#nr13
Info-computational constructivism in modelling of life as cognition – possibilities and limits
Anonymous
This paper will address the open question formulated as: Which levels of abstraction are appropriate in the synthetic modelling of life and cognition? within the framework of info-computational constructivism - an approach to natural phenomena as computational processes on informational structures.
At present we lack the common understanding of the process of life and cognition in living organisms with the details of co-construction of informational structures and processes in embodied, embedded cognizing agents in general, including artifactual cognitive agents.
Accepting Maturana and Varela’s identification of life with cognition (as self-generating process of interaction with the environment) I present the info-computational constructive approach to living beings as cognizing agents and suggest studying mechanisms of cognition, from the simplest to the most complex living systems, in order to be able to create classes of artifactual cognizing agents on different level of organization. The argument builds on Kauffman’s understanding of agency and self-organization.
Key Words – Computing nature, Info-computationalism, Morphological computing, Information physics, Evolution with Self-organization and Autopoiesis.
Introduction. Life as info-computational generative process of cognition.
“Living systems are cognitive systems, and living as a process is a process of cognition. This statement is valid for all organisms, with or without a nervous system.” (Maturana & Varela, 1980)
This paper presents a study within info-computational constructive framework of the life process as knowledge generation in living agents from the simplest living organisms to the most complex ones. Here <knowledge> of a primitive life form is very basic indeed – it is <knowledge> how to act in the world. An amoeba <knows> how to search for food and how to avoid dangers.
Since for a human it is impossible to grasp reality at once at all levels of organization, I analyze life as cognitive processes unfolding in a layered structure of nested information network hierarchies with corresponding computational dynamics (information processes) – from molecular, to cellular, organismic and social levels.
The aim is to indicate how computational approaches, dominant in (human) knowledge production today with new computational models under current development, together with insights from the research on theory of information and bioinformatics may be related to life itself understood as cognition.
The description of the conceptual framework of info-computationalism can be found in (Dodig-Crnkovic & Müller, 2011) (Dodig-Crnkovic, 2009) (Dodig-Crnkovic, 2006). The relationship between natural computing (such as biocomputing, DNA-computing, social computing, quantum computing, etc) and the traditional Turing machine model of computation is elaborated in (Dodig-Crnkovic, 2012a)(Dodig-Crnkovic, 2011a) (Dodig-Crnkovic, 2011b) (Dodig-Crnkovic, 2010a). Constructing/generation of knowledge within info-computational framework is discussed in (Dodig-Crnkovic, 2007) (Dodig-Crnkovic, 2010b)(Dodig-Crnkovic, 2010c)(Dodig-Crnkovic, 2008).
The problem of the relationship between closed and open systems, that is complementarity of constructive and axiomatic approaches is addressed in (Burgin & Dodig-Crnkovic, 2013).
Finally 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).
The Computing Nature
Computer pioneer Zuse was the first to suggest (in 1967) that the physical behavior of the entire universe is being computed on a basic level, possible to model on cellular automata, by the universe itself which he referred to as “Rechnender Raum” or Computing Space/Cosmos. Consequently, Zuse was the first pancomputationalist (naturalist computationalist), followed by many others like Ed Fredkin, Stephen Wolfram and Seth Lloyd – to name but a few. According to the idea of computing nature (naturalist computationalism or pancomputationalism) one can view the time development (dynamics) of physical states in nature as information processing (natural computation). Such processes include self-assembly, developmental processes, gene regulation networks, gene assembly in unicellular organisms, protein-protein interaction networks, biological transport networks, and similar. (Dodig-Crnkovic & Giovagnoli, 2013)
Within info-computationalism, two basic concepts information and computation (the dynamics of informational structure) are mutually interdependent (Dodig-Crnkovic, 2011a) (Chaitin, 2007) – so the framework is a synthesis of informational structural realism and natural computationalism.
