Challenges, Agents and Coordination: how an action ontology can help us tackle both practical and foundational problems QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. Francis Heylighen Evolution, Complexity and Cognition group Vrije Universiteit Brussel Ontology Philosophy of what is • What reality is constituted of Most basic elements or concepts • E.g. matter, ideas, energy, fields, spirit… Building blocks of all higher level theories • Different ontologies result in different models • This has practical implications for solving problems Newtonian Ontology The world is constituted out of particles • • • Permanent pieces of matter Moving in space and time following fixed “laws of Nature” Shortcomings • No explanation for emergent phenomena: complexity, evolution, mind, life, society, intelligence… • No meaning or purpose Need for a process ontology Change is basic • • Not objects, but processes are primary Allows for novelty, creativity, evolution Complexity is basic • No primitive, independent elements Phenomena only exist in relation/interaction to others • Everything is connected • Whole is more than sum of the parts Some precursors Heraclitus • You can never step in the same river twice Process Metaphysics • Whitehead, Bergson, Teilhard… Valentin Turchin: • “cybernetic ontology of action” My own history ± 1976 ± 1984 ± 1987 1990Õs ± 2000 ± 2007 2009 2010-now relational principle, generalized natural selection Òstructural languageÓformalism discovery of cybernetics self-organization & evolution of cooperation multi-agent systems stigmergy life is an adventure challenges & coordination The basic element Action = elementary process • Transforming some condition X into a different condition Y • X→Y Y Interpretations • if X, then Y • X = “cause”, Y = “effect” • X = “initial state”, Y = “next state” • X = “condition” (for action to occur), Y = “action” (creation of new condition) X Action Examples Elementary particle reaction • n p + e- + e (Beta decay of neutron) Chemical reaction • 2H2 + O2 2H2O (production of water) Causal rule • Glass falls → Glass breaks More examples Action of thermostat • Temperature < 21° → switch on heating Animal action • Smell food → eat food Human action • See friend greet friend Conditions What are the conditions X and Y in X→ Y? • Condition = distinguishable class of situations • “state of the world” at the beginning of the action Distinguished by the actions possible in that state • states differ if and only if possible actions differ • Observation/distinction is an action Formally: state = set of all potential actions • action performed => state changes Bootstrapping logic Action is defined as change of state State is defined as collection of possible actions Action is the true primitive • State is a more complex, derived concept • But which fits in better with our “classical” intuition Example: n p + e- + e • state n (neutron) defined by reactions in which it participates • e.g. ability to decay into a proton, electron and neutrino • proton, electron, etc. are similarly defined by the actions in which they take part Agents Agent = part of condition necessary for action But which is not affected by action • A+X→A+Y • A = agent or catalyst of the action X → Y Agents have a certain invariance or stability • “objects” rather than processes Agents are produced by variation and selection • stable conditions survive longer than unstable ones • => they will become more common Physics Particle = simplest possible agent • Fermion (e.g. proton, neutron, electron…) • • Invariant during action: A + X → A + Y • Observed via boson (e.g. photon) exchanges Example: • e- → e- + • photographic plate + → photographic plate + trace Space-Time Network of actions determines “causal structure” • light-cone separates “time-like” from “space-like” connections • Actions without parallel actions are “horismotic” (= “light-like”) • Particles follow time-like trajectories Topology of space and time can be reconstructed from this causal structure (Kronheimer & Penrose, 1967) Conclusion: particles, space and time emerge from networks of actions, not vice-versa Macroscopic Causality Particular action: • X + B (background conditions) → Y + B’ • Every X + B state is unique General Action • X→Y X reduced to a general category including many unique states Abstraction is made of the background Either because it does not affect the action, or is invariant (agent) Example • Dropping + B → falling + B’ • B = gravitation, weight, object heavier than air, etc. Directionality Actions tend to have a preferred direction • X → Y, but not Y → X • In general irreversible This produces attractors in the state space • regions that you can enter but not leave This implies equifinality • Different initial states lead to the same final states Phase portrait attractor attractor attractor Goal-directedness Attractors = implicit goals of actions/agents • i.e. situations that all actions go towards • and will return to even when perturbed Fitness = “attractivity” of a state = underlying goal/value of all agents Examples: • Physics: goal = minimal potential energy • Biology: goal = maximal survival and reproduction • Psychology: goal = maximal happiness • Economics: goal = maximal “utility” (benefit) The Intentional Stance Action: A+X→A+Y Agent A has • Belief or Sensation about the situation it is in • Intention about what action to do next • initial condition X to which A reacts Action Y that A performs Desire or Goal Attractor to which A’s actions eventually lead Intentional vs. causal Causal stance: • A + X (cause) → A + Y (effect) • Effect fully determined by cause => no need for goal Intentional and causal stances are formally equivalent • Causal stance is typical for mechanistic models • Intentional stance is typical for “mental” explanations Advantages of intentional stance • Can deal with more complex and intelligent agents • Does not require full information about causes Since end states are to some degree