1
CiE 2014: Language, Life, Limits
Modeling of Life as Cognitive Info-Computation
Gordana Dodig-Crnkovic
Abstract.This article presents naturalist approachto cognition understood as network of info-computational autopoietic processes in living systems. It providesconceptualframeworkfor the existing models,relating them with new empirical and theoretical insights into the phenomena of cognition. It elucidates three fundamental questions:what cognition is, how cognition works and what cognition doeson different levels of organization of living matter. By explicating info-computational character of cognition and its evolution its agent-dependency, generative mechanisms we better understand its life-sustaining and life-propagating role. The info-computational approach can contribute to rethinking cognition as process of bio-chemical computation in living beings, from the simplest to the most complex ones. This can both help to improve ourunderstandingof living systems and to further develop cognitive computing.
Introduction
It is a remarkable fact that even after half a century of research in cognitive science, cognition still lacks commonly accepted definition(Lyon, 2005).E.g.Neisser’s textbook description of cognition as“all the processes by which sensory input is transformed, reduced, elaborated, stored, recovered and used” (Neisser, 1967) is so broad that it would ascribe cognition to present day robots. Currentlythe field of cognitive robotics is being developed in hope to learn by construction what cognition might be and then returning to cognitive systems in nature to find what solutions nature have evolved. The process of two-way learning (Rozenberg & Kari, 2008) startsfrom nature by reverse engineeringexisting cognitive agents, simultaneously trying to design cognitive computational artifacts.We have a lot to learn from natural systems about how to computationally model cognition and how to engineer cognitive computers. (Modha et al., 2011)
Until recently only humans were commonly accepted as cognitive agents (anthropogenic approach in Lyon). Some were ready to ascribe certain cognitive capacities to all apes, and some perhaps to all mammals. The lowest level cognition for those with the broadest view of cognition included all organisms with nervous system. Only a few were ready to go below that level. Among those very few, the first ones ready to acknowledge cognitive agency of organisms without nervous system, were Maturana and Varela (Maturana & Varela, 1980; Maturana, 1970)who argued that cognition and life are identical.Lyon’s classification, besides anthropogenic approach, has a biogenic approach based on self-organizing complex systems and autopoiesis. The focus of present paper will be on biogenic cognition.The adoption of the definition of Maturana and Varela is motivated by the wish to provide a theory that would include all living organisms and artificial cognitive agents within the same framework.
The Computing Nature, Computational Naturalism and Minimal Cognition
Naturalism is the view that the nature is the only reality. It describes nature through its structures, processesand relationships between them, relying on scientific approach. Naturalism studies evolution of all natural world,including the life and development of humanity as a part of nature. Computational naturalism (pancomputationalism, naturalist computationalism) is a view that the nature is a huge network of computational processes, whichaccording to physical laws compute (dynamically develop) its own next state from the current one. Representatives of this approach are Zuse, Fredkin, Wolfram, Chaitin and Lloyd who proposed different varieties of computational naturalism. According to the idea of computing nature one can view the time development (dynamics) of physical states in nature as information processing (natural computation). Such processes include self-assembly, self-organization, developmental processes, gene regulation networks, gene assembly, protein-protein interaction networks, biological transport networks, social computing, evolutionary processes and similar processes of morphogenesis (creation of form). The idea of computing nature and the relationships between two basic concepts of information and computation are explored in(Dodig-Crnkovic & Giovagnoli, 2013)(Dodig-Crnkovic & Burgin, 2011).
