Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
The Role of Cognition in Cognitive Systems
From Robots to Primatology Joanna J. Bryson
Artificial Models of Natural IntelligenceUniversity of Bath
Institute of Cognitive and Evolutionary Anthropology University of Oxford
http://www.cs.bath.ac.uk/ai/AmonI.html
Outline
• Introduction to Intelligence & Cognition
• Where do you put it in a Cognitive System? Behavior Oriented Design
• Primates & Cognition
Intelligence
• What matters is expressing the right behavior at the right time: action selection.
Tony J. Prescott, Joanna J. Bryson, and Anil K. Seth, “Introduction. Modelling Natural Action Selection”, Philosophical Transactions of the Royal Society -- B, Biology 362(1485): 1521–1529, Sept 2007.
Intelligence
• What matters is expressing the right behavior at the right time: action selection.
• Conventional AI planning searches for an action sequence, requires set of primitives.
Tony J. Prescott, Joanna J. Bryson, and Anil K. Seth, “Introduction. Modelling Natural Action Selection”, Philosophical Transactions of the Royal Society -- B, Biology 362(1485): 1521–1529, Sept 2007.
Intelligence
• What matters is expressing the right behavior at the right time: action selection.
• Conventional AI planning searches for an action sequence, requires set of primitives.
• Learning searches for the right parameter values, requires primitives and parameters.
• Evolution and development are learning.
Tony J. Prescott, Joanna J. Bryson, and Anil K. Seth, “Introduction. Modelling Natural Action Selection”, Philosophical Transactions of the Royal Society -- B, Biology 362(1485): 1521–1529, Sept 2007.
Combinatorics• If . . .
– an agent knows 100 actions (e.g. eat, drink, sleep, step, turn, lift, grasp, poke, flip...), and – it has a goal (e.g. go to Madagascar)
• Then . . .
– Finding a one-step plan may take 100 acts. – A two-step plan may take 1002 (10,000). – For unknown number of steps, may search forever, missing critical steps or sequence.
Intelligence & Design
Joanna J. Bryson and Lynn Andrea Stein, “Modularity and Design in Reactive Intelligence”, in The Seventeenth International Joint Conference on Artificial Intelligence (IJCAI), Seattle WA, pp. 1115–1120, Morgan Kaufmann, 2001.
Intelligence & Design• Combinatorics is the problem, search is the only
solution.
Joanna J. Bryson and Lynn Andrea Stein, “Modularity and Design in Reactive Intelligence”, in The Seventeenth International Joint Conference on Artificial Intelligence (IJCAI), Seattle WA, pp. 1115–1120, Morgan Kaufmann, 2001.
Intelligence & Design• Combinatorics is the problem, search is the only
solution.
• The task of intelligence is to focus search.
• Called bias (learning) or constraint (planning).
• Most behavior has no or little real-time search.
Joanna J. Bryson and Lynn Andrea Stein, “Modularity and Design in Reactive Intelligence”, in The Seventeenth International Joint Conference on Artificial Intelligence (IJCAI), Seattle WA, pp. 1115–1120, Morgan Kaufmann, 2001.
Intelligence & Design• Combinatorics is the problem, search is the only
solution.
• The task of intelligence is to focus search.
• Called bias (learning) or constraint (planning).
• Most behavior has no or little real-time search.
• For natural intelligence, most focus evolves.
Joanna J. Bryson and Lynn Andrea Stein, “Modularity and Design in Reactive Intelligence”, in The Seventeenth International Joint Conference on Artificial Intelligence (IJCAI), Seattle WA, pp. 1115–1120, Morgan Kaufmann, 2001.
Intelligence & Design• Combinatorics is the problem, search is the only
solution.
• The task of intelligence is to focus search.
• Called bias (learning) or constraint (planning).
• Most behavior has no or little real-time search.
• For natural intelligence, most focus evolves.
• For artificial intelligence, most focus designed.Joanna J. Bryson and Lynn Andrea Stein, “Modularity and Design in Reactive Intelligence”, in The Seventeenth International Joint Conference on Artificial Intelligence (IJCAI), Seattle WA, pp. 1115–1120, Morgan Kaufmann, 2001.
