Question? Leave a message!




Consciousness and Cognitive Systems

Consciousness and Cognitive Systems
Intelligent Control and Cognitive Systems brings you... Consciousness and Cognitive Systems Joanna J. Bryson University of Bath, United KingdomConsciousness Cognitive Systems Can an artificial cognitive system be • conscious Who cares • Why care • What is consciousness in the first place •KC Green, Gun Show, 513Cognitive Systems Philosophy Science fiction uses robots and aliens to • examine the human condition; the future to examine the present. AI does the same thing. • ... but, AI is also real. • Well, some of it is real. • Some of it is tangled with Sci Fi. •Roadmap for Conscious Machines 5. Executive 1. (1) Disembodied 6. Emotional 1. (0) Isolated 7. Selfconscious 1. Decontrolled 8. Empathic 2. Reactive 9. Social 3. Adaptive 10.Humanlike 4. Attentional 11.Super Conscious Arrabales et al 2009Roadmap for Conscious Machines 1. (1) Disembodied 1. (0) Isolated 1. Decontrolled 2. Reactive Sensing to action: intelligence 3. Adaptive 4. Attentional Unconsciousness is more conscious Arrabales et al 2009Roadmap for Conscious Machines 5. Executive multiple goals (unconscious 2) 6. Emotional “human like” 7. Selfconscious knows about self 8. Empathic knowledge (k) of others 9. Social k of other’s k of self 10.Humanlike use Interweb to extend mind 11.Super Conscious multiple streamsConsciousness = Like ME From an AI even Computer Science • perspective, many of these criteria are easy to achieve. E.g. perfect self knowledge. • Consciousness is easy but combinatorics is • hard – computational explanation for biological phenomenon of unconsciousness Bryson, Philosophy Magazine, 2007What’s Consciousness it is nor hand, nor foot, nor arm, nor SG5UT Robotic Arm face, nor any other part belonging to a man. Tad McGeer's passive dynamic walker Glenn Matsumura, Wired 2007 Chuck Rosenberg’s IT, 1997If this can be a hand... ...what could a mind be likeModelling Natural Intelligence One of the best ways • to understand how something works is to build it yourself. AI is used in scientific • modelling, but also in Philosophy. Dennett: “Intuition Pumps”Consciousness as per Dennett The term conscious is itself culturally • evolved. May not refer to any one psychological • phenomenon. Like light before modern physics. •Dennett vs. The Cartesian Theatre Consciousness cannot work by infinite • recursion. Must be composed of nonconscious • elements. Nothing inside you is conscious; you are. •Multiple Drafts / The Attentional Spotlight There are many candidate parallel • processes which could be conscious. Only one is – leaves trace in episodic • memory. Not necessarily determined in order, e.g. • if driving may ‘see’ something you hit only after you hear the bump. ∴ Not really conscious all the time •Syst em sof t ware ( 0 t h) Syst em sof t ware ( commercial processor) Periperhal Mot ion Vergence Ullmanesque Physical schema Saccades based st ereo visual rout ines based obj. recog. “Building VOR Smoot h pursuit Face popout s Face remembering Face recognit ion Brains Head/ body/ eye/ coord Head/ eye coord Gest ure recognit ion Facial gest ure recog. Body mot ion recog. for Own hand t racking Specif ic obj. recog. Generic object recog. Bodies”, Bring hands Hand Grasping, Bodybased met aphors midline linking t ransfer Brooks DOF reduct ion DOF reduct ion Bat t ing st at ic Stein (specific coords) (generic coords) object s Body st abilit y, Body+arm reaching Body mimicry (1993), leaning, rest ing Manipulat ion t urn t aking MIT AI Sound localizat ion Soundbased manip. Voice/ face assoc lab tech Sound/ mot ion correl Human voice ext ract ion Prot o language report Tone ident ificat ionVoice t urn t aking Visual imagery Symbolizat ion 1439. Ment al rehearsal Imaginat ion Mult ipledraf t s emergence Sept 1 Sept 1 Sept 1 Sept 1 Sept 1 1993 1994 1995 1996 1997Fill In and Confabulation Things like the driving story the fact we are never aware of our blind spot unless we really go out of the way to test for it indicate we cannot trust our intuitions about consciousness. Dennett Critics Some people really hate these ideas. • Chalmers is the main antiDennett • champion. Chalmers’ Hard Problem: Explaining • qualia. How do you know someone else sees • red the same way you doThe Zombie Problem (seriously) A standard problem in philosophy: how • would you tell if someone wasn’t conscious Dennett: