For most problems in biology, physiology, artificial life, and medicine, it is not feasible to model an entire emergent agent at the level of its underlying components. We circumvent this by declaring the agent's very existence a feature of the model and focusing on some subset of its overall dynamics. Since this strategy reduces the organism's existence to a starting assumption, models that incorporate the possibility of death must also declare the structure of a life-death boundary, an extrinsically defined viability constraint [1]. With these equations in hand, we can begin to ask which configurations of an agent and its environment will lead to survival and which will lead to death. Unfortunately, these models are often highly nonlinear, and the typical structures of dynamical systems theory (invariant sets and their corresponding manifolds) are insufficient for constructing a skeleton of the system's possible paths and survival outcomes. 

Viability Space Decomposition is a framework designed to analyze systems of nonlinear ordinary differential equations with extrinsic viability constraints, building on top of the existing foundation of dynamical systems. Specifically, we demonstrate that it is possible to derive new global manifolds, which provide a global map of the possible paths the system can take when paired with the classical invariant sets and their manifolds. This method has been demonstrated for models of single isolated agents and multiple agents in interaction [2,3]. Importantly, these structures can only be derived if one views the essential physiological variables, behavioral elements, and environment as a unified dynamical system. Currently, work is being done to characterize new bifurcations in this framework, demonstrate the math in higher dimensional systems, and extend it to models with more detailed spatial relationships between agents.

References

[1] W.R. Ashby, "Design for a brain," 2nd Ed., Springer, 1960.

[2] C. McShaffrey and R. D. Beer, “Decomposing viability space,” in ALIFE 2023: Ghost in the Machine: Proceedings of the 2023 Artificial Life Conference, MIT Press, 2023.  Article Link.

[3] C. McShaffrey and R. D. Beer, “Dissecting viability in multi-agent systems,” in ALIFE 2024: Proceedings of the 2024 Artificial Life Conference, MIT Press, 2024.  Article Link.

Unlike the extrinsic viability constraints that we use in most of our models, real organisms do not have their physiological limits imposed by some outside observer. Instead, their viability constraints are intrinsically generated and conditional on the underlying network of recurrent processes that continuously regenerate them. While we cannot build a complete model of an emergent cell or multicellular organism, we can still make progress on a theory of intrinsic viability by using idealized models of metastable systems. Building on Randall Beer's characterization of autopoiesis in Conway's Game of Life [1,2,3] we recently demonstrated that it is possible to derive the intrinsic viability constraint of a glider from first principles [4]. Work has also been done to generalize these findings to other cellular automata that approximate Euclidean space [5]. I am interested in both further exploring what a minimal cellular automata like the Game of Life has to say about intrinsic viability, and scaling up to more physically realistic models. 

References

[1] R.D. Beer, "The cognitive domain of a glider in the Game of Life," in Artificial Life, 2015.

[2] R.D. Beer, "Characterizing autopoiesis in the Game of Life," in Artificial Life, 2015.

[3] R. D. Beer, “An integrated perspective on the constitutive and interactive dimensions of autonomy,” in ALIFE 2020: Proceedings of the 2020 Artificial Life Conference, MIT Press, 2020.  Article Link.

[4] R. D. Beer, C. McShaffrey, and T. M. Gaul, “Deriving the intrinsic viability constraint of an emergent individual from first principles,” in ALIFE 2024: Proceedings of the 2024 Artificial Life Conference, MIT Press, 2024.  Article Link.

[5] T. M. Gaul, “Autopoiesis in RealLife Euclidean automata,” in ALIFE 2024: Proceedings of the 2024 Artificial Life Conference, MIT Press, 2024.  Article Link.

When I studied cognitive science at Vassar, I was told that the goal of the field was, "to study minds, wherever we may find them." Unlike the traditional neuroscientific perspectives that I had been exposed to previously, this was much more liberating. The professors encouraged us to think critically and pluralistically about the various definitions of cognition and how they apply to natural and artificial systems alike. I began to consider cognitive science as broadly being the interdisciplinary study of autonomous systems behaving in their environment, and got hands-on experience working with both robotics and live organisms. Through both my coursework and research, I began to view cognitive science and biology as strongly overlapping fields. 

 Towards the end of my undergraduate education, I read Maturana and Varela's Autopoiesis and Cognition: The Realization of the Living [1]. This work, although almost entirely verbal in its formulation, convinced me that it should be possible to construct a formal theory connecting the constitution of living systems with their cognitive dynamics. This inspired me to do my graduate studies with Randall Beer, who has been advancing formal theories of agent-environment interaction as well as autopoiesis [2,3]. 

 More recently, other sources have come forth proposing the fruitful link between cognitive science and biology from the study of basal cognition [4] to the manipulation of agential materials or purposeful matter in bioengineering [5,6]. 

I believe that blurring the line between biology and cognitive science has the potential to fully transform how we understand the behavior and physiological limits of living agents. 

References

TO BE INCLUDED SOON

[1]