Cognition
Cognition
Definition:
Cognition is on-line (real-time) search.
Cognition
Definition:
Cognition is on-line (real-time) search.
Consequence:
Cognition is bad.
Cognition
Cognition
• Why is cognition / individual search bad?
• Slow
• Uncertain
Cognition
• Why is cognition / individual search bad?
• Slow
• Uncertain
• Unpopular in most species.
• Plants
• Protozoa
Only think when you don’t know what’s going on
• Cognition is costly.
• Time, errors metabolism.
• Value of investment can be estimated from own experience or mother’s (maternal effects).
Deary et al. (2004)(Schaie et al. 2004; Kotrschal
& Taborsky in prep.)Joanna J. Bryson, “Age-Related Inhibition and Learning Effects: Evidence from Transitive Performance”, in Proceedings of the 31st Annual Meeting of the Cognitive Science Society (CogSci 2009) pp. 3040–3045.
Outline
• Introduction to Intelligence & Cognition
• Where do you put it in a Cognitive System? Behavior Oriented Design
• Primates & Cognition
Architecture
• Where do you put the cognition?
• Really: How do you bias / constrain / focus cognition (learning, search) so it works?
Behavior Oriented Design
• All search (learning, planning) is done within modules with specialised representations.
• Specialized representations promote reliability of search; also determine decomposition.
• Modules provide perception, action, memory. Arbitration via hierarchical dynamic plans.
• Iterative / agile test & development cycle.
(Bryson 2001, 2003)
BOD Action SelectionParallel-rooted, Ordered, Slip-stack Hierarchical (POSH) action selection:
• Some things need to be checked at all times: drive collection.
• Some things only need considering in particular context: competences.
• Some things reliably follow from others: action patterns.
BOD Robot Example
• Behaviour Library — per platform.
• POSH plan — per “species” / goal set.
• Memory — per individual.
(ATAL 1997, PhD 2001)
Joanna J. Bryson “The Behavior-Oriented Design of Modular Agent Intelligence”, Agent Technologies, Infrastructures, Tools, and Applications for e-Services, R. Kowalszyk, J. P. Müller,H. Tianfield and R. Unland, eds., pp. 61–76, Springer, 2003.
Directioncurrentpreferred∗directions
narrow, has dir?, pick open dir
correct dir, lose dir, move, move view? ��� � ��
�
�
�� � �Action
Selection
directionwhich-directionsense-ring-mask
move, move view?
������������
P-Memory
sonar-historysonar-expect �� C-Sense
sensor-ring-vector
csense
��
csense�����
������
compound-sense����������
������������
Bump
∗bumps
reg bump,bumped
���������������������
bump-fuse��
� ���
��
� �Robot
sonar���������������
��������������infra-red
��
bumpers������������
�������������
bumpx, ynext∗
bump-fuse
��
DP-Map
∗landmarks pick near neighbor, pick further neighbor
untried near neighbor?, untried far neighbor? ��� � ��
�
�
�� � �Action
Selection
DP-Landx,yin-dirout-dir
in dp, entered dp��������������
����������������
E-Memory
∗directions∗times
done-that����
������
continue untriedkeep going�����������
�������������
� � � � ��
�
�
�� � � � �
Robot(and C-Sense)
csense, odometry
��
direction, time�������
����������
life (D)
talk [1/120 Hz]
(worth talking�)
speak
sense (C) [7 Hz]
bump (bumped�) yelp reg bump back off clear bump lose direction
look compound sense
walk (C)
halt (has direction�)
(move view ’blocked)
lose direction
start (has direction⊥) pick open dir
continue move narrow (move view ’clear) correct dir
wait snore sleep
walk (C)
halt (has direction�)
(move view ’blocked)
lose direction
cogitate route (C)
enter dp (in dp ⊥)
(entered dp⊥)
lose direction greet dp
leave dp (in dp �)
(entered dp�)
dismiss dp
pick direction (C)
look up
(untried near neighbor
�)
pick near neighbor