In the computing nature, cognition should be studied as a natural process. If we adopt the biogenetic approach to cognition, the important question is what is the minimal cognition.Recently numbers of empirical studies have revealed unexpected richness of cognitive behaviors (perception, information processing, memory, decision making) in as simple organisms as bacteria. Single bacteria are too small to be able to sense anything but their immediate environment, and they live too short to be able to memorize significant amount of data, so of interest for us are bacterial colonies, swarms and filmsthat have shown to possess unanticipated complexity of behaviors which can be characterized as “minimal cognition”. (Duijn, Keijzer, & Franken, 2006)(Ben-Jacob, Shapira, & Tauber, 2006; Ben-Jacob, 2008, 2009)(Ng & Bassler, 2009; Waters & Bassler, 2005)
Apart from bacteria and similar organisms without nervous system, even plants are typically taught of as living systems without cognitive capacities.However, plants have been found to possess memory (in their bodily structures that change as a result of past events) and the ability to learn (plasticity, ability to adapt through morphodynamics), anticipate and direct their behavior accordingly. Plantsare argued to possess rudimentary forms of knowledge, according to(Pombo, O., Torres J.M., Symons J., 2012) p. 121 (Rosen 1985) p. 7 and Popper (Popper, 1999) p. 61.
Notice that in the similar way as we ascribea degree of cognition to all living organisms, we might ascribe a degree of consciousness in a sense of self-perception (Goertzel, 1994),andit wouldof course not imply consciousness in the anthropogenic sense. In this article we focus onprimitive cognition as totality of processes of self-generation and self-maintenance that enables organisms to survive and adequately use information from the world. Understanding of cognitionin degrees can help us better understand the step between inanimate and animate matter – from the first autocatalytic chemical reactions to the first autopoietic proto-cells.
Informational Structure of Reality for a Cognitive Agent
When we talk about computing nature, we can ask: what is the “hardware” for this computation? We, as cognitive agents interacting with the universe through information exchange, experience cognitively the universe as information. The informational structural realism (Floridi, 2003)(Sayre, 1976)(Stonier, 1997) is a framework that takes information as the fabric of the universe (for an agent). The physicists Zeilinger (Zeilinger, 2005) and Vedral (Vedral, 2010) suggest that information and reality are one. For a cognizing agent in the informational universe, the dynamical changes of its informational structures make it a huge computational network where computation is understood as information dynamics (information processing). The substrate, the “hardware” is information that defines data-structures on which computation is going on.
Info-computationalism is a synthesis of informational structural realism and natural computationalism (pancomputationalism) - the view that the universe computes its own next state from the previous one (Chaitin, 2007). It builds on two basic complementary concepts: information (structure) and computation (the dynamics of informational structure) as described in (Dodig-Crnkovic, 2011a) and (Dodig-Crnkovic, 2006)(Dodig-Crnkovic, 2014).
The world exists as potential information, corresponding to Kant’s das Ding an sich. Through interactions, this potential information becomes actual information, “a difference that makes a difference” according to (Bateson, 1972).Acognizing agent in a process of interaction uncovers aspects of the world. Shannon describes the process as the conversion of latent information into manifest information (McGonigle & Mastrian, 2012). Even though Bateson’s definition of information as a difference that makes a difference (for an agent) is a widely cited one, there is a more general definition that includes the fact that information is relational and subsumes Bateson’s definition:
Information expresses the fact that a system is in a certain configuration that is correlated to the configuration of another system. Any physical system may contain information about another physical system. (Hewitt, 2007) (Italics added)
Combining Bateson and Hewitt insights, on the basic level, information is a difference in one physical system that makes a difference in another physical system.
When discussing cognition of special interest is the notion ofagent, i.e. a system able to act on its own behalf(Dodig-Crnkovic, 2014). Agency has been explored in biological systems by (Kauffman, 1995)(Kauffman, 1993)(Deacon, 2011). The world as it appears to an agent depends on the type of interaction through which the agent acquires information.. Potential information in the world is obviously much richer than what we observe, containing invisible worlds of molecules, atoms and sub-atomic phenomena, distant cosmological objects and the like. (Dodig-Crnkovic & Müller, 2011) Agents communicate by exchanging messages (information) that helps them coordinate their actions based on the (partial) information they possess.
