Modeling Life as Cognitive Info-computation

1. Computability in Europe 2014: Language, Life, Limits. June 23-27, Budapest Modeling Life as Cognitive Info-computation Gordana Dodig Crnkovic Professor of Computer Science School of Innovation, Design and EngineeringMälardalen University, Sweden http://www.idt.mdh.se/~gdc/

2. Mälardalen University Sweden12,000 students and around 900 employees, of which 67 are professors

3. What is Cognition? After half a century of research in cognitive science, cognition still lacks a commonly accepted definition (Lyon, 2005). Textbook description of cognition: “all the processes by which sensory input is transformed, reduced, elaborated, stored, recovered and used” (Neisser, 1967) is so broad that it includes present day robots. On the other hand, the Oxford dictionary definition: “the mental action or process of acquiring knowledge and understanding through thought, experience, and the senses” applies only to humans. *Mental = relating to the mind. Mind is set of processes in which consciousness, perception, affectivity, judgment, thinking, and will are based.

4. Cognitive science Biology /embodiment/embeddedness Biology /embodiment/embeddedness http://cacm.acm.org/magazines/2011/8/114944-cognitive-computing/fulltext Biology /embodiment/embeddedness Biology /embodiment/embeddedness (situatedness) http://en.wikipedia.org/wiki/Cognitive_science

5. Cognition, different levels of understanding • Traditional anthropogenic approach to cognition* – only humans are cognitive agents • Biogenic approaches* – cognition is ability of all living organisms, no matter how “primitive” – goes a level below the complexity of human language – to complex systems chemical signaling and regulation processes. (Maturana & Varela, 1980; Maturana, 1970), argued that cognition and life are identical processes. • New sub-biotic approaches to cognition assume that it is possible to construct cognitive agents starting from abiotic systems – a level below biogenic cognition. The question is if abiotic systems can be considered cognitive, in what sense and on which level. * (Lyon, 2005)

6. Anthropogenic Cognition vs. Anthropogenic Intelligence • (Anthropogenic) Cognition is the PROCESS by which humans acquire, integrate and generate knowledge. It is the result of attention, perception, memory, and executive functions of learning and behavior generation (information integration and transformation of perception into higher order symbols; comparison of incoming information with the information stored in memory together with value system and biological drives) • (Anthropogenic) Intelligence is the ABILITY to understand and reason upon (i.e. structure and interrelate and operate upon) what is perceived, memorized and learned. • Intelligence as ABILITY is based on cognition as PROCESS.

7. Similarly, Biogenic and Abiotic Cognition vs. Biogenic and Abiotic (Artifactual) Intelligence • (Biogenic and Abiotic) Cognition is the PROCESS by which simple living organisms acquire, integrate and generate <knowledge>. It is the result of attention, perception, memory, and executive functions of learning and behavior generation. • (Biogenic and Abiotic) Intelligence is the ABILITY to structure, interrelate and operate upon information that is perceived, memorized and learned. • Intelligence as ABILITY is based on cognition as PROCESS.

8. Connecting Anthropogenic with Biogenic and Abiotic Cognition We focus on cognition and propose the common framework for understanding Anthropogenic, Biogenic and Abiotic Cognition. We argue that (as in the rest of biology) – nothing makes sense except for in the light of evolution (Dobzhansky, 1973) and the cognition as a process can only be understood in the light of evolution. Regarding abiotic systems we will compare their “cognitive behavior” with living organisms, and draw conclusions.

9. Living as a process is a process of cognition • “A cognitive system is a system whose organization defines a domain of interactions in which it can act with relevance to the maintenance of itself, and the process of cognition is the actual (inductive) acting or behaving in this domain. Living systems are cognitive systems, and living as a process is a process of cognition. This statement is valid for all organisms, with and without a nervous system.” (Maturana, 1970) • In 1991, Kampis proposed a unified model of computation as the mechanism underlying biological processes through “self-generation of information by non-trivial change (self-modification) of systems” (Kampis, 1991. Self-Modifying Systems in Biology and Cognitive Science: A New Framework for Dynamics, Information and Complexity).