keep going
(continue untried�)
pick previous direction
desperate look up
(untried far neighbor
�)
pick further neighbor
start (has direction⊥) ask directions
continue move narrow (move view ’clear) correct dir
Statistical Testing of BOD Action Selection
Tests performed in Tyrellʼs (1993) “Simulated Environment”
0 1 2 3 4 5 6 7 8 92
4
6
8
10
12
14
(Sparse)Std (Sparse)Var1 (Sparse)Var2 (Sparse)Var3
Fitn
ess
life (D)
flee (C) (sniff predator t)
freeze (see predator t) (covered t) (hawk t) hold still
run away (see predator t) pick safe dir go fast
look observe predator
mate (C) (sniff mate t)
inseminate (courted mate here t) copulate
court (mate here t) strut
pursue pick dir mate go
triangulate (getting lost t) pick dir home go
home 1::5 (late t) (at home⊥) pick dir home go
check 1::5 look around
exploit (C) (day time t)
use resource (needed res avail t) exploit resource
leave pick dir go
sleep at home (at home t) (day time⊥) sleep
N NE E SE S SW W NW
UTReproduce
1.4
T U
Move ActionsMate
-0.08
Court
P. Mate Rand. Dir P. Den R. Den All Dirs
Clean Leavethis Sq
CleanSleep
Mate Court
ApproachMate
Explore For Mates
Explore
Sleep
ApproachP. Den
Approach R. Den
Sleepin Den Clean
Keep
DirtinessLow HealthNight Proxfrom DenDistance
-0.10
-0.05
-0.01
-0.05 -0.05-0.15
Courted Mate in Sq
Mate in SqReceptive
No Denin Sq
Den in Sq
No Den in Sq
in SqDen
-0.02
-0.02-0.25
-0.30-0.04
= small negative activation
= positive activation
= small positive activation
= zero activation
= large positive activation(1.0)
Joanna J. Bryson, “Hierarchy and Sequence vs. Full Parallelism in Action Selection”, Simulation of Adaptive Behavior 6, pp. 147–156. 2000.
BOD Monkey Science
Joanna J. Bryson and Jonathan C. S. Leong “Primate Errors in Transitive ‘Inference’” Animal Cognition, 10(1):1–15, January 2007.
� � ��
�
�
�� � �
ActionSelection
apparatustest-board
reward
find-color, reward-found, new-test,
no-test, finish-test, save-result, rewarded��
monkeyvisual-attention
hand
grasping, noises,grasp-seen
��
sequenceseq
sig-difweight-shift
make-choice,
learn-from-reward��
rule-learner*attendants*rule-seqs
current-focuscurrent-rule
target-chosen, focus-rule, pick-block,priority-focus, rules-from-reward����������������������
�������������������������
look-at��������������������
���������������������
(Animal Cog 2007)
0 50 100 150 200 250 300 350 400 450 5000
0.1
0.2
0.3
0.4
0.5 P1 P2a
P2bP2c
P3 T1 T2a T2
• If each agent has a 1% chance of discovering a skill (e.g. making cheese) in its lifetime and there are 4000 agents, probably some agents will know the skill.
• If it is easier to learn the skill from a knowledgeable agent than by discovery, then selective pressure for culture.
• Inclusive fitness c < b × r (Hamilton 1964; West et al 2007).
Combinatorics vs Culture
BOD Experiments in Social Learning in VR
Mark A. Wood and Joanna J. Bryson, “Skill Acquisition Through Program-Level Imitation in a Real-Time Domain”, IEEE Transactions on Systems, Man and Cybernetics Part B—Cybernetics, 37(2):272–285, April 2007.
Basic Result: Still intractable without an enormous amount of
prior information.
Extension of Roy 1999 (PhD) to realtime planning.
Fortunately, Priors Easy to Insert with BOD
Samuel J. Partington and Joanna J. Bryson, “The Behavior Oriented Design of an Unreal Tournament Character”, The Fifth International Working Conference on Intelligent Virtual Agents, pp. 466–477, Springer, 2005.Cyril Brom, Jakub Gemrot, Michal Bída, Ondrej Burkert, Sam J. Partington and Joanna J. Bryson, “POSH Tools for Game Agent Development by Students and Non-Programmers”, in The Nineth International Computer Games Conference: AI, Mobile, Educational and Serious Games, pp. 126–133, 2006.