Information Self-Structuring through Morphological/Physical/Intrinsic Computation and PAC Algorithms
When talking about computational models of biological phenomena, we must emphasize that within info-computational framework computation is defined as information processing. This differs from the traditional Turing machine model of computation thatis analgorithm/effective procedure/recursive function/formal language. The Turing machine is a logical construct, not a physical device(Cooper, 2012).Modeling computing nature adequately, including biological information processing, with its self-generating and learning real-time properties requires new models of computation such as interactive and networked concurrent computation models, as argued in (Dodig-Crnkovic & Giovagnoli, 2013) and (Dodig-Crnkovic, 2011b) with the reference to (Hewitt, 2012) and (Abramsky, 2008).
Computation in general can be described as a self-generating system consisting of anetwork of programs(Goertzel, 1994), a model inspired by self-modifying systems of(Kampis, 1991). In the development of general theory of networked physical information processing, we generalize the idea of computation. Examples of new computing paradigmsinclude natural computing (Rozenberg, Bäck, & Kok, 2012)(MacLennan, 2004)(Nunes de Castro, 2007)(Cardelli, 2009); superrecursive algorithms (Burgin, 2005); interactive computing (Wegner, 1998); actor model (Hewitt, 2012) and similar“second generation” models of computing(Abramsky, 2008).
In the context of novel models of computation Valiant’s ecorythms or algorithms satisfying “Probably Approximately Correct” criteria(PAC) are of special interest as they explicitly model natural systems“learning and prospering in a complex world”. (Valiant, 2013) The difference between PAC learning algorithms and Turing machine model is that the latter does not interact with the environment, and thus does not learn.It has unlimited resources, both space (memory) and time, and even thoughit is sequential, it does not operate in real time. In order to computationally model living nature,we need suitable resource-aware learning algorithms, such as ecorithms, described by Valiant:
“The model of learning they follow, known as the probably approximately correct model, provides a quantitative framework in which designers can evaluate the expertise achieved and the cost of achieving it. These ecorithms are not merely a feature of computers. I argue in this book that such learning mechanisms impose and determine the character of life on Earth. The course of evolution is shaped entirely by organisms interacting with and adapting to their environments.”(Valiant, 2013)p.8
Cognitive capacity in living systems depends on the specific morphology of organisms that enables perception, memory, information processing and agency. As argued in (Dodig-Crnkovic, 2012), morphology is the central idea in the understanding of the connection between computation and information.
The process of mutual evolutionary shaping between an organism and its environment is a result of information self-organization. Here, both the physical environment and the physical body of an agent can be described by their informational structurethat consists of data as atoms of information. Intrinsic computational processes, which drive changes of informational structures, resultsfrom operation of physical laws. The environment provides an organism with a variety of inputs in the form of both information and matter-energy, where the difference between information and matter-energy is not in the kind, but in the use organism makes of it. As there is no information without representation(Landauer, 1991), all information is carried by some physical carrier (light, sound, radio-waves, chemical molecules, etc.). The same physical object can be used by an organism as a source of information and as a source of nourishment/matter/energy. In general, the simpler the organism, the simpler the information structures of its body, the simpler the information carriers it relies on, and the simpler its interactions with the environment.
Cellular Computation
The environment is a resource, but at the same time it also imposes constraints that limit an agent’s possibilities. In an agent that can be described as a complex informational structure, constraints imposed by the environment drive the time development (computation) of its structures, and thus even its shape and behavior, to specific trajectories. This relationship between an agent and its environment is called structural coupling by (Maturana & Varela 1980) Experiments with bacteria performed by Ben-Jacob and Bassler illustrate the case in point:
“bacteria sense the environment and perform internal information processing (according to the internally stored information) to extract latent information embedded in the complexity of their environment. The latent information is then converted into usable or “active” information that affects the bacterium activity as well as intracellular changes.”(Ben-Jacob, 2009)
Bacteria interact with the environment,sense and extract its latent information. This information triggers cognitive processes that result in changes in its structure, function and behavior. Information can be seen as inducing “an internal condensed description (model or usable information) of the environment, which guides the organism’s functioning”. This is a process of intracellular computation, which proceeds via “gene computation circuits or gene logical elements. The circuits can either represent gene circuits or regulatory pathways.” As bacteria multiply by cell division,
“individuals in a growing colony begin to respond to the colony itself (i.e., information flow from the colony to the individual), these individuals respond by regulating their movements, growth rates, various tasks they perform, the chemical signals they send to other bacteria, and even their gene-network state (phenotypic state) according to the received signals.” (Ben-Jacob, 2009)
While individual bacteria (a few microns in size) can sense only a small area between replications, a colony that is composed of billions of bacteria can sense a large area and over long time periods.In a colony there is distributed coordination of tasks. Essential in the process of self-regulation of colony is the communication between units. Each unit is an autonomous system with internal information-management capabilities: interpretation, processing and storage of information. Ben-Jacob has found that “complex colonial forms (patterns) emerge through the communication-based singular interplay between individual bacteria (the micro-level) and the colony (the macro-level).“ (Waters & Bassler, 2005) describe the process of ”quorum-sensing” and communication between becteria that use two kind of languages – intra-species and inter-species chemical signalling. That is how they are capable of building films consisting of variety of species.
