View
236
Download
0
Category
Tags:
Preview:
Citation preview
EE141
EE690 EE690 Design of Design of Embodied IntelligenceEmbodied Intelligence
Janusz StarzykJanusz StarzykEECS, Ohio UniversityEECS, Ohio University
EE141
“…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. His family roots are in Kolomyje – so he once told “as with all bright people, my roots are in Poland”
“…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. He co-founded Palm Computing and Handspring Inc.
IntelligenceIntelligence
AI’s holy grailFrom Pattie Maes MIT Media Lab
EE141
Traditional Artificial Intelligence Embodied Intelligence (EI) Challenges of EI
We need to know how We need means to implement it We need resources to build and sustain its
operation Promises of EI
To economy To society
OutlineOutline
EE141
IntelligenceIntelligence
From http://www.indiana.edu/~intell/map.shtml
Mainstream Science on Intelligence December 13, 1994: An Editorial With 52 Signatories, History, and Bibliography 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.
EE141
Various Definitions of Intelligence 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.
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.
EE141
Traditional AITraditional AI Embodied Intelligence Embodied Intelligence Abstract 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”
Embodiment knowledge is implicit in the
fact that we have a body– embodiment is a
foundation for brain development
Intelligence develops through interaction with environment Situated in a specific
environment Environment is its best
model
EE141
Design principles of intelligent systemsDesign principles of intelligent systemsfrom Rolf Pfeifer “Understanding of Intelligence”, 1999
Interaction with complex environment
cheap design ecological balance redundancy principle asynchronous parallel, loosely
coupled processes sensory-motor
coordination value principle
Agent
Drawing by Ciarán O’Leary- Dublin Institute of Technology
EE141
Embodied Intelligence 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, anxiety or scarcity of resources– EI learns so it must have associative self-organizing memory– Knowledge is acquired by EI
EE141
Embodiment of MindEmbodiment of Mind
Definition Embodiment of a mind
is a mechanism
under the control of the intelligence core
that contains
sensors and actuators connected to the core through communication channels.
EnvironmentEnvironment
Intelligence core
Embodiment
Sensors
Actuators
EE141
Embodiment of MindEmbodiment of Mind
Necessary for development of intelligence
Hosts brain’s interfaces that interact with environment
Not necessarily constant or in the form of a physical body
Boundary transforms modifying brain’s self-determination EnvironmentEnvironment
Intelligence core
Embodiment
Sensors
Actuators
EE141
Brain learns own body’s dynamic Self-awareness results from
identification with own embodiment Embodiment can be extended by
using tools and machines Successful operation depends on
correct perception of environment and own embodiment
Embodiment of MindEmbodiment of Mind
EE141
Broca’sarea
Parsopercularis
Motor cortex Somatosensory cortex
Sensory associativecortex
PrimaryAuditory cortex
Wernicke’sarea
Visual associativecortex
Visualcortex
Brain OrganizationBrain Organization
While we learn its functionscan we emulate its operation?
EE141
How can we design intelligence?How can we design intelligence?
We need to know how We need means to
implement it We need resources to
build and sustain its operation
EE141
Requirements for Embodied IntelligenceRequirements for Embodied Intelligence
State oriented Learns spatio-temporal patterns
Situated in time and space
Learning Perpetual learning
Screening for novelty
Value driven Goal creation
Competing goals Emergence
Artificial evolution
Self-organization
EE141
INPUT OUTPUT
Simulation or
Real-World System
TaskEnvironment
Agent Architecture
Long-term Memory
Short-term Memory
Reason
ActPerceive
RETRIEVAL LEARNING
EI Interaction with EnvironmentEI Interaction with Environment
From Randolph M. Jones, P : www.soartech.com
EE141
Challenges of Embodied IntelligenceChallenges 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 reinforcement interfaces Energy, temperature, pressure, acceleration level Teacher input
Development of motor interfaces Arms, legs, fingers, eye movement
EE141
Challenges of Embodied Intelligence (cont.)Challenges of Embodied Intelligence (cont.) Finding algorithmic solutions for
Association, memory, sequence learning, invariance building, representation, anticipation, value learning (pain reduction), 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
V. Mountcastle argues that all regions of the brain perform the same computational algorithm
Groups of neurons (minicolumns) have the same structure and are connected in a pseudorandom way
Minicolumns organized in macrocolumns
VB Mountcastle (2003). Introduction [to a special issue of Cerebral Cortex on columns]. Cerebral Cortex, 13, 2-4.
