Cognitive Neuroscience and Embodied Intelligence Emergence of the Embodied Intelligence How to Motivate a Machine ? Janusz Starzyk School of Electrical Engineering and Computer Science, Ohio University, USA www.ent.ohiou.edu/~starzyk EE141 Outline Traditional Artificial Intelligence Embodied Intelligence (EI) Embodiment of Mind EI Interaction with Environment How to Motivate a Machine Goal Creation Hierarchy GCS Experiment Motivated Learning Challenges of EI We need to know how to organize it We need means to implement it We need resources to build and sustain its operation Promises of EI To economy EE141 To society Intelligence AI’s holy grail From Pattie Maes MIT Media Lab “…Perhaps the last frontier of science – its ultimate challenge- is to understand the biological basis of consciousness and the mental process by which we perceive, act, learn and remember..” from Principles of Neural Science by E. R. Kandel et al. E. R. Kandel won Nobel Price in 2000 for his work on physiological basis of memory storage in neurons. “…The question of intelligence is the last great terrestrial frontier of science...” from Jeff Hawkins On Intelligence. Jeff Hawkins founded the Redwood Neuroscience Institute devoted to brain research EE141 Is what is intelligence? EE141 Various Definitions of Intelligence The American Heritage Dictionary: The capacity to acquire and apply knowledge. The faculty of thought and reason. Webster Dictionary: The act or state of knowing; the exercise of the understanding. The capacity to know or understand; readiness of comprehension; Wikipedia – The Free Encyclopedia: The capacity to reason, plan, solve problems, think abstractly, comprehend ideas and language, and learn. Kaplan & Sadock: The ability to learn new things, recall information, think rationally, apply knowledge and solve problems. On line dictionary dict.die.net The ability to comprehend; to understand and profit from experience The classical behavioral/biologists: The ability to adapt to new conditions and to successfully cope with life situations. Dr. C. George Boeree, professor in the Psychology Department at Shippensburg University: A person's capacity to (1) acquire knowledge (i.e. learn and understand), (2) apply knowledge (solve problems), and (3) engage in abstract reasoning. Stanford University Professor of Computer Science Dr. John McCarthy, a pioneer in AI: The computational part of the ability to achieve goals in the world. Scientists in Psychology: Ability to remember and use what one has learned, in order to solve problems, adapt to new situations, and understand and manipulate one’s reality. EE141 Intelligence Mainstream Science on Intelligence December 13, 1994: An Editorial With 52 Signatories, by Linda S. Gottfredson, University of Delaware Intelligence is a very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience. From http://www.indiana.edu/~intell/map.shtml EE141 Animals’ Intelligence Defining intelligence through humans is not appropriate to design intelligent machines: – Animals are intelligent too Dog IQ test: Dogs can learn 165 words (similar to 2 year olds) Average dog has the mental abilities of a 2-year-old child (or better) They would beat a 3- or 4-year-old in basic arithmetic, Dogs show some basic emotions, such as happiness, anger and disgust “The social life of dogs is very complex - more like human teenagers interested in who is moving up in the pack, who is sleeping with who etc,“ says professor Stanleay Coren from University of British Columbia Border collies, poodles, and german shepards are the smartest dogs EE141 Traditional AI Abstract intelligence Embodied Intelligence attempt to simulate “highest” human faculties: – language, discursive reason, mathematics, abstract problem solving Environment model Condition for problem solving in abstract way “brain in a vat” EE141 Embodiment knowledge is implicit in the fact that we have a body – embodiment supports brain development Intelligence develops through interaction with environment Situated in environment Environment is its best model Design principles of intelligent systems from Rolf Pfeifer “Understanding of Intelligence” Design principles synthetic methodology time perspectives emergence diversity/compliance frame-of-reference complete agent principle From: www.spectrum.ieee.org/.../biorobot11f-thumb.jpg EE141 Design principles of intelligent systems from Rolf Pfeifer “Understanding of Intelligence”, 1999 EE141 Interaction with complex environment ecological balance redundancy principle parallel, loosely coupled processes asynchronous sensory-motor coordination value principle cheap design Agent Drawing by Ciarán O’Leary- Dublin Institute of Technology The principle of “cheap