“The colony behaves much like a multi-cellular organism. As such, it can collect information about the environment over long times and extended distances and process it. Then, it performs distributed information processing to asses the situation and self-organize into different patterns as needed to function better in the encountered conditions.” (Ben-Jacob, 2009)
Experiments show “the distributed informationprocessing of the colony as a whole, including self-alteration and broadcasting messagesto initiate alterations in other bacteria.Such self-plasticity and decision-making capabilitieselevate the level of bacterial cooperationduring colonial self-organization.” “ new features collectivelyemerge during biotic self-organizationwhich involves natural information processingon every level, from the individual cells via cellmodules to the whole colony. The cells thusco-generate new information that is used tocollectively assume newly engineered cell traitsand abilities that are not explicitly stored inthe genetic information of the individuals. Bacteria cannot genetically store allthe information required for creating the colonialpatterns-the required information iscooperatively generated as self-organizationproceeds by bacterial communication, information-management,and self-plasticity capabilities. “Thus, thebacteria need only have genetically stored theguidelines for producing these capabilities andusing them to generate new information asrequired.”
Ben-Jacob conclude that bacteria ”can perform natural distributedinformation processing, learn from past experience,and possibly alter the genome organizationor even create new genes to better copewith novel challenges.” All those processes can be modelled as distributed concurrent computation in networks of programs.
Empirical results on the cognitive abilities of bacteria swarms, colonies and films by (Ben-Jacob et al., 2006; Ben-Jacob, 2008, 2009)(Ng & Bassler, 2009; Waters & Bassler, 2005) support the result about evolution, learning and cognition of (Harms, 2006)who proved a theorem that natural selection will always lead a population to accumulate information, and so to 'learn' about its environment. Okasha (Okasha, 2005) points out that “any evolving population 'learns' about its environment, in Harms' sense, even if the population is composed of organisms that lack minds entirely, hence lack the ability to have representations of the external world at all.”
Self-Organization, Cognitive Info-Computation and Evolution of Life
From bio computing we learn that in living organisms the biological structure (hardware) is at the same time a program (software) that controls the behavior of that hardware. Already in 1991 Kampis proposed a unified model of computation (in his book Self-Modifying Systems in Biology and Cognitive Science: A New Framework for Dynamics, Information and Complexity)as the mechanism underlying biological processes through “self-generation of information by non-trivial change (self-modification) of systems” (Kampis, 1991). This process of self-organization and self-generation of information is what is elsewhere described as morphological computation on different levels of organization of natural systems. Current research interest in adaptive networks goes in the same direction.(Dodig-Crnkovic & Giovagnoli, 2013)
However, understanding of the basic evolutionary mechanisms of accumulating information, at the same time increasing the information-processing capacities of organisms (memory, anticipation, computational efficiency), is only the first step towards a fully-fledged evolutionary understanding of cognition, but probably the most difficult one, as it requires a radical redefinition of fundamental concepts of information, computation and cognition in naturalist terms. According to Maturana,