Informational structural realism (Floridi, 2003) takes information to be the fabric of the universe (for an agent). As a consequence the process of dynamical changes of the universe makes the universe a huge computational network where computation is information processing. As it corresponds to the dynamic of processes that exist in the universe, it is necessarily both discrete and continuous, on both symbolic and sub-symbolic[1] level. Information and computation are two fundamental and inseparable elements necessary for naturalizing cognition and knowledge. (Dodig-Crnkovic, 2009)
Physicists Zeilinger (Zeilinger, 2005) and Vedral (Vedral, 2010) suggest the possibility of seeing information and reality as one. This is in accord with informational structural realism which says that reality is made of informational structures (Floridi, 2009)(Floridi, 2008) (Sayre, 1976) as well as with info-computational epistemology (Dodig-Crnkovic, 2009) based on informational structural realism in conjunction with natural computationalism. Reality for an agent is informational and agent-dependent (observer-dependent) and consists of structural objects, which are adjusted to the shared reality of agents community of practice. This brings together metaphysical views of Wiener (“information is information, not matter or energy”) and Wheeler (“it from bit”) with Zuse, Fredkin, Lloyd, Wolfram and others view of computing nature.
In sum: information is the structure, the fabric of reality. The world exists independently from us (realist position of structural realism) in the form of proto-information, the potential form of existence corresponding to Kant’s das Ding an sich. That proto-information becomes information (“a difference that makes a difference” according to (Bateson, 1972)) for a cognizing agent in a process of interaction through which specific aspects of the world get uncovered.
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.” (italics added) (Hewitt, 2007)
This has profound consequences for epistemology and relates to the ideas of participatory universe, (Wheeler, 1990) endophysics (Rössler, 1998) and observer-dependent knowledge production as understood in second-order cybernetics. Combining Bateson and Hewitt insights, on the basic level, information is the difference in one physical system that makes difference in another physical system.
Of special interest with respect to knowledge generation are agents - systems able to act on their own behalf.
The world as it appears to an agent depends on the type of interaction through which the agent acquires information[2]. 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 similar. Our knowledge about this potential information or proto-information which reveals with help of scientific instruments continuously increase with the development of new devices and the new ways of interaction with the world, both theoretical and material constructs (Dodig-Crnkovic & Mueller, 2009).
Information and Computation in Cognizing Agents
“Intelligence organizes the world by organizing itself.” (Piaget, 1955)
“Ontologically, Eigenvalues and objects, and likewise, ontogenetically, stable behavior and the manifestation of a subject’s ‘grasp’ of an object cannot be distinguished.” (Foerster, 1977) p. 280 (italics added)
Studies in biology, ethology and neuroscience, which have increased our knowledge of biological cognitive functions, have led to the insight that the most important feature of cognition is its ability to deal efficiently with complexity. This, together with the increase in power of electronic computing brings us closer to adequate modelling of intelligent behaviour. From the computationalist point of view intelligence may be seen as capacity based on several levels of data processing in a cognizing agent, as argued by Minsky. Data, information, perceptual images and knowledge are organized in a multiscale model of the brain and nervous system, up to the emergent level of consciousness according to (Minsky, 1986)(Minsky, 2011). Multiresolutional models have proven to be a good way of studying complexity in biological systems, and they are also being implemented in artificial intelligence, AI (Goertzel, 1993).
The advantage of computational approaches is their testability. Cognitive robotics research, e.g. presents us with a sort of laboratory where our understanding of cognition can be tested in a rigorous manner. From cognitive robotics it is becoming evident that cognition and intelligence are closely related to agency. Anticipation, planning and control are essential features of intelligent agency. A similarity has been found between the generation of behaviour in living organisms and the formation of control sequences in artificial systems. (Pfeifer & Bongard, 2006)(Pfeifer, Lungarella, & Iida, 2007)
Information produced from sensory data processed by an agent is a result of perception. From the point of view of data processing, perception can be seen as an interface between the data (the world) and an agent’s perception of the world. (Hoffman, 2009) criticizes traditional view of perception as a true picture of the world.
“Instead, our perceptions constitute a species-specific user interface that guides behavior in a niche. Just as the icons of a PC's interface hide the complexity of the computer, so our perceptions usefully hide the complexity of the world, and guide adaptive behavior. This interface theory of perception offers a framework, motivated by evolution, to guide research in object categorization. ”
Thus, perception cannot be cut off on one side of the interface, inside an agent and its brain. Patterns of information are both in the world and in the functions and structures of the agent. Information is the difference in the world that makes difference in an agent.
With perception as an interface, sensorimotor activities play a central role in realizing this function of connecting the inside with the outside worlds of an agent, endogenous with the exogenous. Perception has co-evolved with sensorimotor skills of an organism. Enactive approach to perception (Noë, 2004) emphasizes the role of sensorimotor abilities, that can be connected with the changing informational interface between an agent and the world, and thus increasing information exchange.
