Competency in Navigating Arbitrary Spaces as an Invariant for Analyzing Cognition in Diverse Embodiments

Abstract

One of the most salient features of life is its capacity to handle novelty and namely to thrive and adapt to new circumstances and changes in both the environment and internal components. An understanding of this capacity is central to several fields: the evolution of form and function, the design of effective strategies for biomedicine, and the creation of novel life forms via chimeric and bioengineering technologies. Here, we review instructive examples of living organisms solving diverse problems and propose competent navigation in arbitrary spaces as an invariant for thinking about the scaling of cognition during evolution. We argue that our innate capacity to recognize agency and intelligence in unfamiliar guises lags far behind our ability to detect it in familiar behavioral contexts. The multi-scale competency of life is essential to adaptive function, potentiating evolution and providing strategies for top-down control (not micromanagement) to address complex disease and injury. We propose an observer-focused viewpoint that is agnostic about scale and implementation, illustrating how evolution pivoted similar strategies to explore and exploit metabolic, transcriptional, morphological, and finally 3D motion spaces. By generalizing the concept of behavior, we gain novel perspectives on evolution, strategies for system-level biomedical interventions, and the construction of bioengineered intelligences. This framework is a first step toward relating to intelligence in highly unfamiliar embodiments, which will be essential for progress in artificial intelligence and regenerative medicine and for thriving in a world increasingly populated by synthetic, bio-robotic, and hybrid beings.

Keywords: anatomical morphospace; basal cognition; physiology.

Figure 1

Multi-scale competency architecture (MCA). (A) The MCA is implemented by biological systems in which every level of organization traverses various spaces toward preferred regions. Subcellular systems (molecular networks) navigate the transcriptional space (B), while collections of cells navigate the anatomical morphospace, such as planarian tissues that can be pushed into regions of the space corresponding to diverse species’ head shapes without genomic editing (C) (image by Alexis Pietak [31]). Higher-order systems distort the energy landscapes for their subsystems (via virtual “objects” in that space) to enable their components’ local homeostatic mechanisms to achieve goals that are adaptive at the higher level systems’ space. This links the intelligence (or competent navigation) of spaces to simple energy minimization dynamics. Panels (A,B) are courtesy of Jeremy Guay of Peregrine Creative.

Figure 2

Diverse spaces within which living systems navigate. Transcriptional space is the space of possible gene expression patterns, taken with permission from [59]. Examples of a transcriptional state space of a two-gene network (mutual inhibition of genes (A,B)) and the associated epigenetic landscape in the two-dimensional state space are shown. The dynamical state of the network maps to a point in the space; changes in gene activity represent walks in the space. (B) Physiological space is the space of possible physiological states, simplified to just three parameters such as intracellular ion concentrations, taken with permission from [70]. Individual cells occupy regions of the space and can move between states by opening and closing specific ion channels. The functional state (region of the space) is a function of all of the parameters and large-scale variables, such as V_mem (resting potential), which refer to numerous microstates composed of individual ion levels. (C) Navigating spaces to thrive despite novel stressors, taken with permission from [71]. Planaria exposed to barium chloride experience head deprogression because barium is a blocker of potassium channels, making it impossible for the neural tissues in the head to maintain a normal physiology. The flatworms soon regenerate a new head which is barium-insensitive. Transcriptomic analysis showed only a handful of genes whose expression was altered in the barium-adapted heads. Because barium is not something planaria encounter in the wild, this example shows the ability of the cells to navigate transcriptional space to identify a set of genes that enable them to resolve a novel physiological stressor. The mechanism by which they rapidly determine which of many thousands of genes should be up- or downregulated in this scenario is not understood. (The cells do not turn over fast enough to allow a hill-climbing search, for example.)