IDEs for Dynamic
Plans
• Advanced BOD Environment.
• Ubiquitous robotics requires AI (or servicing AI) by graduates with second class honours.
Joanna J. Bryson, Tristan J. Caulfield and Jan Drugowitsch, “Integrating Life-Like Action Selection into Cycle-Based Agent Simulation Environments”, in Proceedings of Agent 2005:Generative Social Processes, Models, and Mechanisms. pp. 67–81, Argonne National Laboratory 2006.
IDEs for Dynamic
Plans
• Advanced BOD Environment.
• Ubiquitous robotics requires AI (or servicing AI) by graduates with second class honours.
This work was & is again being funded by
aerospace.
Joanna J. Bryson, Tristan J. Caulfield and Jan Drugowitsch, “Integrating Life-Like Action Selection into Cycle-Based Agent Simulation Environments”, in Proceedings of Agent 2005:Generative Social Processes, Models, and Mechanisms. pp. 67–81, Argonne National Laboratory 2006.
Outline
• Introduction to Intelligence & Cognition
• Where do you put it in a Cognitive System? Behavior Oriented Design
• Primates & Cognition
Why BOD Works• Modularity: problem spaces, combat
combinatorics, allow locally-optimal representations.
• Should use ordinary (OO) code (arbitrarily powerful but also access to primitives.)
• Hierarchical action selection for arbitration.
• Dedicated, high-frequency goal / attention switching, keeps hierarchical AS responsive.
• Agile development, refactoring (Beck 2000).
Subsumption (Brooks 1986)
Subsumption (Brooks 1986)• Emphasis on
sensing to action (via Augmented FSM).
Subsumption (Brooks 1986)• Emphasis on
sensing to action (via Augmented FSM).
• Very complicated, distributed arbitration.
Subsumption (Brooks 1986)• Emphasis on
sensing to action (via Augmented FSM).
• Very complicated, distributed arbitration.
• No learning.
Sept 1
1 9 9 3
Sept 1
1 9 9 4Sept 1
1 9 9 5
Sept 1
1 9 9 6
Sept 1
1 9 9 7
Syst em sof t ware ( 0 t h) Syst em sof t ware ( commercial processor)
Periperhal Mot ion
Saccades
VOR
Smoot h pursuit
Vergence
based st ereo
Ullman-esque
visual rout ines
Face pop-out s Face remembering Face recognit ion
Gest ure recognit ion Facial gest ure recog. Body mot ion recog.
Own hand t racking
Physical schema
based obj. recog.
Bring hands
midline
Hand
linking
Grasping,
& t ransf er
Specif ic obj. recog. Generic object recog.
Body-based met aphors
DOF reduct ion
( specif ic coords)
DOF reduct ion
( generic coords)Bat t ing st at ic
ob ject s
Head/ eye coord
Body st abilit y ,
leaning, rest ing
Head/ body / ey e/ coord
Body+arm reaching Body mimicry
Manipulat ion t urn t aking
Sound localizat ion
Sound/ mot ion correl Human voice ext ract ion
Sound-based manip. Voice/ f ace assoc
Voice t urn t aking
Prot o language
Visual imagery Symbolizat ion
Imaginat ionMent al rehearsal
Mult iple-draf t s emergence
Tone ident if icat ion
“Building Brains
for Bodies”, Brooks & Stein (1993), MIT AI lab tech report 1439.
Sept 1
1 9 9 3
Sept 1
1 9 9 4Sept 1
1 9 9 5
Sept 1
1 9 9 6
Sept 1
1 9 9 7
Syst em sof t ware ( 0 t h) Syst em sof t ware ( commercial processor)
Periperhal Mot ion
Saccades
VOR
Smoot h pursuit
Vergence
based st ereo
Ullman-esque
visual rout ines
Face pop-out s Face remembering Face recognit ion
Gest ure recognit ion Facial gest ure recog. Body mot ion recog.
Own hand t racking
Physical schema
based obj. recog.
Bring hands
midline
Hand
linking
Grasping,
& t ransf er
Specif ic obj. recog. Generic object recog.