10. Information, computation, cognitionAgent-centered Hierarchies of Levels information computation cognition Fruit fly brain neurons Fruit fly larva In this lecture I will present a unified framework for modeling of information, computation and cognition from an agents perspective. Fruit fly brain micrograph http://www.sciencedirect.com/science/article/pii/S0378437104014839

11. Starting from anthropogenic perspective: The brain development - Cognition as biological phenomenon “The brain development may be carried out based on the basic body-organization-blueprints that are specific to an animal species depending on their strategy to survive in an environment. To understand how our brains are established in the course of evolution, we have been conducting a comparison of the structure and function of the gene that are essential for establishing body organization and brain development in a wide rage of animals with nervous system.” http://lcn.brain.riken.jp/tool_kit_evolution.htm

12. Wonders of evolution – the smallest insect with brain, smaller than an amoeba • Size of the smallest insect and two protozoans in comparison. • Megaphragma mymaripenne. • Paramecium caudatum. • Amoeba proteus. • Scale bar for A–C is 200 μm. B and C are made up of a single cell, A the wasp complete with eyes, brain, wings, muscles, guts – is actually smaller. This wasp is the third smallest insect alive. the smallest nervous systems of any insect, consisting of just 7,400 neurons. Housefly has 340,000 Honeybee has 850,000. 95% of the wasps’s neurons have no nucleus. http://www.sciencedirect.com/science/article/pii/S1467803911000946 The smallest insects evolve anucleate neuronsArthropod Structure & Development, Volume 41, Issue 1, January 2012, Pages 29–34

13. Information, computation, cognition Agent-centered Hierarchy of Levels Natural information processing Henry Markram (2012) The Human Brain Project, Scientific American 306, 50 – 55 Human connectome http://outlook.wustl.edu/2013/jun/human-connectome-project http://www.nature.com/scientificamerican/journal/v306/n6/pdf/scientificamerican0612-50.pdf The Human Brain Project

14. Current brain research initiatives The Human Brain Project (HBP) is a large scientific research project, directed by the École polytechnique fédérale de Lausanne and largely funded by the European Union, which aims to simulate the complete human brain on supercomputers to better understand how it functions. The BRAIN Initiative (Brain Research through Advancing Innovative Neurotechnologies, also referred to as the Brain Activity Map Project) is a proposed collaborative research initiative announced by the Obama administration on April 2, 2013, with the goal of mapping the activity of every neuron in the human brain.Based upon the Human Genome Project, the initiative has been projected to cost more than $300 million per year for ten years. Source: Wikipedia

15. Current brain research initiatives The Allen Institute conducting and completing large-scale brain mapping projects for the last 10 years. In early 2012 launched three additional major research initiatives to drive critical advances in understanding how the brain works and develops. • Neural Coding (understanding how information is encoded and decoded in the mammalian brain) • Molecular Networks (understanding how information is encoded and decoded within a cell) • Cell Types (large-scale descriptive resources of human and mouse brain cell types at molecular, morphological and connectional levels) • Atlasing (collection of online public resources integrating extensive genomic and neuroanatomic data) http://www.alleninstitute.org/science/research_programs/index.html

16. The Strategy of Info-Computational Approach • Even though anthropogenic approach to cognition is the oldest and by far the most dominant one, it is the most difficult approach to the most complex problem – embodied human brain. • The study of biogenic and abiotic cognition can help us trace evolutionary roots of cognitive capacities in living organisms (biogenic) and construct (abiotic) artifact with cognitive and intelligent behavior (cognitive computing and cognitive robotics). • Therefore we start with simplest living systems such as bacteria to try to understand the basis of their cognitive behavior in informational structures and their dynamics (computational processes).

17. Information, computation, cognition. Agency-based Hierarchies of Levels Introducing generalized concepts of information and computation. Short summary of the argument: • Information presents a structure consisting of differences in one system that cause the differences in another system. In other words, information is <observer>*-relative. • Computation is information processing (dynamics of information). It is physical process of morphological change in the informational structure (physical implementation of information, as there is no information without physical implementation.) *<> brackets indicate that the term is used in a broader sense than usually.