the zombie idea is incoherent. • (likes Brooks, embodiment theory.) • Consciousness is what it’s like to act human. • There’s nothing else. • Critics: Dennett thinks we’re all zombies •Popular Theories of Consciousness Consciousness is selfawareness. • Consciousness requires language. • Consciousness is the root of ethical • obligation / a soul. Consciousness is a special pattern of • energy (Dahaene) Consciousness is a special level of • information integration (Tononi)What People Like in Consciousness Theories We’ll never understand consciousness. • We will understand it, but not in 100 years. • I have a quantitative, scientific measure of • consciousness, but it will take 60 years until we can check if I’m right (Tononi). Only humans are conscious. •Currently the most popular theory in Cognitive Systems Research is Barr’s Global Workspace Theory Upcoming slides by Murray ShanahanNeural Parallelism An animal’s nervous system is massively parallel Massive parallelism surely underpins human cognitive prowess So how are the massively parallel computational resources of an animal’s central nervous system harnessed for the benefit of that animal How can they orchestrate a coherent and flexible response to each novel situation Nature has solved this problem. How 22Global Workspace Architecture Parallel Unconscious Specialist Processes Global Workspace Multiple parallel specialist processes compete and cooperate for access to a global workspace If granted access to the global workspace, the information a process has to offer is broadcast back to the entire set of specialists 23Conscious vs NonConscious Global workspace theory (Baars) hypothesises that the mammalian brain instantiates such an architecture It also posits an empirical distinction between conscious and nonconscious information processing Information processing in the parallel specialists is nonconscious Only information that is broadcast is consciously processed 24Combining a GW with Internal Simulation Cortical area 1 Cortical area 2 Cortical area 3 GW Broadcast Selection / competition It’s possible to combine an internal sensorimotor loop with mechanisms for broadcast and competition, and thereby marry the simulation hypothesis with global workspace theory 25Remember / Revision Prescott after Brooks corrected Wooldridge xScience Evolution Selection requires variation – occurs • between existing options (and their combinations mutations). History matters – understanding it helps • explain what we think. Some combination of what works well • and what we were lucky someone thought of – culture.Combining a GW with Internal Simulation Cortical area 1 Cortical area 2 Cortical area 3 GW Broadcast Selection / competition It’s possible to combine an internal sensorimotor loop with mechanisms for broadcast and competition, and thereby marry the simulation hypothesis with global workspace theory 28A Biologically Nonimplausible Implementation Global neuronal workspace Built out of spiking neurons with transmission delays Cortical columns comprise 32 × 32 fully connected nets Workspace nodes comprise 16 × 16 topographically mapped regions Cortical columns trained to associate successively presented pairs of images Column 1 Column 3 using STDP Column 2 29Controlling a Robot The inner sensorimotor loop can be embedded in a larger system and used to control a robot This results in a form of “cognitivelyenhanced” action selection icing The implemented action selection architecture Is based on salience and winnertakesall Pretty much Imposes a veto at final motor output stage Maes nets Modulates salience as a result of internal simulation again. Releases veto when salience exceeds a threshold The parallelism of the GW architecture enables the inner loop to explore alternatives 30Search vs Time Combinatorics is the problem, search is the only • solution. The task of intelligence is to focus search. • Called bias (learning) or constraint (planning). • Most behaviour has no or little realtime search. • For natural intelligence, most focus evolves. • Physical/cognitive constraints limit search space. •Hypothesis Consciousness cognition are that mental stuff that takes time. (Treisman Gelade, Cognitive Psychology1980)Ex 1: find the green TEx 1: find the green T T T T T T T T T T T T T T T T T T T T T T T T T T T TEx 1: find the green TEx 1: find the green T L T L L T L T L L L T T T T T T T L L T L L T T L