design” intelligent agents: “cheap” exploitation of ecological niche economical (but redundant) exploitation of specific physical properties of interaction with real world EE141 Principle of “ecological balance” balance / task distribution between morphology neuronal processing (nervous system) materials environment balance in complexity given task environment match in complexity of sensory, motor, and neural system EE141 The redundancy principle redundancy prerequisite for adaptive behavior partial overlap of functionality in different subsystems sensory systems: different physical processes with “information overlap” EE141 Generation of sensory stimulation through interaction with environment multiple modalities constraints from morphology and materials generation of correlations through physical process basis for crossmodal associations EE141 The principle of sensory-motor coordination Holk Cruse self-structuring of sensory data through interaction with environment physical process — not „computational“ prerequisite for learning EE141 •no central control •only local neuronal communication •global communication through environment neuronal connections The principle of parallel, loosely coupled processes Intelligent behavior emergent from agent-environment interaction Large number of parallel, loosely coupled processes Asynchronous Coordinated through agent’s –sensory-motor system –neural system –interaction with environment EE141 So what is an Embodied Intelligence ? EE141 Embodied Intelligence Definition Embodied Intelligence (EI) is a mechanism that learns how to survive in a hostile environment – Mechanism: biological, mechanical or virtual agent with embodied sensors and actuators – EI acts on environment and perceives its actions – Environment hostility is persistent and stimulates EI to act – Hostility: direct aggression, pain, scarce resources, etc – EI learns so it must have associative self-organizing memory – Knowledge is acquired by EI EE141 Embodied Intelligence EI mimics biological intelligent systems, extracting general principles of intelligent behavior and applying them to design intelligent agents. Knowledge is not entered into such systems, but rather is a result of their successful interaction with the environment. Embodied intelligent systems adapt to unpredictable and dynamic situations in the environment by learning, which gives them a high degree of autonomy. Learning in such systems is incremental, with continuous prediction of the input associations based on the emerging models - only new information is registered in the memory. EE141 What is Embodiment of a Mind? EE141 Embodiment of a Mind Embodiment of a mind is a part of environment under control of the mind It contains intelligence core and sensory motor interfaces to interact with environment It is necessary for development of intelligence It is not necessarily constant or in the form of a physical body Boundary of embodiment transforms modifying brain’s self-determination EE141 Embodiment Sensors channel Environment Intelligence core Actuators channel Embodiment of Mind Brain learns own body’s dynamic Self-awareness is a result of identification with own embodiment Embodiment can be extended by using tools and machines Successful operation is a function of correct perception of environment and own embodiment EE141 Requirements for Embodied Intelligence State oriented Learns spatio-temporal patterns Situated in time and space Learning Perpetual learning Screening for novelty Value driven Pain detection Pain management Goal creation Competing goals Emergence artificial evolution self-organization EE141 EI Interaction with Environment Agent Architecture Reason Short-term Memory Perceive Act RETRIEVAL LEARNING Long-term Memory INPUT OUTPUT Task Environment Simulation or Real-World System EE141 From Randolph M. Jones, P : www.soartech.com Sensory Inputs Coding Kandel Fig. 30-1 Kandel Fig. 23-5 Richard Axel, 1995 Visual, auditory, olfactory, tactile, smell -> motor How do we process and represent sensory information? EE141 Hip Trunk Arm Hand Foot Face Tongue Larynx Challenges of Embodied Intelligence EE141 Challenges of Embodied Intelligence Development of sensory interfaces Active vision Speech processing Tactile, smell, taste, temperature, pressure sensing Additional sensing – Infrared, radar, lidar, ultrasound, GPS, etc. – Can too many senses be less useful? Development of pain sensors Energy, temperature, pressure, acceleration level Teacher input Development of motor interfaces Arms, legs, fingers, eye movement EE141 Embodiment Sensors Intelligence core Actuators Environment Challenges of Embodied Intelligence (cont.) Finding algorithmic solutions for Association, memory, sequence