were fun and thought provoking. But I failed to see how they could be usefully applied towards a better understanding of the real world systems that I wanted to study like blockchain networks, or The Federal Reserve. This is what led me to seek out and place my focus on the methods of systems science that aim for deep architectural understanding.
Synthesis?
So, returning to the question of agents vs. architectures, I appreciate how this week’s readings support my growing intuition that there is a need for integration.
Many of the complex adaptive systems we are interested in—like the human body, a smart home, or a branch of a national government, can be simultaneously viewed as agents or as structures. It’s a matter of perspective.
The Handbook of Applied Systems Science identifies three core methods in systems science - bottom-up ABMs, top-down system dynamics, and middle-out network analysis.4
I wonder, how can we develop tools that let us efficiently combine insights from each approach to better understand the interplay between agents and architectures?
Simon, H. A. (1962). The Architecture of Complexity | SpringerLink. https://link.springer.com/chapter/10.1007/978-1-4899-0718-9_31
Simon, H. A. (1977). The Organization of Complex Systems. In H. A. Simon (Ed.), Models of Discovery: And Other Topics in the Methods of Science (pp. 245–261). Springer Netherlands. https://doi.org/10.1007/978-94-010-9521-1_14
Heylighen, F., Cilliers, P., & Gershenson, C. (2006). Complexity and Philosophy (arXiv:cs/0604072). arXiv. https://doi.org/10.48550/arXiv.cs/0604072
Neal, Z. (2016). Handbook of Applied System Science. Routledge & CRC Press. https://www.routledge.com/Handbook-of-Applied-System-Science/Neal/p/book/9780415843348
I always fall back on a systems processes view. More tools and more perspectives than just ABM, top-down dynamics, and network analysis (which, granted, can cover a lot).
Systems/agents interact to organize as more complex systems/architectures.
Systems/architectures emerge from the same systems processes--systems/agents organizing into networks, forming hierarchies, circulating information/material/energy, in states with state transitions, adapting and co-evolving with their environments, all following systemic life cycles.
If you know how healthy, functioning systems work generally, then you can identify when they go wrong--pathologies in systems. Bounded hierarchies/modules that prevent the circulation of information, material, and energy through the system. Difficulties in individuals' capacities to organize with others to improve their work and the system. Inadequate input, problems with stocks and flows. Too much or too little growth and not enough balancing. Inability to evolve internally, to co-evolve with the environment. Etc.