L T TTime Consciousness Sometimes time is determined by the • number of steps you need to do (e.g. counting to yourself, searching a screen.) But sometimes it seems to be determined • by something else…Learning and Time Lookingtime experiments rely • on reactiontime delay being indicative of surprise. Flattening of reaction times • correlates with failure to notice shift in reward Looking time schedule, but no impact on research e.g. performance (Rapp et al Santos, Spelke 1998).Allocating Time Attention I. Individuals allocate more time when less certain (Bryson 2009; 2010). II. Species allocate in response to niche e.g. tamarins insects (Hauser 1999). III.Species allocate inversely with age (Rapp et al 1998, Bryson 2009; 2010). IV.Individuals allocate inversely with urgency (Shadlen and Newsome, 1998; Bogacz et al., 2006).A Theory of Conscious Attention The basic function of conscious awareness • is to update important models (learn). Time is allocated in proportion to • uncertainty by inhibiting action. Not to choose immediate action • If new action is favoured due to model • updates, may affect immediate behaviour.Consciousness for AI Only need C if system learns, and learning • relies on a bottlenecked cognitive resource. In this case, allocating C to tasks you are • doing in proportion to how uncertain you are about them is a pretty good guess. Also attend to other novel / unpredicted • by your internal model events (deer in the headlights).Point of Intervention 1. Action selection as usual. 2. Inhibit action expression while selected action is in mind, update models. 3. If new action becomes more salient, insight. Flush plan start over. • 4. Update of models may not have immediate impact on behaviour.Conclusions Bryson 2011, 2012 The basic function of awareness is not to • choose actions, but to inhibit actions once selected and learn about their situation. A costly (in terms of time) allocation of • resources for learning, varies in application by species and by individual situation. Easy to build. •Are There Already Conscious AI Systems Charlie Kemp, GA Tech Andrea Thomaz, MIT“If the best the roboticists can hope for is the creation of some crude, cheesy, secondrate artificial consciousness, they still win.” D. C. Dennett (1994), “The Practical Requirements for Making a Conscious Robot”, Philosophical Transactions: Physical Sciences and Engineering, 349 p. 137 (133146).How does this relate to other theories of consciousnessRoadmap for Conscious Machines 5. Executive 1. (1) Disembodied 6. Emotional 1. (0) Isolated 7. Selfconscious 1. Decontrolled 8. Empathic 2. Reactive 9. Social 3. Adaptive 10.Humanlike 4. Attentional 11.Super Conscious Arrabales et al 2009Self Consciousness Consciousness of self: limited like all • consciousness to likely useful search space. Much facilitated in humans by language • instruction probably less in other species. Google Search treats its own pages like other’s: • selfawareness neither necessary nor sufficient for consciousness.Self Consciousness Consciousness of self: limited like all • consciousness to likely useful search space. Much facilitated in humans by language • instruction probably less in other species. Google Search treats its own pages like other’s: • selfawareness neither necessary nor sufficient for consciousness.Language Helps Symbolic representation allows more • compact and / or less emotionallysalient representations. Learn concepts from others; shared • consciousness of events (Dennett 2008). Not a prerequisite for this basic functional • component of action selection.Ethics Consciousness: culturallyevolved concept • of uncertain age and origin (Dennett, 2001). May refer to no single psychological • phenomenon. Ethics: Coevolve with social order. • Much relies on assigning responsibility: • covaries with but not determined by consciousness. If robots are conscious… …do we have to think harder about moralityGratuitous XKCD (Munroe 2012)ICCS Conclusions Learned about autonomous intelligence by • programming robots. Learned about interacting social intelligence • (a little) by programming ABM. Learned a marketable skill by programming • a game. Please teach me by filling in the unit review • form – we really do read the free text
sharer
Presentations
Free
Document Information
Category:
Presentations
User Name:
Dr.BenjaminClark
User Type:
Teacher
Country:
United States
Uploaded Date:
21-07-2017