ANNALYSIS Paper No. 1 - The Sociotechnical Structure of AI Alignment

A literature review and applied analysis examining AI alignment as a sociotechnical system. This work synthesizes existing alignment research and applies a series of sociological frameworks with which to understand how institutional structures, incentives, and human coordination shape AI safety outcomes.

R. Ann Medalia

Luhman's Social Systems Theory and Safe AI Alignment

Keywords: systems theory, autopoieses, inner alignment, outer alignment

Author’s Note

Synthesis and analysis were conducted in the course of independent research. The ideas presented herein are not accredited to or sponsored by any institution. I have no conflicts of interest to disclose. 

Systems Approach

Niklas Luhmann was a German sociologist (1927–1998) best known for his prominent contributions to the theory of social systems (Best, 2003). Luhmann’s social systems theory, hereafter referred to as ‘sytems theory,’ is a sociological framework manifested from the work of theorists from multiple fields of research. Systems theory as used by Luhmann is classified under the broader systems approach philosophy, which describes an antecedently unprecedented sociological school of thought, centering  interdependent systems (Montuori, 2011) rather than individuals.

Systems approach includes many interdisciplinary frameworks and concepts, including cybernetics, chaos theory, complexity theory, and general systems theory (Montuori, 2011). The interdisciplinary nature of systems theory is critical both to the function of the theory itself, and in its implications for safe artificial intelligence (AI) alignment.

Defining a System 

Put simply, a system is something whose state changes over time according to [any] rules. The key qualifiers are that a system has a state, that it evolves over time, and that its evolution follows mathematical rules. Systems, in systems approach, are structures, collectives, or groups in which variables change over time. 

Although the term ‘systems’ is used by sociologists, the behavior of a system can be described mathematically through equations that demonstrate the system’s trajectory, revealing pattern movement toward a particular long-term state (Kelso, 2001). This mathematical approach is best described in the words of J.A.S. Kelso: 

“Fundamentally, a dynamical system pertains to anything—such as the behavior of a human being or a social group—that evolves over time. Mathematically, if X1, X2, …, Xn are the variables characterizing the system's behavior, a dynamical system is a system of equations stipulating the temporal evolution of X→, the vector of all permissible X-values. Therein lies the rub: in the social and behavioral sciences one has to find these Xs and identify their dynamics on a chosen level of description.” (Kelso, 2001)

Within Luhmann’s systems theory, examples of systems include law, economics, politics, and science, each of which is controlled by its own internal precedent and coded communication. 

Cybernetics 

The concept of cybernetics was developed during World War II in response to the rising demand for more sophisticated technology (Straussfogel, von Schilling, 2009). The term ‘cybernetics’ was coined by Norbert Wiener (1894–1964), a renowned American mathematician, philosopher, and MIT professor (Tulchinsky, Varavikova, 2014) who was the first to synthesize the work of several key scholars working on the concept of cybernetics. Wiener’s term refers to subgroups, ‘systems’, that are dependent on each other to function; it is based on the biological processes of feedback loops, conscious behvior, and communication between living organisms, (Straussfogel, von Schilling, 2009). 

Of course, the term also contributed as a useful tool with which to examine and explain mechanistic and artificial systems, where the behavior of each unit is compartmentalized, (Tulchinsky, Varavikova, 2014). The mechanistic systems in turn produced rapid, exponential contributions to the study of cybernetics, but the developing field of cybernetics would go beyond machines. The concept of control systems (biological, mechanical), which monitor output and adjust based on feedback, are central to the concept of cybernetics. Wiener said about biological control systems: 

“...communication and control are the most important factors that allow organisms to maintain homeostasis [stability] within a greater tide of growing entropy in the universe. The key mechanism of homeostasis is ‘negative feedback’, whereby a system utilizes information from the environment to limit the effects of change. Positive feedback mechanisms, by contract, are the processing of environmental information as a catalyst for change.” (Straussfogel, von Schilling, 2009)

Wiener, like many sociologists who would later study general systems theory, placed a particular emphasis on feedback loops within control systems. 

Cybernetics also spans into epistemology (influencing human cognition and knowledge itself) (Straussfogel, von Schilling, 2009). This expansion, in particular, has strong implications for the development practices in machine learning. 

Complexity Theory 

Complexity theory, another framework under the umbrella of systems approach, examines how open systems behave when growing internal fluctuations push them past a critical point, triggering self-organization, adaptation, and the emergence of new patterns and relationships among their components (Etkin, 2016). Central to complexity theory is the emphasis on the unpredictability of complex systems, rather than treating systems as strictly deterministic (Etkin, 2016). The theory has particularly strong implications in risk management processes...

...

Citations 

Best, S. (2003). Functionalist perspectives: theorising systems and structures. In Functionalist perspectives: Theorising systems and structures (pp. 15–46). SAGE Publications Ltd. https://doi.org/10.4135/9781446219621.n2

Etkin, David. (2016). Disaster Theory, An Interdisciplinary Approach to Concepts and Causes. Butterworth-Heinemann.  https://www.sciencedirect.com/book/monograph/9780128002278/disaster-theory

Kelso, J.A.S., Editor(s): Smelser, Neil J., Baltes, Paul B. (2001). Self-organizing Dynamical Systems. (pp. 13844-13850). International Encyclopedia of the Social & Behavioral Sciences, Pergamon. https://www.sciencedirect.com/science/chapter/referencework/abs/pii/B0080430767005684

Luhmann, N. (1995). Social Systems. Stanford University Press. https://www.sup.org/books/sociology/social-systems

Luhmann, N. (2012). Introduction to Systems Theory. Polity. https://www.politybooks.com/bookdetail?book_slug=introduction-to-systems-theory--9780745645711

Montuori, A. (2011). Encyclopedia of Creativity (Second Edition) - Systems Approach (pp. 414–421). Academic Press. https://www.sciencedirect.com/science/chapter/referencework/abs/pii/B9780123750389002120?via%3Dihub

Straussfogel, D., von Schilling, C. (2009). International Encyclopsedia of Human Geography - Systems Theory (pp. 151–158). Elsevier. https://www.sciencedirect.com/science/article/pii/B9780080449104007549

Tulchinsky, Theodore H., Varavikova, Elena A. (2014). The New Public Health (Third Edition) - Planning and Managing Health Systems (pp.613–641). Elsevier. https://www.sciencedirect.com/book/monograph/9780124157668/the-new-public-health