Body-based met aphors
DOF reduct ion
( specif ic coords)
DOF reduct ion
( generic coords)Bat t ing st at ic
ob ject s
Head/ eye coord
Body st abilit y ,
leaning, rest ing
Head/ body / ey e/ coord
Body+arm reaching Body mimicry
Manipulat ion t urn t aking
Sound localizat ion
Sound/ mot ion correl Human voice ext ract ion
Sound-based manip. Voice/ f ace assoc
Voice t urn t aking
Prot o language
Visual imagery Symbolizat ion
Imaginat ionMent al rehearsal
Mult iple-draf t s emergence
Tone ident if icat ion
“Building Brains
for Bodies”, Brooks & Stein (1993), MIT AI lab tech report 1439.
Sept 1
1 9 9 3
Sept 1
1 9 9 4Sept 1
1 9 9 5
Sept 1
1 9 9 6
Sept 1
1 9 9 7
Syst em sof t ware ( 0 t h) Syst em sof t ware ( commercial processor)
Periperhal Mot ion
Saccades
VOR
Smoot h pursuit
Vergence
based st ereo
Ullman-esque
visual rout ines
Face pop-out s Face remembering Face recognit ion
Gest ure recognit ion Facial gest ure recog. Body mot ion recog.
Own hand t racking
Physical schema
based obj. recog.
Bring hands
midline
Hand
linking
Grasping,
& t ransf er
Specif ic obj. recog. Generic object recog.
Body-based met aphors
DOF reduct ion
( specif ic coords)
DOF reduct ion
( generic coords)Bat t ing st at ic
ob ject s
Head/ eye coord
Body st abilit y ,
leaning, rest ing
Head/ body / ey e/ coord
Body+arm reaching Body mimicry
Manipulat ion t urn t aking
Sound localizat ion
Sound/ mot ion correl Human voice ext ract ion
Sound-based manip. Voice/ f ace assoc
Voice t urn t aking
Prot o language
Visual imagery Symbolizat ion
Imaginat ionMent al rehearsal
Mult iple-draf t s emergence
Tone ident if icat ion
“Building Brains
for Bodies”, Brooks & Stein (1993), MIT AI lab tech report 1439.
it is nor hand, nor foot, nor arm, nor face, nor any other
part belonging to a man.
What’s Consciousness?
Glenn Matsumura, Wired 2007
SG5-UT Robotic Arm
Tad McGeer's passive dynamic walker
Chuck Rosenbergʼs IT, 1997
Dennett (2008)“Contents arise, get revised, contribute to... the modulation of behavior, and in the process leave their traces in memory...”
“Only [commonality is] the historical property of having won a temporally local competition with sufficient decisiveness... to enable recollection...”
Dennett (2008)“Contents arise, get revised, contribute to... the modulation of behavior, and in the process leave their traces in memory...”
“Only [commonality is] the historical property of having won a temporally local competition with sufficient decisiveness... to enable recollection...”
• Characteristics:
1. Selection from concurrent options.
2. Indicated by episodic memory.
Function-Based Theory
• Consciousness is holding one stimulus in mind while searching options primed by it for a better response.
• Only triggered when next action isn’t obvious (reflexive or trained).
• Side effect: special types of learning.
• Side effect: long reaction times. Focus attention longer when less certain.
Joanna J. Bryson, “Crude, Cheesy, Second-Rate Consciousness”, The Second AISB Symposium Computing and Philosophy, Mark Bishop (ed), pp. 10–15, Edinburgh UK, April 2009.
Monkeys Learning New Rewards (or not)
• Monkeys that learn chained pairs of values (A>B; B>C; C>D; D>E; E>F) normally are faster at assessing stimuli the further they are on the chain (B>E faster than B>D).
• Elderly monkeys are always fast.
• Elderly monkeys also don’t learn when you change the reward scheme -- not aware?
Herb Terrace, Columbia, NY
Joanna J. Bryson, “Age-Related Inhibition and Learning Effects: Evidence from Transitive Performance”, in Proceedings of the 31st Annual Meeting of the Cognitive Science Society (CogSci 2009) pp. 3040–3045.