independent of initial states Interpretations The intentional stance can be interpreted metaphysically QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. Panpsychism: all phenomena have “mindlike qualities” • E.g. particles have rudimentary “consciousness” (Chalmers) Animism: all phenomena are “sentient” agents In fact: interpretations are a question of personal preference • they are all formally equivalent, even including the Newtonian interpretation QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. Challenges Intentional agents: typically living organisms or people Basic value = maximizing fitness Challenge = condition that potentially elicits action from the agent • because performing that action may lead to a fitness increase • at least relative to not performing the action Challenges are intrinsically meaningful conditions Cognition Tackling problems (complex challenges) requires • Selecting which challenge(s) to take on • Selecting which actions to perform for a given challenge Intelligence = ability to make good selections Knowledge = interiorized decision rules: • Anticipate challenge: if X, expect Y X→Y • Choose action: if Y, do Z Y→Z if X, do Z X→Z Planning: • Make inference: Challenge Types Positive: opportunity to increase fitness Negative: danger of losing fitness Expected: goals, threats (“anti-goals”) Unexpected: diversions, disturbances, affordances Perceived: prospect As yet invisible: mystery Course of Action Intended/anticipated sequence of actions • from present state to present goal Will need correction because of diversions • Disturbances → counteract • Affordances → exploit • Neutral diversions → change course Course of action Without diversions Course of action with diversions With diversions Prospect and Mystery The course of action (path ahead) is only partly anticipatable Prospect (perceived challenges) is always mixed up with mystery (as yet invisible challenges) Prospect and Mystery prospect mystery Course of action ? prospect mystery prospect agent Stigmergy Stigmergy = stimulation of actions by the results of actions • Primitive mechanism of coordination between actions/agents Agent A performs action: A+X→A+Y • X = initial challenge that elicits action • Y = result, “trace” left by the action There is stigmergy if Y too is a challenge • for the same or for another agent • in that case, Y will trigger a subsequent action • E.g. A’ + Y → A’ + Z Propagation of challenges Stigmergy => (branching) chain of challenges producing new challenges • E.g. A + X → A + Y, A’ + Y → A’ + Z , ... Z U A” X Y A A’ S A’ A A”’ A W A’ V Example: building a house tubing plumbers builders foundations carpente rs electricia walls windows ns plasterer s painters plastered walls finished house electricity walls + carpenters → house with windows (+ carpenters) house with windows + electricians → house with electricity (+ electricians) Example: office organization Coordination Actions are coordinated when • There is minimal friction • Overall loss of fitness because of interaction E.g. conflict, obstruction There is maximal synergy Overall gain in fitness because of interaction E.g. cooperation, complementarity Coordinated actions/agents can achieve much more together than alone Aspects of coordination Alignment • Actions should aim at the same targets Division of labor (parallel, simultaneous) • Actions should be performed by most competent agents Workflow (sequential) • Actions should follow each other efficiently Aggregation • Results of actions should be integrated into coherent whole Alignment Actions pointing in opposite directions obstruct each other • conflict, friction Actions pointing towards the same target reinforce each other • cooperation, synergy Parallel and Sequential Coordination laying electricity roofing parallel Qu i c k Ti m e ™ a n d a No n e d e c o m p re s s o r a re n e e d e d to s e e th i s p i c tu re . plumbing plastering painting sequential Self-organization Variation and natural selection → increase in fitness • Decrease in friction • Increase in synergy • Emergence of coordination between actions/agents Coordinated group of agents = system or organization, e.g. • • agents = atoms system = molecule agents = cells organism system = multicellular System as Network of Actions/Agents E f k l c e j h I O a i d g b S Some Ethical Imperatives Fundamental Value: increase fitness for all agents • By stimulating their individual development • By promoting the coordination of their actions, more specifically: Maximize synergy/cooperation • Promote complementarity / diversity Minimize friction/conflict • Prevent “free riders” Facilitate self-organization Facilitators of self-organization Increased variation / diversity • “order from noise” Easier propagation • More alignment → more pressure to align Stigmergic medium • Registers and broadcasts challenges (e.g. Wikipedia) Hebbian learning • Synergetic connections between actions/agents are reinforced Become easier to use next time Practical applications Self-organizing technologies • Artificial agents, action rules, medium • E.g. computer simulations, self-configuring engineering systems, networks of mobile sensors… Mobilization systems • Produce motivating challenges for individuals Using flow and other criteria: clear goals, immediate feedback, challenges adapted to abilities, variation in challenges, … Minimize boredom, anxiety, confusion, procrastination… • Facilitate coordination E.g. via alignment of goals and terminology, stigmergy and propagation of challenges Conclusion: benefits of action ontology • Generalization of Newtonian ontology • Transcendence of mind-matter dualism • Explanation for emergence, goal-directedness, evolution… • A very simple and practical philosophy • Foundations for metaphysics, epistemology and ethics • A framework for transdisciplinary unification • A methodology for tackling complex problems • A basis for building meaningful narratives • Thus, bridging the gap between the “two cultures”