Human Intelligence - Uniform StructureHuman Intelligence - Uniform Structure
EE141
Selective Processing in Minicolumns Selective Processing in Minicolumns
Sensory inputs are represented by gradually more abstract features in the sensory hierarchy
Use “sameness principle” of the observed objects to detect and learn feature invariance
Learn to store temporal sequences Use random wiring to preselect sensory
features Use feedback for input prediction and screen
input information for novelty Use redundant structures of sparsely connected
processing elements
EE141
Interactions between MinicolumnsInteractions between Minicolumns
Positive Reinforcement
Negative Reinforcement
Sensory Inputs Motor
Outputs
Valuecenter
Sensory processing
Motor processing
Pleasure Pain
Learn Control
Associate
Direct
EE141
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 sensory 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 receive inputs from reinforcement neurons on the lower level receive inputs from sensory neurons provide inputs to motor neurons initiate learning and force exploration
Functions of Neurons in MinicolumnsFunctions of Neurons in Minicolumns
EE141
Selective Sensory-Motor PathwaysSelective Sensory-Motor Pathways
Hierarchical sensory and motor pathways Branching off from more general to more specific Easy to expand to higher levels Gradual loss of interconnect plasticity towards inputs
inputs
internal representations
EE141
Environment……………... … …
Sensory neurons in a minicolumn
Selective Sensory-Motor PathwaysSelective Sensory-Motor PathwaysIncreasing connection’s plasticity
106 neurons
1011 neurons
10 neurons
Sensory pathway
10 neurons
104 neuronsActivation pathway
EE141
General characteristics: Hierarchical structure Minicolumn processing Spatial and temporal association
via interconnections and dual neurons
Feedback connections Selective adaptation
Function: Invariant representation Anticipation Screening for novelty Value learning
... …
... …
... …
Organization of Sensory-motor PathwaysOrganization of Sensory-motor Pathways
EE141
Sensory-motor CoordinationSensory-motor Coordination
R: representationE: expectationA: associationD: directionP: planning Environment
…
…
…
…
D
R
E
A
Sensor path Motor path
Increasing connection’s adaptability’
Goal creation &Value system
EE141
Goal Driven BehaviorGoal Driven Behavior
Goal driven behavior is one of the required elements of intelligence
Perceptions and actions are activated selectively to serve the machine’s objectives
In the existing EI models, the goal is defined by designers and is given to the learning agent
Humans and animals create their own goals
The goal creation may be one of the most important elements of EI mechanism
EE141
Goal Creation Goal Creation
Goal creation ability is essential for developing intelligence
We will create goals based on simple structures interacting with sensory and motor pathways
Goals must be built and understood in a similar way to building perceptions
More complex goals can be understood only if representations for these goals are build
Goal creation should result from EI interaction with environment, by perceiving successes or failures of its actions
EE141
Pain-center and Goal CreationPain-center and Goal Creation
Simple Mechanism Creates hierarchy of
values Leads to formulation of
complex goals Reinforcement
neurotransmitter:• Pain increase• Pain decrease
Forces exploration
+
-
Environment
Sensor
MotorPain level
Dual pain levelPain increase
Pain decrease
(-)
(+)
Excitation
(-)
(-)
(+)
(+)
EE141
Abstract Goal CreationAbstract Goal Creation
The goal is to reduce the primitive pain level Abstract goals are created to satisfy the primitive goals
Expectation
AssociationInhibitionReinforcementConnectionPlanning
- +
PainDual pain
Food
refrigerator
- +
Stomach
Abstract pain(Delayed memory of pain)
“food” becomes a sensory input to
abstract pain center
Sensory pathway(perception, sense)
Motor pathway(action, reaction)
Primitive Level
Level I
Level II
Eat
Open
EE141
The Three Pathways CombinedThe Three Pathways Combined Goal creation, sensory
and motor pathways interact on different hierarchy levels
Pain driven goal creation sets goal priorities
The tree pathways emerge together from interaction with the environment
Pain tree IPain tree IIMotor pathwaySensory pathwayPain center to motor Sensor to motor Sensor to pain center
EE141
How can we design intelligence?How can we design intelligence?
We need to know how We need means to
implement it Hardware self-organization Sparsity of interconnect Functionality requirements EI Design Efforts
We need resources to build and sustain its operation
EE141
We need means to implement itWe need means to implement it
What is needed is a self-organizing hardware Saves design work Increases hardware reliability Improves production yield Lowers design cost
Basic mechanism for self-organized hardware is pruning of random interconnections Nature does it
Activity triggered “growth” of new neurons Spare neurons used in the later stages of learning
EE141
Brain is self-organizing and sparse
Human Brain Human Brain at Birthat Birth 6 Years Old6 Years Old 14 Years Old14 Years Old
Rethinking the Brain, Families and Work Institute, Rima Shore, 1997.
We need means to implement itWe need means to implement it
EE141
Thompson, R. A., & Nelson, C. A. (2001). Developmental science and the media: Early brain development. American Psychologist, 56(1), 5-15.