What’s Consciousness?
• A module for learning new action selection.
• Triggered by uncertainty.
• Detectable due to episodic memories & reaction time.
• A finite resource constantly directed towards what seems most surprising.
“If the best the roboticists can hope for is the creation of some
crude, cheesy, second-rate 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 (133-146).
Outline
• Introduction to Intelligence & Cognition
• Where do you put it in a Cognitive System? Behavior Oriented Design
• Primates & Cognition
• Conclusions
Cognitive Robots
Bad News:
Cognitive Robots
Bad News:
Cognition doesn’t require robots (only rich, dynamic, real-time environments).
Cognitive Robots
Bad News:
Cognition doesn’t require robots (only rich, dynamic, real-time environments).
Robots don’t require cognition.
Cognitive Robots
Bad News:
Cognition doesn’t require robots (only rich, dynamic, real-time environments).
Robots don’t require cognition.
Good News:
Cognitive Robots
Bad News:
Cognition doesn’t require robots (only rich, dynamic, real-time environments).
Robots don’t require cognition.
Good News:
Cognitive robots are still pretty interesting.
What I Learned from Robots
What I Learned from Robots1. Perception is hard -- which explains the brain.
• Lead to specialized representations encapsulated in modules; my method of behavior-module decomposition.
What I Learned from Robots1. Perception is hard -- which explains the brain.
• Lead to specialized representations encapsulated in modules; my method of behavior-module decomposition.
2. Discrete action selection is compatible with continuous acting, provided the primitive `acts’ alter ongoing behaviour supported by modules.
• e.g. motor act sends target velocity, not vector;
• multiple || devices/modules e.g. speech, motion.
The BrainHigher mammals separate sense & action (Central Sulcus).
Chance for Cognition?(images: Carlson)
When Your Robot Must Think...
• Modularity: problem spaces, combat combinatorics, allow locally-optimal representations.
• Hierarchical action selection for real-time arbitration between modules.
• Dedicated, high-frequency goal / attention switching, compensates for hierarchical AS.
• Agile development, refactoring (Beck 2000).
Thanks!
Mark Wood Jan Drugowitsch
Sam Partington
Tristan Caulfield
Cyril Brom (et al)
Jon Leong
The Role of Cognition in Cognitive Systems
From Robots to Primatology Joanna J. Bryson
Artificial Models of Natural IntelligenceUniversity of Bath
Institute of Cognitive and Evolutionary Anthropology University of Oxford
http://www.cs.bath.ac.uk/ai/AmonI.html
Functionalist Assumption: All we care about is producing intelligent behaviour.
• Physical Symbol System Hypothesis (Newell & Simon 1963); Qualia, Chalmers “hard problem” (1995).
• Thinking, consciousness as epiphenomena (Churchland 1988, Brooks & Stein 1993).
Science: We’ll build it to see if we need it.
Action Selection
Evolving Culture• Old theories of limits:
altruism, rate of environmental change.
• Concurrency can accelerate behaviour change (Bryson 2008).
• Čače & Bryson (2005, 2007) show altruistic communication about food is selected for due to niche creation.
Ivana Čače and Joanna J. Bryson, “Agent Based Modelling of Communication Costs: Why Information can be Free”, in Emergence and Evolution of Linguistic Communication C. Lyon, C.L Nehaniv and A. Cangelosi, eds., pp. 305–322, Springer 2007.
BOD Development Cycle1. Initial decomposition ⇒ specification.
2. Scale the system.
i. Code one behavior and/or plan.
ii. Test and debug code (test earlier plans).
iii. Simplify the design.
3. Revise the specification.
BOD Development Cycle1. Initial decomposition ⇒ specification.
2. Scale the system.
i. Code one behavior and/or plan.
ii. Test and debug code (test earlier plans).
iii. Simplify the design.
3. Revise the specification.
1. Specify (high-level) what the agent will do.
2. Describe activities as sequences of actions. competences and action patterns
3. Identify sensory and action primitives from these sequences.
4. Identify the state necessary to enable the primitives, cluster primitives by shared state. behavior modules
5. Identify and prioritize goals / drives. drive collection
6. Select a first (next) behavior to implement.
BOD Development Cycle1. Initial decomposition ⇒ specification.