learning, invariance building, representation, anticipation, value learning, goal creation, planning Development of circuits for neural computing Determine organization of artificial minicolumn Self-organized hierarchy of minicolumns for sensing and motor control Self-organization of goal creation pathway EE141 Human Intelligence – Cortex Uniform Structure V. Mountcastle argues that all regions of the brain perform the same computational algorithm Groups of neurons (minicolumns) connected in a pseudorandom way Same structure Minicolumns organized in macrocolumns VB Mountcastle (2003). Introduction [to a special issue of Cerebral Cortex on columns]. Cerebral Cortex, 13, 2-4. EE141 Mini Columns “The basic unit of cortical operation is the minicolumn … It contains of the order of 80-100 neurons except in the primate striate cortex, where the number is more than doubled. The minicolumn measures of the order of 30-50 m in transverse diameter, separated from adjacent minicolumns by vertical, cell-sparse zones … The minicolumn is produced by the iterative division of a small number of progenitor cells in the neuroepithelium.” (Mountcastle) EE141 Copyright © 2006-2008, all rights reserved, Visualbiotech Artificial Minicolumn Organization Sensory neurons are responsible for representing environment receive inputs from sensors or sensory neurons on lower level represent the environment receive feedback input from motor and higher level neurons help to activate motor and reinforcement neurons Motor neurons are responsible for actions and skills are activated by reinforcement and sensory neurons activate actuators or provide an input to lower level motor neurons provide planning inputs to sensory neurons Reinforcement neurons are responsible for building the value system, goal creation, learning, and exploration EE141 receive inputs from reinforcement neurons on the lower level receive inputs from sensory neurons provide inputs to motor neurons initiate learning and force exploration Sensory-motor Coordination Sensory and motor pathways are interconnected on different hierarchical levels inputs outputs EE141 internal representations Sensory-motor Coordination Goal creation &Value system Sensor path Motor path D … … R … … A R: representation E: expectation A: association D: direction P: planning EE141 Environment ’ adaptability Increasing connection’s E How to Motivate a Machine ? A fundamental question is what motivates an agent to do anything, and in particular, to enhance its own complexity? What drives an agent to explore the environment and learn ways to effectively interact with it? EE141 How to Motivate a Machine ? Pfeifer claims that an agent’s motivation should emerge from the developmental process. He called this the “motivated complexity” principle. Chicken and egg problem? An agent must have a motivation to develop while motivation comes from development? Steels suggested equipping an agent with self-motivation. “Flow” experienced when people perform their expert activity well would motivate to accomplish even more complex tasks. Humans get internal reward for activities that are slightly above their level of development (Csikszentmihalyi). But what is the mechanism of “flow”? Oudeyer proposed an intrinsic motivation system. Motivation comes from a desire to minimize the prediction error. Similar to “artificial curiosity” presented by Schmidhuber. EE141 How to Motivate a Machine ? Can a machine that only implements externally given goals be intelligent? If not how these goals can be created? •There is a need for a hierarchy of values. •Not all values can be predetermined by the designer. •There is a need for motivation to act, explore and learn. •As machine makes new observations about the environment, there is a need to relate them to goals and values and create new goals and values. EE141 How to Motivate a Machine ? Although artificial curiosity helps to explore the environment, it leads to learning without a specific purpose. It may be compared to exploration in reinforcement learning. internal reward motivates the machine to perform exploration. Exploration is needed in order to learn and to model the environment. But is this mechanism the only motivation we need to develop intelligence? Can “flow” ideas explain goal oriented learning? Can we find a more efficient mechanism for learning? I suggest a simpler mechanism to motivate a machine. EE141 How to Motivate a Machine ? I suggest that it is the hostility of the environment, in the definition of EI that is the most effective motivational factor. It is the pain we receive that moves us. It is our intelligence determined to reduce this pain that motivates us to act, learn, and develop. Both