Synaptic Density over the Lifespan Synaptic Density over the Lifespan
EE141
Sparse Connectivity Sparse Connectivity The brain is sparsely connected. A neuron in cortex may have on the order of
100,000 synapses. There are more than 1010 neurons in the brain. Fractional connectivity is very low: 0.001%.
Implications: Connections are expensive biologically. Short local connections are cheaper than
long ones.
EE141
Feature extraction – using WTA Generalization – using overlapping regions
… …
… …h+1
s
h
winner
…… … …
r-lower level region (single WTA in each)
Hierarchical Self-organizing Hierarchical Self-organizing Learning Memory (HSOLM)Learning Memory (HSOLM)
EE141
Hierarchical Self-organizing Learning MemoryHierarchical Self-organizing Learning Memory 3-layer R-net in hierarchical structure Reduced connection density The winner pathway replaces the WTA
Established using feed-back and local competition
…
…
…
… …
winner
Input pattern
h+1
s2
h
s1
number of output links from the lower level nodes
)),(... 2, 1,( 213
1 sshho nnninl
EE141
Wiring in SOLARWiring in SOLAR
Initial wiring and final wiring selection for credit card approval problem
EE141
IR
tim
Excitationlink
1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4
miss
sMUXTP MUX MUX MUX
mit
o1 2 3 4
MUX
momn1 n2 n3
PCN
LFN
LN
Multiple winnerDetection Neuron
FromESN
PreductionNeuron
PredictionMatchingNeuron
Sequence Learning Mechanism
EE141
Software or hardware?Software or hardware?
Sequential Error prone Require programming Low cost Well developed
programming methods
Concurrent Robust Require design Significant cost Hardware prototypes
hard to build
Software Hardware
EE141
Future software/hardware capabilitiesFuture software/hardware capabilities
Human brain complexity
2005 2010 2015 2020 2025 2030 2035 204010
4
105
106
107
108
109
1010
1011
Year
Num
ber
of n
euro
ns
Software Simulation (PC based) Hardware approach (FPGA)
Analog VLSI
100 FPGAs
10 FPGAs
EE141
How can we design intelligence?How can we design intelligence?
We need to know how We need means to
implement it We need resources to
build and sustain its operation
EE141
Major Intelligent Systems Design EffortsMajor Intelligent Systems Design Efforts GOMS - Goals, Operators, Methods, and Selection Rules
Proposed by Card, Moran, and Newell (1983), for modeling and describing human task performance
SOAR - State, Operator, Application, Result Based on the book by Newell, A. (1990), Unified Theories of Cognition.
ACT-R (Adaptive Control of Thought - Rational ) Described by Anderson, J. (1993), Rules of the Mind
Artificial General Intelligence – Adaptive A.I. Inc. http://adaptiveai.com/
Hierarchical Temporal Memory – Numenta Inc.
Irvine Sensors Corporation (Costa Mesa, CA)
3DANN
“Brain On Silicon” will not be just a science fiction!
http://www.numenta.com/
EE141
GOMS - Goals, Operators, Methods, and Selection Rules Inspired by psychology of immediate action Explicit hierarchical task decompositions Explicit pairing of goals with tasks Belief, Goal and Plan representation and
selection
Primary Engineering Activities Cognitive task analysis Design of goal hierarchy Definition of plans and selection rules Definition of primitive operators and resource constraints
EI Design EffortsEI Design Efforts
Proposed by Card, Moran, and Newell (1983),
for modeling and describing human task performance From slide by Dan Glaser
EE141
SOAR - State, Operator, Application, Result Problem space and physical symbols Minimal set of rules to support intelligent behavior Belief, Goal and Plan representation and selection
Primary Engineering Activities Cognitive task analysis Identification of primitive operations Design of belief and goal hierarchies Design of situation interpretation hierarchies Interactions between knowledge units
http://sitemaker.umich.edu/soar
Newell, A. 1990. Unified Theories of Cognition. Cambridge, Massachusetts: Harvard University Press.
EI Design EffortsEI Design Efforts
EE141
ACT-R (Adaptive Control of Thought - Rational ): a cognitive architecture inspired by psychological models of
memory, skills, and learning Optimization-oriented integrated memory, action, and learning Procedural knowledge units with explicit retrieval Uses if then production rules Belief, Goal and Plan representation and selection
Primary Engineering Activities Cognitive task analysis Identification of procedural and declarative knowledge Design of semantic associations and retrievable goal hierarchies Identification of primitive operations with logical and associative context
Anderson, John. Rules of the Mind, Lawrence Erlbaum Associates, Hillsdale, NJ (1993).
EI Design EffortsEI Design Efforts
EE141
EI Design EffortsEI Design Efforts Hierarchical Temporal Memory – Numenta Inc.
