The scaling of goals via homeostasis: an evolutionary simulation, experiment and analysis

  • 2022-11-15 21:48:44
  • Leo Pio-Lopez, Johanna Bischof, Jennifer V. LaPalme, Michael Levin
  • 34


All cognitive agents are composite beings. Specifically, complex livingagents consist of cells, which are themselves competent sub-agents navigatingphysiological and metabolic spaces. Behavior science, evolutionarydevelopmental biology, and the field of machine intelligence all seek an answerto the scaling of biological cognition: what evolutionary dynamics enableindividual cells to integrate their activities to result in the emergence of anovel, higher-level intelligence that has goals and competencies that belong toit and not to its parts? Here, we report the results of simulations based onthe TAME framework, which proposes that evolution pivoted the collectiveintelligence of cells during morphogenesis of the body into traditionalbehavioral intelligence by scaling up the goal states at the center ofhomeostatic processes. We tested the hypothesis that a minimal evolutionaryframework is sufficient for small, low-level setpoints of metabolic homeostasisin cells to scale up into collectives (tissues) which solve a problem inmorphospace: the organization of a body-wide positional information axis (theclassic French Flag problem). We found that these emergent morphogenetic agentsexhibit a number of predicted features, including the use of stress propagationdynamics to achieve its target morphology as well as the ability to recoverfrom perturbation (robustness) and long-term stability (even though neither ofthese was directly selected for). Moreover we observed unexpected behavior ofsudden remodeling long after the system stabilizes. We tested this predictionin a biological system - regenerating planaria - and observed a very similarphenomenon. We propose that this system is a first step toward a quantitativeunderstanding of how evolution scales minimal goal-directed behavior(homeostatic loops) into higher-level problem-solving agents in morphogeneticand other spaces.


Quick Read (beta)

loading the full paper ...