Traditionally, symbolic AI was an attempt to model cognition and intelligence as symbol manipulation, which turned out insufficient. (Clark, 1989) In order to improve and complement symbolic approaches, Smolensky proposed mechanism of an intuitive processor (which is not accessible to symbolic intuition), with a conscious rule interpreter:
“What kinds of programs are responsible for behavior that is not conscious rule application? I will refer to the virtual machine that runs these programs as the *intuitive processor*. It is presumably responsible for all of animal behavior and a huge proportion of human behavior: Perception, practiced motor behavior, fluent linguistic behavior, intuition in problem solving and game-playing--in short, practically all skilled performance.” (Smolensky, 1988)
Sloman has developed interesting ideas about mind as virtual machine running on the brain in (Sloman, 2002) which also addresses the symbol grounding problem.
From the point of view of info-computationalism, a mechanism behind this virtual machine hierarchy is computational self-organization of information, i.e. morphological computing, see (Dodig-Crnkovic, 2012b) and references therein. In his new research programme, Sloman goes a step further studying meta-morphogenesis which is morphogenesis of morphogenesis, (Sloman, 2013) – a way of thinking in the spirit of second order cybernetics.
Info-Computationalist Epistemological Constructivism
The above understanding of cognition is adopted by info-computationalism as it provides a notion of cognition in degrees, which provides a bridge from human-level cognition to minimal cognition in simplest biological forms and intelligent machines (under development). Within the framework of info-computational naturalism (Dodig-Crnkovic, 2009) knowledge is seen as a result of successive structuring of data, where data are simplest information units, signals acquired by a cognizing agent through the senses/ sensors/ (Dodig-Crnkovic, 2007) (Skyrms, 2010). Information is meaningful data, which can be turned into knowledge by an interactive computational process going on in the cognizing agent. Information is always embodied in a physical substrate: signal, molecule, particle or event which will induce change of a structure or a behaviour of an agent (Landauer, 1991). The world (reality) for an agent presents potential information, both outside and within an agent.
Knowledge, on the other hand, always resides in a cognitive agent. Semantics develops as data → information → knowledge structuring process, in which complex structures are self-organized by the computational processing from simpler ones. Meaning of information is thus defined for an agent and a group of agents in a network and it is given by the use information has for them. Knowledge generation as information processing in biological agents presupposes natural computation, defined by MacLennan (MacLennan, 2004) as computation occurring in nature or inspired by that in nature, which is the most general current computation paradigm.
Knowledge Generation as Morphological Computation
Traditional theoretical Turing machine model of computing is equivalent to algorithms/effective procedures, recursive functions or formal languages. Turing machine is a logical device, a model for execution of an algorithm. However, if we want adequately to model computing nature including biological structures and processes understood as embodied physical information processing, highly interactive and networked computing models beyond Turing machines are needed, as argued in (Dodig-Crnkovic & Giovagnoli, 2013). In order to develop general theory of the networked physical information processing, we must also generalize the ideas of what computation is and what it might be. For new computing paradigms, see for example (Rozenberg, Bäck, & Kok, 2012)(Burgin, 2005)(MacLennan, 2004) (Wegner, 1998)(Hewitt, 2012)(Abramsky, 2008). Turing machines form the proper subset of the set of information processing devices.
In the computing nature, knowledge generation should be studied as a natural process. That is the main idea of Naturalized epistemology (Harms, 2006), where the subject matter is not our concept of knowledge, but the knowledge itself as it appears in the world[3] as specific informational structures of an agent. The origin of knowledge in first living agents is not well researched, as the idea still prevails that knowledge is possessed only by humans. However, there are different types of knowledge and we have good reasons to ascribe “knowledge how” and even simpler kinds of “knowledge that” to other living beings. Plants can be said to possess memory (in their bodily structures) and ability to learn (adapt, change their morphology) and can be argued to possess rudimentary forms of knowledge. On the topic of plant cognition see Garzón in (Pombo, O., Torres J.M., Symons J., 2012) p. 121. In his Anticipatory systems (Rosen, 1985) claim as well: “I cast about for possible biological instances of control of behavior through the utilization of predictive models. To my astonishment I found them everywhere[…] the tree possesses a model, which anticipates low temperature on the basis of shortening days” Popper (Popper, 1999) p. 61 ascribes the ability to know to all living: ”Obviously, in the biological and evolutionary sense in which I speak of knowledge, not only animals and men have expectations and therefore (unconscious) knowledge, but also plants; and, indeed, all organisms.”