18. Information, computation, cognition.Agency-based Hierarchies of Levels • Both information and computation appear on many different levels of organisation/abstraction/resolution/granularity of matter/energy in space/time. • Of all agents (entities capable of acting on their own behalf) only living agentshave the ability to actively make choices so to increase the probability of their own continuing existence. This ability of living agents to act autonomously on its own behalf is based on the use of energy/matter and information from the environment.

19. Information, computation, cognition. Agency-based Hierarchies of Levels • Cognition consists of all (info-computational) processes necessary to keep living agent’s organizational integrity on all different levels of its existence. Cognition = info-computation • Cognition is equivalent with the (process of) life.* Its complexity increases with evolution.This complexification is a result of morphological computation. * Maturana, H. & Varela, F., 1980. Autopoiesis and cognition: the realization of the living, Dordrecht Holland: D. Reidel Pub. Co.

20. Information as a fabric of reality “Information is the difference that makes a difference. “Gregory Bateson It is the difference in the world that makes the difference for an agent. Here the world includes agents themselves too. “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.” Carl Hewitt Bateson, G. (1972). Steps to an Ecology of Mind: Collected Essays in Anthropology, Psychiatry, Evolution, and Epistemology pp. 448–466). University Of Chicago Press. Hewitt, C. (2007). What Is Commitment? Physical, Organizational, and Social. In P. Noriega, J. Vazquez, Salceda, G. Boella, O. Boissier, & V. Dign (Eds.), Coordination, Organizations, Institutions, and Norms in Agent Systems II (pp. 293 –307). Berlin, Heidelberg: Springer Verlag.

21. Information structures as a fabric of reality (thus structured/organized data) for an agent Informational structural realism (Floridi, Sayre) argues that information (for an agent) constitutes the fabric of reality: Reality consists of informational structures organized on different levels of abstraction/resolution. See also: Van Benthem and Adriaans (2008) Philosophy of Information, In: Handbook of the philosophy of science series. http://www.illc.uva.nl/HPI Ladyman J. and Ross D., with Spurrett D. and Collier J. (2007) Every Thing Must Go: Metaphysics Naturalized, Oxford UP Floridi, L. (2008) A defence of informational structural realism, Synthese161, 219-253. Sayre, K. M. (1976) Cybernetics and the Philosophy of Mind, Routledge & Kegan Paul, London.

22. The relational definition of information Combining definitions of Bateson: “Information is a difference that makes a difference.”(Bateson, 1972) and Hewitt: ”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), we get: Information is defined as the difference in one physical system that makes the difference in another physical system.

23. Structure vs. process For all living agents, information is the fabric of reality. But: the knowledge of structures is only half a story. The other half are changes, processes – information dynamics. (In classical formulation: being and becoming.) Information processing will be taken as the most general definition of computation. This definition of computation has a profound consequence – if computation is the dynamics of informational structures of the universe, the dynamics of the universe is a network of computational processes (natural computationalism). Dodig-Crnkovic, G., Dynamics of Information as Natural Computation, Information 2011, 2(3), 460-477; Selected Papers from FIS 2010 Beijing, 2011.

24. Reality for an agent - informational structure with computational dynamics Information is defined as the difference in one physical system that makes the difference in another physical system. This reflects the relational character of information and thus agent-dependency which calls for agent-based or actor models. As a synthesis of informational structural realism and natural computationalism, I propose info-computational structuralism that builds on two basic concepts: information (as a structure) and computation (as a dynamics of an informational structure) (Dodig-Crnkovic, 2011). (Dodig-Crnkovic & Giovagnoli, 2013) Information and computation are two basic and inseparable elements necessary for naturalizing <cognition>. (Dodig-Crnkovic, 2009)

25. Computational modeling in cognitive science Symbolic modeling evolved from the computer science paradigms using the technologies of Knowledge-based systems - "Good Old-Fashioned Artificial Intelligence" (GOFAI). Used in expert systems and cognitive decision making, and extended to socio-cognitive approach. Subsymbolic modeling includes Connectionist/neural network models.