2. Scale the system.
i. Code one behavior and/or plan.
ii. Test and debug code (test earlier plans).
iii. Simplify the design.
3. Revise the specification.
BOD Development Cycle1. Initial decomposition ⇒ specification.
2. Scale the system.
i. Code one behavior and/or plan.
ii. Test and debug code (test earlier plans).
iii. Simplify the design.
3. Revise the specification.
Simplify the Design
Use the simplest representations.
• Plans:
• primitives, action patterns, competences.
• drives only if need to always check.
• Behavior modules / memory:
• none, deictic, specialized, general.
(Bryson, AgeS 2003)
Simplify the DesignTrade off representations: plans vs. behaviors
• Use simplest plan structure unless redundancy (split primitives for sequence, add variable state in modules).
• If competences too complicated, introduce primitives or create more hierarchy.
• Split large behaviors, use plans to unify.
• All variable state in modules (deictic).(Bryson, AgeS 2003)
References
References• Cyril Brom and Joanna J. Bryson, “Action Selection for
Intelligent Systems”, white paper for euCognition, 7 August 2006.
References• Cyril Brom and Joanna J. Bryson, “Action Selection for
Intelligent Systems”, white paper for euCognition, 7 August 2006.
• Joanna J. Bryson, “Cross-Paradigm Analysis of Autonomous Agent Architecture”, Journal of Experimental and Theoretical Artificial Intelligence 12(2):165-190, 2000.
References• Cyril Brom and Joanna J. Bryson, “Action Selection for
Intelligent Systems”, white paper for euCognition, 7 August 2006.
• Joanna J. Bryson, “Cross-Paradigm Analysis of Autonomous Agent Architecture”, Journal of Experimental and Theoretical Artificial Intelligence 12(2):165-190, 2000.
• Joanna J. Bryson and Lynn Andrea Stein, “Architectures and Idioms: Making Progress in Agent Design”, The Seventh International Workshop on Agent Theories, Architectures and Languages (ATAL), Boston, 2000.
“History as Evolution” Hypothesis
Joanna J. Bryson, “Cross-Paradigm Analysis of Autonomous Agent Architecture”, Journal of Experimental and Theoretical Artificial Intelligence 12(2):165-190, 2000.
“History as Evolution” Hypothesis
• If an architecture is around for a while, and it changes, the change was probably selected, adaptive.
Joanna J. Bryson, “Cross-Paradigm Analysis of Autonomous Agent Architecture”, Journal of Experimental and Theoretical Artificial Intelligence 12(2):165-190, 2000.
“History as Evolution” Hypothesis
• If an architecture is around for a while, and it changes, the change was probably selected, adaptive.
• This is particularly likely if the change goes against the stated theories of the architectureʼs makers.
Joanna J. Bryson, “Cross-Paradigm Analysis of Autonomous Agent Architecture”, Journal of Experimental and Theoretical Artificial Intelligence 12(2):165-190, 2000.
“History as Evolution” Hypothesis & Correlary
“History as Evolution” Hypothesis & Correlary• If an architecture is around for a while,
and it changes, the change was probably selected, adaptive.
“History as Evolution” Hypothesis & Correlary• If an architecture is around for a while,
and it changes, the change was probably selected, adaptive.
• If similar features occur in a lot of architectures with different phylogenies, those features are probably adaptive.
“History as Evolution” Hypothesis & Correlary• If an architecture is around for a while,
and it changes, the change was probably selected, adaptive.
• If similar features occur in a lot of architectures with different phylogenies, those features are probably adaptive.
• If you want to make a contribution to a field, describe your best innovations in terms of well-known systems.
Productions
• From sensing to action (c.f. Skinner; conditioning; Witkowski 2007.)
• These work -- basic component of intelligence.
• The problem is choice (search).
• Requires an arbitration mechanism.
Production-Based Architectures
• Expert Systems: allow choice of policies, e.g. recency, utility, random.
• SOAR: problem spaces (from GPS), impasses, chunk learning.