Uses hierarchy of spatio-temporal associations Discover causes in the world Learns complex goal-oriented behavior Makes predictions while learning Combines unsupervised and supervised learning to make associations Aims at exceeding human level cognition
Primary Engineering Activities Cognitive structure analysis Pattern learning in hierarchical memories Temporal sequence learning mechanism Goal creation
http://www.numenta.com/
Die Stack
Irvine Sensors Corporation (Costa Mesa, CA)
EE141
EI Design EffortsEI Design Efforts Artificial General Intelligence – Adaptive A.I. Inc.
Has general cognitive ability Acquires knowledge and skills Uses bidirectional, real-time interaction Selects information and uses an adaptive attention Unsupervised and self-supervised learning Adaptive and self-organizing data structure Processes dynamic patterns
Primary Engineering Activities Explicitly engineering functionality Conceptual design General proof of concept Animal level cognition
http://adaptiveai.com/
Irvine Sensors Corporation (Costa Mesa, CA)
3DANN
“Brain On Silicon” will not be just a science fiction!
EE141
How can we design intelligence?How can we design intelligence?
We need to know how We need means to
implement it We need resources to
build and sustain its operation
EE141From Ray Kurzwail, The Singularity Summit at Stanford, May 13, 2006
Resources – Evolution of ElectronicsResources – Evolution of Electronics
EE141From Ray Kurzwail, The Singularity Summit at Stanford, May 13, 2006
Clock Speed (doubles every 2.7 years)
EE141
Doubling (or Halving) timesDoubling (or Halving) times
Dynamic RAM Memory “Half Pitch” Feature Size 5.4 years Dynamic RAM Memory (bits per dollar) 1.5 years Average Transistor Price 1.6 years Microprocessor Cost per Transistor Cycle 1.1 years Total Bits Shipped 1.1 years Processor Performance in MIPS 1.8 years Transistors in Intel Microprocessors 2.0 years Microprocessor Clock Speed 2.7 years
From Ray Kurzwail, The Singularity Summit at Stanford, May 13, 2006
EE141
0%
20%
40%
60%
80%
100%
1999
2002
2005
2008
2011
2014
% Area Memory
% Area ReusedLogic
% Area New Logic
Percent of die area that must be occupied by memory to maintain SOC design productivity
Design Productivity Gap Design Productivity Gap Low-Value Designs? Low-Value Designs?
Source = Japanese system-LSI industry
EE141
2002 2010 2020 2030
Biomimetics and Bio-inspired SystemsBiomimetics and Bio-inspired SystemsImpact on Space Transportation, Space Science and Earth ScienceImpact on Space Transportation, Space Science and Earth Science
Mis
sio
n C
om
ple
xity
Biological Mimicking
Embryonics
Extremophiles
DNA Computing
Brain-like computing
Self Assembled Array
Artificial nanoporehigh resolution
Mars in situlife detector
Sensor Web
Biological nanoporelow resolution
Skin and Bone
Self healing structureand thermal protection
systems
Biologically inspired aero-space systems
Space Transportation
EE141
NanoEngineer-1 is an Open Source (GPL) project sponsored by Nanorex, Inc. It is an interactive 3D nanomechanical
engineering program
From Nanorex Inc.
Revolutions in electronics and computing will allow reconfigurable, autonomous,
"thinking" spacecraft
http://ipt.arc.nasa.gov/nanotechnology.html
Planetary Gear
Universal Joint
NanosystemsNanosystems
EE141
Promises of embodied intelligencePromises of embodied intelligence To society
Advanced use of technology– Robots– Tutors– Intelligent gadgets
Society of minds– Superhuman intelligence– Progress in science– Solution to societies’ ills
To industry Technological development New markets Economical growth
ISAC, a Two-Armed Humanoid RobotISAC, a Two-Armed Humanoid RobotVanderbilt UniversityVanderbilt University
EE141
Transhumanist Transhumanist valuesvalues
Nothing wrong about “tampering with nature” Individual choice in the use of enhancement technologies; Peace, international cooperation, anti-proliferation of WMDs Improving understanding (research and public debate; critical thinking;
open-mindedness; scientific progress; open discussion of the future) Getting smarter (individually; collectively; develop machine intelligence) Pragmatism; engineering and entrepreneur-spirit; “can-do” attitude Diversity (species, race, religious creed, sexual orientation, life style, etc.) Saving lives (life extension, anti-aging research, and cryonics)
Copyright Institute for Ethics and Emerging Technologies 2004-2005
EE141
Sounds like science fictionSounds like science fiction
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
EE141
Embodied Artificial IntelligenceEmbodied Artificial IntelligenceBased 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).
Recommended