Figure 3

Anatomical morphospace and its navigation by cellular collective intelligence. (A) Example of morphospace—the space of possible shapes—for coiled shells (taken with permission from [87]). Three parameters (rate of increase in the size of the generated shell cross section per revolution, the distance between the cross section and the coiling axis, and the rate of translation of the cross section along the axis per revolution) define a space within which many taxa can be found. (B) Space of possible planarian heads defined by possible values of three morphogen values in a computational model (taken with permission from [88]). (C,D) The idea of morphospace and different species of animals as mathematical transformations of coordinates in those spaces was originally proposed by D’Arcy Thompson (panels taken with permission from [89]). Traversals of morphospace can be seen in regeneration, such as for the salamander limb, which will continue to grow when amputated at any position (brought to a new region of morphospace for the limb) until the system reaches the correct state (the shape of a normal limb), at which point it stops ((E) panel by Jeremy Guay of Peregrine Creative), or in the ability of both normal and scrambled tadpole faces to rearrange until a correct frog craniofacial morphology is reached ((F,G) taken from [90]). (H) Remodeling, de novo embryogenesis, and regeneration are all examples of biological systems’ abilities to navigate from starting positions in morphospace “s1”–“s4” and reach the target morphology goal state “G” while avoiding the local maxima “LM”. Morphospace plasticity (I) includes the ability of higher-level constraints to activate diverse underlying molecular mechanisms as needed. For example, (I) tubulogenesis in the amphibian kidney normally works via cell–cell communication, but when the cells are forced to be very large (by induced polyploidy), this reduces the number of cells and eventually leads to switching to using cytoskeletal bending to form the same diameter of tube from just one cell bending around itself (panel by Jeremy Guay from [91,92]).

Figure 4

Isomorphism between neural bioelectricity and preneural (developmental) bioelectricity. (A) Neural cells compute by forming networks in which each cell can use ion channels to establish a specific resting potential and selectively communicate it to connected neighbors through electrical synapses known as gap junctions. (B) Neural dynamics are actually a speed-optimized variant of a much more ancient system. All cells use ion channels, and most cells form electrical connections with their neighbors. (C) In the brain, DNA-specified ion channel hardware in neurons enables bioelectric computation, a kind of software that can be impacted by experiences (stimuli). By enabling fast communication over long distances, the synaptic architecture depicted in (A) enables the brain to control the physiological dynamics of muscle cells and hence move the body in three dimensions. (D) Prior to the development of specialized, high-speed neurons, preneural bioelectric networks exploited the same architecture of physiological software implemented by the same ion channel hardware. The information-processing and memory features of bioelectrical networks were used to control the movement of the body configuration through the morphospace by controlling cell behaviors [104,137]. Images by Jeremy Guay of Peregrine Creative.

Figure 5

A proposed model in which evolution pivots the same strategies (and some of the same molecular mechanisms) to navigate different spaces. Each level of organization solves problems in its own space, and systems evolved from navigating the metabolic, physiological, transcriptional, anatomical, and finally (when the muscle and nervous systems evolve) 3D space of traditional behavior. Other spaces, such as linguistic space, are possible in more advanced forms. Images by Jeremy Guay of Peregrine Creative.

Figure 6

Similar strategies are seen in diverse biological systems at all levels for navigating problem spaces. One example is spreading out and then pulling back from regions that are non-attractors. (A) Physarum slime molds spreading throughout a maze and then pulling back from every location except the shortest path between two food sources (taken with permission from [144]). (B) Neurons often prune back after forming a set of network connections (taken with permission from [145]). (C) Evolutionary exploration finds high fitness peaks, and then populations pull back from the valleys. Panel (C) by Jeremy Guay of Peregrine Creative.

Figure 7

Schematic specifications of a Markov blanket (MB) comprising sensory (s) and active (a) states that are intermediate between the external or environmental (e) states and the internal (i) states of some system of interest. The MB is a boundary in the joint system-environment state space. It may be partially implemented by a structure (here, a plasma membrane) in a 3D space. The MB states (s and a) can be thought of as an API between the system and its environment. Taken from [157] and used with permission.

Figure 8

Generic two-system interaction mediated by a Markov blanket. (a) Any MB can be considered a boundary B in the joint state space of a system S and its environment E. The physical interaction between S and E, here represented by the Hamiltonian (total energy) operator H_SE, is defined at this boundary. (b) S must obtain free energy from and exhaust waste heat into E. The boundary B must therefore include a thermodynamic sector in addition to the sensory (s) and active (a) sectors. The states of this thermodynamic sector are observationally inaccessible and hence uninformative to S. Taken from [78] with a CC-BY license.

Figure 9

Actions in one space enable (or constrain) actions in another space. These relations function both from the bottom up and the top down in the scale hierarchy. Gene expression, for example, provides the components needed to enable a particular morphology, which in turn enables behaviors that enable the free energy production required to drive further gene expression. Enabling and constraining relations function, in general, both from the bottom up and the top down in the scale hierarchy. Image by Jeremy Guay of Peregrine Creative.