Active Inference, Evolutionary Transition, and the Deep Roots of Complex Societies
Total Page:16
File Type:pdf, Size:1020Kb
Social morphogenesis as enactive agency Active inference, evolutionary transition, and the deep roots of complex societies Avel Gu´enin{Carlut avelguenin.github.io May 21, 2021 Kairos Research kairos-research.org Table of Contents Introducing active inference to cultural evolutionists Clarifying active inference Emerging agency as active inference 1 Introducing active inference to cultural evolutionists The Free Energy Principle [17, 11, 18, 1] Dynamical systems, given sufficiently regular boundary conditions, spontaneously self-organise so as to minimise their expected variational free energy, effectively maximising evidence for the world-model they enact. (En)active inference is the process by which dynamical systems climb evidence gradient, ie autonomously enact adaptive agency [31]. from Friston [18] Dynamics as inference 2 (En)active inference is the process by which dynamical systems climb evidence gradient, ie autonomously enact adaptive agency [31]. from Friston [18] Dynamics as inference The Free Energy Principle [17, 11, 18, 1] Dynamical systems, given sufficiently regular boundary conditions, spontaneously self-organise so as to minimise their expected variational free energy, effectively maximising evidence for the world-model they enact. 2 Dynamics as inference The Free Energy Principle [17, 11, 18, 1] Dynamical systems, given sufficiently regular boundary conditions, spontaneously self-organise so as to minimise their expected variational free energy, effectively maximising evidence for the world-model they enact. (En)active inference is the process by which dynamical systems climb evidence gradient, ie autonomously enact adaptive agency [31]. from Friston [18] 2 Active inference can work by altering one's environment to fit their expectations, which corresponds to the evolutionary process known as niche construction [8, 13]. from Veissi`ereet al. [39] Humans in particular build shared expectations through engagement with common social and material affordances, allowing adaptative cultural niche construction [23, 6] by thinking through other minds [39, 40]. Cultural niche construction as active inference 3 Humans in particular build shared expectations through engagement with common social and material affordances, allowing adaptative cultural niche construction [23, 6] by thinking through other minds [39, 40]. Cultural niche construction as active inference Active inference can work by altering one's environment to fit their expectations, which corresponds to the evolutionary process known as niche construction [8, 13]. from Veissi`ereet al. [39] 3 Cultural niche construction as active inference Active inference can work by altering one's environment to fit their expectations, which corresponds to the evolutionary process known as niche construction [8, 13]. from Veissi`ereet al. [39] Humans in particular build shared expectations through engagement with common social and material affordances, allowing adaptative cultural niche construction [23, 6] by thinking through other minds [39, 40]. 3 • Occam's razor : Is Active Inference effectively predictive of new facts about human cognition ? [7, 14, 42] • The Optimisation problem : If Active Inference is verified, how is everything not always adaptive ? [9, 10, 16] • The Enactive objective : Can Active Inference account for non-representational engagement with a landscape of affordances ? [2, 4, 5, 16, 22, 26] Common criticism to Active Inference 4 • The Optimisation problem : If Active Inference is verified, how is everything not always adaptive ? [9, 10, 16] • The Enactive objective : Can Active Inference account for non-representational engagement with a landscape of affordances ? [2, 4, 5, 16, 22, 26] Common criticism to Active Inference • Occam's razor : Is Active Inference effectively predictive of new facts about human cognition ? [7, 14, 42] 4 • The Enactive objective : Can Active Inference account for non-representational engagement with a landscape of affordances ? [2, 4, 5, 16, 22, 26] Common criticism to Active Inference • Occam's razor : Is Active Inference effectively predictive of new facts about human cognition ? [7, 14, 42] • The Optimisation problem : If Active Inference is verified, how is everything not always adaptive ? [9, 10, 16] 4 Common criticism to Active Inference • Occam's razor : Is Active Inference effectively predictive of new facts about human cognition ? [7, 14, 42] • The Optimisation problem : If Active Inference is verified, how is everything not always adaptive ? [9, 10, 16] • The Enactive objective : Can Active Inference account for non-representational engagement with a landscape of affordances ? [2, 4, 5, 16, 22, 26] 4 Clarifying active inference The Active Inference framework does not add anything to the explanation of how cognition or culture works. It provides physical grounding to the study of cognition [11] by explaining why dynamical systems must self-organise toward adaptive coupling with their environment. Self-organised adaptive coupling Explanation and (scientific) prediction 5 It provides physical grounding to the study of cognition [11] by explaining why dynamical systems must self-organise toward adaptive coupling with their environment. Self-organised adaptive coupling Explanation and (scientific) prediction The Active Inference framework does not add anything to the explanation of how cognition or culture works. 