26. Computational modeling in cognitive science Dynamical systems theory closely related to ideas about the embodiment of mind and the environmental situatedness of human cognition based on physiological and environmental events. The most important here is the dimension of time. Neural-symbolic integration techniques putting symbolic models and connectionist models into correspondence. Bayesian models of brain function whichassume that the nervous system maintains internal probabilistic models that are updated by neural processing of sensory information using methods approximating those of Bayesian probability.

27. It is important to notice:Computationalism is not what it used to be… … that is, the thesis that persons are Turing machines. Turing Machine is a model of computation equivalent to algorithm and it may be used for description of different processes in living organisms. We need computational models for the basic characteristics of life as the ability to differentiate and synthesize information, make a choice, to adapt, evolve and learn in an unpredictable world. That requires computational mechanisms and models which are not mechanistic and predefined as Turing machine.* * We need learning such as PAC Probably Approximately Correct – Leslie Valiant

28. Computationalism is not what it used to be …… that is the thesis that persons are Turing machines Computational approaches that are capable of modelling adaptation, evolution and learning are found in the field of natural computation and computing nature. Cognitive computing and cognitive robotics are the attempts to construct abiotic systems exhibiting cognitive characteristics. It is argued that cognition comes in degrees, thus it is meaningful to talk about cognitive capabilities of artifacts, even though those are not meant to assure continuing existence, which was the evolutionary role of cognition in biotic systems.

29. Turing computation: “Turing on Super-Turing and adaptivity” according to Siegelmann “Biological processes are often compared to computation and modeled on the Universal Turing Machine. While many systems or aspects of systems can be well described in this manner, Turing computation can only compute what it has been programmed for.(…) Yet, adaptation, choice and learning are all hallmarks of living organisms. This suggests that there must be a different form of computation capable of this sort of calculation. (…) Super-Turing model is both capable of modeling adaptive computation, and furthermore, a possible answer to the computational model searched for by Turing himself.” Hava T. Siegelmann, Turing on Super-Turing and adaptivity, Progress in Biophysics and Molecular Biology, http://www.sciencedirect.com/science/article/pii/S0079610713000278

30. Actor model of concurrent distributed computation “In the Actor Model [Hewitt, Bishop and Steiger 1973; Hewitt 2010], computation is conceived as distributed in space, where computational devices communicate asynchronouslyand the entire computation is not in any well-defined state. (An Actor can have information about other Actors that it has received in a message about what it was like when the message was sent.) Turing's Model is a special case of the Actor Model.” (Hewitt, 2012) Hewitt’s “computational devices” are conceived as computational agents – informational structures capable of acting on their own behalf.

31. Actor model of concurrent distributed computation Actors are the universal primitives of concurrent distributed digital computation. In response to a message that it receives, an Actor can make local <decisions>, create more Actors, send more messages, and designate how to respond to the next message received. For Hewitt, Actors become Agents only when they are able to process expressions for commitments including the following: Contracts, Announcements, Beliefs, Goals, Intentions, Plans, Policies, Procedures, Requests, Queries. In other words, Hewitt’s Agents are human-like or if we broadly interpret the above capacities, life-like Actors.

32. Actor model of concurrent distributed computation Unlike other models of computation that are based on mathematical logic, set theory, algebra, etc. the Actor model is based on physics, especially quantum physics and relativistic physics. (Hewitt, 2006) Summary of interactions between particles described by the Standard Model. http://en.wikipedia.org/wiki/Standard_Model

33. Computing nature and nature inspired computation If it looks like a duck, if it walks like a duck and it quacks like a duck, is it a duck? (If it looks like computation is it computation?) Peter J. Denning. 2007. Computing is a natural science. Commun. ACM 50, 7 (July 2007), 13-18. DOI=10.1145/1272516.1272529 http://doi.acm.org/10.1145/1272516.1272529

34. Computing cells: self-generating systems “a component system is a computer which, when executing its operations (software) builds a new hardware.... [W]e have a computer that re-wires itself in a hardware-software interplay: the hardware defines the software and the software defines new hardware. Then the circle starts again.” Kampis (1991) p. 223 Kampis (1991) Self-Modifying Systems in Biology and Cognitive Science. A New Framework For Dynamics, Information, and Complexity, Pergamon Press Dodig Crnkovic, G. (2011). Significance of Models of Computation from Turing Model to Natural Computation. Minds and Machines, (R. Turner and A. Eden guest eds.) Volume 21, Issue 2, p.301. Complex biological systems must be modeled as self-referential, self-organizing "component-systems" (George Kampis) which are self-generating and whose behavior, though computational in a general sense, goes far beyond Turing machine model.

