Self-maintaining Patterns of Behaviour
I’ve recently published a couple of papers that may be of interest to people in this community.
Partially inspired by conversations at the (fantastic) eSMCs summer school, in San Sebastian, 2011, both papers present analysis of a newly developed computational model which is used to investigate how the organisational property of precarious self-maintenance that is typically associated with life could also underlie habits, seen as precarious self-maintaining patterns of sensorimotor behavior.
The principle underlying the model is simple; a robot is controlled by a system that reinforces visited sensorimotor trajectories, so that the robot is more likely to later re-enact those sensorimotor trajectories. Like a self-maintaining cell, certain patterns of sensorimotor behavior, habits, can also be precarious; perpetually decaying, yet sometimes counteracting that decay through processes of self-maintenance.
The first paper was published about a week ago in Frontiers in Human Neuroscience. It presents the latest version of this “habit-based controller” in detail, including (i) how it can be trained to perform certain behaviors (ii) an explanation of what habits are in the model and (iii) a first exploration of what kind of self-maintaining behaviors emerge when the controller is randomly initialized. This paper was a collaboration with Xabier Barandiaran who, among other contributions, provided a substantial philosophical and historical contextualization of the work.
The second paper, presented earlier this month at ALIFE in NYC, uses the same model, but coupled to a robot with a biological essential variable, “blood sugar,” that needs particular forms of behavior to be regulated within viability limits if it is to be considered healthy. Because the habit-based controller can only reinforce/stabilize patterns of behavior that are repeated, when the state of the essential variable is included as an interoceptive sensed variable, behaviors spontaneously emerge that regulate/stabilize the essential variable. This work relates to Ezequiel’s and Hiro’s work on homeostatic robotics, and provides a demonstration of how the essential variables of a habit can be the same as (or at least tightly intertwined with) the essential variables of the biological organism performing the behavior. To pique your interest, here are also a couple of videos of trials from this paper.
I am always happy to discuss this work — please give me a shout if you have questions or comments. I hope you get a chance to take a look, and I look forward to our next get-together!
All the best,