are needed - hostility of the environment and intelligence that learns how to reduce the pain. Thus pain is good. Without pain there would be no intelligence. Without pain we would not be motivated to develop. Fig. englishteachermexico.wordpress.com/ EE141 Motivated Learning I suggest a goal-driven mechanism to motivate a machine to act, learn, and develop. A simple pain based goal creation system is explained next. It uses externally defined pain signals that are associated with primitive pains. Machine is rewarded for minimizing the primitive pain signals. Definition: Motivated learning (ML) is learning based on the self-organizing system of goal creation in embodied agent. Machine creates higher level (abstract) goals based on the primitive pain signals. It receives internal rewards for satisfying its goals (both primitive and abstract). ML applies to EI working in a hostile environment. EE141 Pain-center and Goal Creation for ML Simple Mechanism Creates hierarchy of Dual pain level Pain increase values (-) Leads to formulation of + (-) complex goals (+) Pain comparators release (+) reinforcement (+) neurotransmitter: (-) • Pain increase Pain level Pain decrease inhibitory • Pain decrease Excitation excitatory Forces exploration EE141 Sensor Environment Motor Pain-center and Goal Creation for ML expectation EE141 Dual pain memory Pain increase + (-) (-) (+) Pain detection - (+) Pain decrease Stimulation Pain detection/goal creation center Reinforcement neurotransmitter Sensory neuron Motor neuron Sensor activation Missing objects Motor Abstract Goal Creation for ML The goal is to reduce the primitive pain level Abstract goals are created if they satisfy the primitive goals Sensory pathway (perception, sense) Motor pathway (action, reaction) refrigerator Open - + food”becomes a “ sensory input to abstract pain center Abstract pain (Delayed memory of pain) Food Eat - Association Inhibition Reinforcement Connection Planning Expectation EE141 Level II Level I + Dual pain Pain Primitive Level Stomach Abstract Goal Hierarchy Sensory pathway (perception, sense) Motor pathway (action, reaction) Job Hierarchy of abstract goals is created if they satisfy the primitive goals - Activation Stimulation Inhibition Reinforcement Echo Need Expectation EE141 Sugar level Spend Level II Eat Level I + Food - Level III + Money - Work + Primitive Level The Three Pathways Combined Goal creation, sensory and motor pathways interact on different hierarchy levels Pain driven goal creation sets goal priorities Pain tree I Pain tree II Motor pathway Sensory pathway Pain center to motor Sensor to motor Sensor to pain center EE141 EI Interaction with Environment EI Architecture Pain Perceive Goal Creation Competing goals Act Planning INPUT OUTPUT Task Environment Simulation or Real-World System EE141 EI machine interacts with environment using its three pathways Goal Creation Experiment in ML PAIR # SENSORY MOTOR INCREASES DECREASES 1 Food Eat sugar level food supplies 8 Grocery Buy food supplies money at hand 15 Bank Withdraw money at hand spending limits 22 Office Work spending limits job opportunities 29 School Study job opportunities - Sensory-motor pairs and their effect on the environment EE141 Goal Creation Experiment in ML WTA WTA Mk Pk Sk 1 1 Gk Pl Bk S P B G Goal Creating Neural Network EE141 M Goal Creation Experiment in ML 10 UA 10 1 S2 Ps B wBP M2 1 G wPG P Trainable connections between pain, bias, and goal neurons EE141 Goal Creation Experiment in ML Pain Primitive Hunger 1 Pain 0 0 200 300 400 Lack of Food 500 600 100 200 300 400 Empty Gorcery 500 600 100 200 300 400 Discrete time 500 600 0.5 0 0 Pain 100 0.5 0 0 Pain signals in CGS simulation EE141 Goal Creation Experiment in ML Goal Scatter Plot 40 35 30 Goal ID 25 20 15 10 5 0 0 100 200 300 400 Discrete time 500 600 Action scatters in 5 CGS simulations EE141 Goal Creation Experiment in ML Pain Pain Pain Pain Pain Primitive Hunger 0.5 0 0.2 0.1 0 0.2 0.1 0 0.2 0.1 0 0.1 0.05 0 0 100 200 300 Lack of Food 400 500 600 0 100 200 300 Empty Gorcery 400 500 600 0 100 200 300 Lack of Money 400 500 600 0 100 200 300 400 Lack of JobOpportunitites 500 600 0 100 200 500 600 300 Discrete time 400 The average pain signals in 100 CGS simulations EE141 Goal Creation Experiment in ML Comparison between GCS and RL EE141 Compare RL (TDF) and ML (GCS) Mean primitive pain Pp value as a function of the number of iterations: - green line for TDF -blue line for GCS. Primitive pain ratio with pain threshold 0.1 EE141 Compare RL (TDF) and ML (GCS) Comparison of execution time on log-log scale TD-Falcon green GCS blue Combined efficiency of GCS 1000 better than TDF Problem solved Conclusion: embodied intelligence, with motivated learning