35. Computation is implemented at different levels of resolution – Computing architecture Some layered computational architectures

36. Computation as information processing.Data to information via computation Elements of information processing in an information system Computational processes on information structures

37. Cognitive information processing is not what it used to be … Cognitive Information Processing theory of learning according to (Gagné, 1985): This is an old and simplistic idea of cognition as information processing. Missing in this scheme are feedback loops that are absolutely essential for cognition and learning. Also missing is information integration from different sensors and couplings to actuators. Memory is not a passive storage but an active ingredient in perception, that is both used for recognition and anticipation. Cognitive / Information Processing Theory of Learning according to (Gagné, 1985)

38. Multisensori information integration Information integration is critical for the brain to interact effectively with our multisensory environment. The human brain integrates information from multiple senses with prior knowledge to form a coherent and more reliable percept of its environment. (learning) Within the cortical hierarchy, multisensory perception emerges in an interactive process with top-down prior information constraining the interpretation of the incoming sensory signals. Marcin Schröder in the book Computing Nature adresses the Dualism of Selective and Structural Information, describing information integration. http://www.birmingham.ac.uk/research/activity/behavioural-neuro/comp-cog-neuro/index.aspx

39. Cognition: Agency-based hierarchies of levels. World as information for an agent Potential information Cognition Actual information for an agent From: http://www.alexeikurakin.org http://www.tbiomed.com/content/8/1/4 scale-invariance of self-organizational dynamics of energy/matter at all levels of organizational hierarchy

40. Agency-based hierarchies of levels. World as information for an agent Actual Information C-elegans Potential information Outside reality for C-elegans Interaction interface for C-elegansCognition C. Elegans has 302 neurons (humans have 100 billion). The pattern of connections between neurons has been mapped out decades ago using electron microscopy, but knowledge of the connections is not sufficient to understand (or replicate) the information processor they represent, for some connections areinhibitorywhile others are excitatory. http://www.33rdsquare.com/2013/07/david-dalrymple-update-on-project.html

41. Reality for an agent – an observer-dependent reality Reality for an agent is an informational structure with which agent interacts. As systems able to act on their own behalf and make sense (use) of information, cognitive agents are of special interest with respect to <knowledge>* generation. This relates to the idea of participatory universe, (Wheeler, 1990) “it from bit” as well as to endophysics or “physics from within” where an observer is being within the universe, unlike the “god-eye-perspective” from the outside of the universe. (Rössler, 1998) *<knowledge> for a very simple agent can be the ability to optimize gains and minimize risks. (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.”

42. An illustration: Agent-dependent multiscale modeling of complex chemical system Observer-centric model – enhanced resolution where observation is made – where chemical reaction takes place The Nobel Prize in Chemistry 2013 “for the development of multiscale models for complex chemical systems” ... Karplus, Levitt and Warshel managed to make Newton's classical physics work side-by-side with the fundamentally different quantum physics.The strength of classical physics was that calculations were simple and could be used to model large molecules but no way to simulate chemical reactions for which chemists use quantum physics. But such calculations require enormous computing power. Nobel Laureates in chemistry devised methods that use both classical and quantum physics. In simulations of how a drug couples to its target protein in the body, the computer performs quantum theoretical calculations on those atoms in the target protein that interact with the drug. The rest of the large protein is simulated using less demanding classical physics. Today the computer is just as important a tool for chemists as the test tube. Simulations are so realistic that they predict the outcome of traditional experiments. http://www.nobelprize.org/nobel_prizes/chemistry/laureates/2013/advanced-chemistryprize2013.pdf