5 Explanation and (scientific) prediction The Active Inference framework does not add anything to the explanation of how cognition or culture works. It provides physical grounding to the study of cognition [11] by explaining why dynamical systems must self-organise toward adaptive coupling with their environment. Self-organised adaptive coupling 5 The Active Inference framework does not entail that living systems manage to find an optimum in free-energy, making them always adaptive. It entails that existing living systems have historically integrated information helping them gravitate around the stablest attractors they could actually explore [12]. Only historyless systems need reach an optimum. Optimality and history 6 It entails that existing living systems have historically integrated information helping them gravitate around the stablest attractors they could actually explore [12]. Only historyless systems need reach an optimum. Optimality and history The Active Inference framework does not entail that living systems manage to find an optimum in free-energy, making them always adaptive. 6 Optimality and history The Active Inference framework does not entail that living systems manage to find an optimum in free-energy, making them always adaptive. It entails that existing living systems have historically integrated information helping them gravitate around the stablest attractors they could actually explore [12]. Only historyless systems need reach an optimum. 6 Agents under active inference do not engage with their niche by integrating cognitive representations of its features. Predictive knowledge is embodied in the material structure and enacted in the situated activity of an agent [31]. A state-of-the-art statistical model of the woodlice's niche From [? ] (Cognitive) prediction and enaction 7 Predictive knowledge is embodied in the material structure and enacted in the situated activity of an agent [31]. A state-of-the-art statistical model of the woodlice's niche From [? ] (Cognitive) prediction and enaction Agents under active inference do not engage with their niche by integrating cognitive representations of its features. 7 (Cognitive) prediction and enaction Agents under active inference do not engage with their niche by integrating cognitive representations of its features. Predictive knowledge is embodied in the material structure and enacted in the situated activity of an agent [31]. A state-of-the-art statistical model of the woodlice's niche From [? ] 7 Emerging agency as active inference The dense, hierarchical settlements known as City-States developped as the result of long-distance trade [37] and taxation [34], accelerated by the later development of bookkeeping [19] and other institutions [38]. Unknown source City-States as collective organisms The coevolution of functional specialisation of individual [41], writing as a new information system [21], and centralised administration [35] make the evolution of City-States a clear case of evolutionary transition toward collective organism. The City-State complex 8 City-States as collective organisms The coevolution of functional specialisation of individual [41], writing as a new information system [21], and centralised administration [35] make the evolution of City-States a clear case of evolutionary transition toward collective organism. The City-State complex The dense, hierarchical settlements known as City-States developped as the result of long-distance trade [37] and taxation [34], accelerated by the later development of bookkeeping [19] and other institutions [38]. Unknown source 8 The City-State complex The dense, hierarchical settlements known as City-States developped as the result of long-distance trade [37] and taxation [34], accelerated by the later development of bookkeeping [19] and other institutions [38]. Unknown source City-States as collective organisms The coevolution of functional specialisation of individual [41], writing as a new information system [21], and centralised administration [35] make the evolution of City-States a clear case of evolutionary transition toward collective organism. 8 The emergence of collective organisms is precedented