based on goal creation system, effectively integrates environment modeling and decision making – thus it is poised to cross the chasm EE141 Reinforcement Learning Single value function Measurable rewards Can be optimized Predictable Objectives set by designer Maximizes the reward Motivated Learning One for each goal Learning effort increases with complexity Always active EE141 Internal rewards Cannot be optimized Potentially unstable Multiple value functions Unpredictable Sets its own objectives Solves minimax problem Always stable Learns better in complex environment than RL Acts when needed How can we make human level intelligence? We need to know how We need means to implement it We need resources to build and sustain its operation EE141 Resources – Evolution of Electronics EE141 From Ray Kurzwail, The Singularity Summit at Stanford, May 13, 2006 EE141 By Gordon E. Moore EE141 Clock Speed (doubles every 2.7 years) EE141 From Ray Kurzwail, The Singularity Summit at Stanford, May 13, 2006 Doubling (or Halving) times EE141 Dynamic RAM Memory “Half Pitch” Feature Size Dynamic RAM Memory (bits per dollar) Average Transistor Price 5.4 years 1.5 years 1.6 years Microprocessor Cost per Transistor Cycle Total Bits Shipped Processor Performance in MIPS Transistors in Intel Microprocessors Microprocessor Clock Speed 1.1 years 1.1 years 1.8 years 2.0 years 2.7 years From Ray Kurzwail, The Singularity Summit at Stanford, May 13, 2006 EE141 From Ray Kurzwail, The Singularity Summit at Stanford, May 13, 2006 EE141 From Hans Moravec, Robot, 1999 Software or hardware? Software Sequential Error prone Require programming Low cost Well developed programming methods EE141 Hardware Concurrent Robust Require design Significant cost Hardware prototypes hard to build Future software/hardware capabilities 11 10 10 10 g alo n A SI L V 9 Number of neurons 10 (F ch a ro pp a are w d r Ha 8 10 7 10 re tw a Sof 6 10 io u lat m i S Human brain complexity A) G P d) ase b C n (P 5 10 4 10 2005 2010 2015 2020 2025 Year EE141 2030 2035 2040 Why should we care? EE141 Source: SEMATECH Design Productivity Gap Low-Value Designs? Percent of die area that must be occupied by memory to maintain SOC design productivity 100% 80% 60% % Area Memory 40% % Area Reused Logic 20% % Area New Logic 19 99 20 02 20 05 20 08 20 11 20 14 0% Source = Japanese system-LSI industry EE141 Self-Organizing Learning Arrays SOLAR * Self-organization * Sparse and local interconnections * Dynamically reconfigurable * Online data-driven learning Integrated circuits connect transistors into a system -millions of transistors easily assembled -first 50 years of microelectronic revolution Self-organizing arrays connect processors into a system -millions of processors easily assembled -next 50 years of microelectronic revolution EE141 Promises of embodied intelligence To society Advanced use of technology – Robots – Tutors – Intelligent gadgets Intelligence age follows – Industrial age – Technological age – Information age Society of minds – Superhuman intelligence – Progress in science – Solution to societies’ ills To industry Technological development New markets Economical growth EE141 ISAC, a Two-Armed Humanoid Robot Vanderbilt University Biomimetics and Bio-inspired Systems Mission Complexity Impact on Space Transportation, Space Science and Earth Science 2002 2010 2020 2030 Embryonics Self Assembled Array Space Transportation Biologically inspired aero-space systems Sensor Web Brain-like computing Extremophiles Mars in situ life detector Skin and Bone Self healing structure and thermal protection systems EE141 Biological nanopore low resolution Artificial nanopore high resolution DNA Computing Biological Mimicking Sounds like science fiction EE141 If you’re trying to look far ahead, and what you see seems like science fiction, it might be wrong. But if it doesn’t seem like science fiction, it’s definitely wrong. From presentation by Feresight Institute Embodied Artificial Intelligence Based on: [1] E. R. Kandel et al. Principles of Neural Science, McGraw-Hill/Appleton & Lange; 4 edition, 2000. [2] F. Inda, R. Pfeifer, L. Steels, Y. Kuniyoshi, “Embodied Artificial Intelligence,” International seminar, Germany, July 2003. [3] R. Chrisley, “Embodied artificial intelligence, ” Artificial Intelligence, vol. 149, pp.131-150, 2003. [4] R. Pfeifer and C. Scheier, Understanding Intelligence, MIT Press, Cambridge, MA, 1999. [5] R. A. Brooks, “Intelligence without reason,” In Proc. IJCAI-91. (1991) 569-595 . [6] R. A. Brooks, Flesh and Machines: How Robots Will Change Us, (Pantheon, 2002). [7] R. Kurzweil The Age of Spiritual Machines: When Computers Exceed Human Intelligence, (Penguin, 2000). EE141 Questions? EE141 EE141