43. Info-computational framework and levels The question of levels of organization/levels of abstraction for an agent is analyzed within the framework of info-computational constructivism, with natural phenomena modeled as computational processes on informational structures. Info-computationalism is a synthesis of informational structuralism (nature is an informational structure for an agent) (Floridi, Sayre) and natural computationalism/pancomputationalism(nature computes its future states from its earlier states) (Zuse, Fredkin, Wolfram, Chaitin, Lloyd). Two central books presenting the diversity of research on information and computation: Adriaans P. and van Benthem J. eds. 2008. Philosophy of Information (Handbook of the Philosophy of Science) North Holland. Rozenberg, G., T. Bäck, and J.N. Kok, eds. 2012. Handbook of Natural Computing. Berlin Heidelberg: Springer.

44. Life as cognition. Autopoiesis as self-reflective process ”Living systems are cognitive systems, and living as a process is a process of cognition. This statement is valid for all organisms, with and without a nervous system.” Humberto Maturana, Biology of Cognition, 1970 • Maturana and Varela (1980) define "autopoiesis" as follows: An autopoietic system is a system organized (defined as a unity) as a networkofprocessesofproduction (transformation and destruction) ofcomponentsthatproduces the components, suchthat: • throughtheirinteractions and transformations continuouslytheyregenerate and realize the networkofprocesses (relations) thatproducedthem; and • theyconstitute it (the system) as a concreteunity in the space in whichthey (the components) exist by specifying the topologicaldomainofitsrealization as such a network.

45. Living agents – basic levels of cognition A living agent is an entity acting on its own behalf, with autopoietic properties that is capable of undergoing at least one thermodynamic work cycle. (Kauffman, 2000) This definition differs from the common belief that (living) agency requires beliefs and desires, unless we ascribe some primitive form of <belief> and <desire> even to a very simple living agents such as bacteria. The fact is that they act on some kind of <anticipation> and according to some <preferences> which might be automatic in a sense that they directly derive from the organisms morphology. Even the simplest living beings act on their own behalf.

46. Living agents – basic levels of cognition Although a detailed physical account of the agents capacity to perform work cycles and so persist* in the world is central for understanding of life/cognition, as (Kauffman, 2000) (Deacon, 2007) have argued in detail, present argument is primarily focused on the info-computational aspects of life. Given that there is no information without physical implementation (Landauer, 1991), computation as the dynamics of information is the execution of physical laws. *Contragrade processes (that require energy and do not spontaneously appear in nature) become possible by connecting with the orthograde (spontaneous) processes which provide source of energy.

47. Living agents – basic levels of cognition Kauffman’s concept of agency (also adopted by Deacon) suggests the possibility that life can be derived from physics. That is not the same as to claim that life can be reduced to physics that is obviously false. However, in deriving life from physics one may expect that both our understanding of life as well as physics will change. We witness the emergence of information physics (Goyal, 2012) (Chiribella, G.; D’Ariano, G.M.; Perinotti, 2012) as a possible reformulation of physics that may bring physics and life/cognition closer to each other.

48. Levels of organization of life/cognition The origin of <cognition> in first living agents is not well researched, as the idea still prevails that only humans possess cognition and knowledge. However, there are different types of <cognition> and we have good reasons to ascribe simpler kinds of <cognition> to other living beings. Bacteria collectively “collects latent information from the environment and from other organisms, process the information, develop common knowledge, and thus learn from past experience”(Ben-Jacob, 2009) 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 simple forms of cognition. Ben-Jacob, E. (2009). Learning from Bacteria about Natural Information Processing. Annals of the New York Academy of Sciences, 1178, 78–90.

49. Evolution of the Nervous System • Nerve net – jellyfish • Simple brain & nerve cord – flatworm • Brain & nerve cord with ganglia – earthworm • Increasing forebrain – fish, bird & human • Olfactory – fish • Complex behavior – birds • Reasoning & cognition – humans Dee Unglaub Silverthorn, Human Physiology- an Integrated Approach, 3rd ed

50. Evolution of the Nervous System Figure 9-1: Evolution of the nervous system