• ACT-R: (Bayesian) utility, problem spaces (reluctantly, from SOAR/GPS.)
Soar
• Productions operate on predicate database.
• If conflict, declare impasse, reason (search).
• Remember resolution: chunk
!"
!"#$%&'($&#)*
+,-./
0".%*
1-%/&"#
2.3"%
4-/(5$/
67.895-
0:/$-8/
+895-8-#$.$&"#
#$%&'$&
()*+,'-.
/-*0
1,2,34,356
7)89.*.,
:-;,<,35,1,5=9=-3>:650, 7-*<?>@ ABBB 7/C$
&*5D=<@7-*<
EFD@7-*<
CG2<-H,4
C3.,<I*5,9
&:JK&+
F<*22,<7-*<L>@ ABBM
D=<@7-*<
C39.<)5.-@7-*<
N=O;
PII=5=,3567-*<M>@ ABBQ :
D=<@7-*<
N,<-@7-*<
PR.,<3*0
&*9+9
1,9.<)5.=H,
$2,<*.-<97=3O0,>7.*., 7-*<!>@ AB?B
P&@7-*<
SJ@7-*<
PR.,<3*0>
E,0,*9,%&:7-*<T>@ AB?M
/,3,<*0
J,*<3=3OEA@7-*<:;)3+=3O 7-*<U>@ AB?T
%3=H,<9*0
7)8O-*0=3O
EA@7-*<
162*<@7-*<
$V7!
J=92V<,I,<,35,9 7)8O-*09 7-*<Q>@ AB?U
76G8-0
769.,G9
N,)<=9.=5
7,*<5;
V<-4)5.=-3
769.,G9
%3=H,<9*0
F,*+>#,.;-4
WDV7>Q
J=92F,*+
#,.;-497-*<A>@ AB?Q &-6>&*9+9
V<-80,G>
72*5,9
Soar
!"
!"#$%&'($&#)*
+,-./
0".%*
1-%/&"#
2.3"%
4-/(5$/
67.895-
0:/$-8/
+895-8-#$.$&"#
#$%&'$&
()*+,'-.
/-*0
1,2,34,356
7)89.*.,
:-;,<,35,1,5=9=-3>:650, 7-*<?>@ ABBB 7/C$
&*5D=<@7-*<
EFD@7-*<
CG2<-H,4
C3.,<I*5,9
&:JK&+
F<*22,<7-*<L>@ ABBM
D=<@7-*<
C39.<)5.-@7-*<
N=O;
PII=5=,3567-*<M>@ ABBQ :
D=<@7-*<
N,<-@7-*<
PR.,<3*0
&*9+9
1,9.<)5.=H,
$2,<*.-<97=3O0,>7.*., 7-*<!>@ AB?B
P&@7-*<
SJ@7-*<
PR.,<3*0>
E,0,*9,%&:7-*<T>@ AB?M
/,3,<*0
J,*<3=3OEA@7-*<:;)3+=3O 7-*<U>@ AB?T
%3=H,<9*0
7)8O-*0=3O
EA@7-*<
162*<@7-*<
$V7!
J=92V<,I,<,35,9 7)8O-*09 7-*<Q>@ AB?U
76G8-0
769.,G9
N,)<=9.=5
7,*<5;
V<-4)5.=-3
769.,G9
%3=H,<9*0
F,*+>#,.;-4
WDV7>Q
J=92F,*+
#,.;-497-*<A>@ AB?Q &-6>&*9+9
V<-80,G>
72*5,9
Soar• Soar has serious engineering.
!"
!"#$%&'($&#)*
+,-./
0".%*
1-%/&"#
2.3"%
4-/(5$/
67.895-
0:/$-8/
+895-8-#$.$&"#
#$%&'$&
()*+,'-.
/-*0
1,2,34,356
7)89.*.,
:-;,<,35,1,5=9=-3>:650, 7-*<?>@ ABBB 7/C$
&*5D=<@7-*<
EFD@7-*<
CG2<-H,4
C3.,<I*5,9
&:JK&+
F<*22,<7-*<L>@ ABBM
D=<@7-*<
C39.<)5.-@7-*<
N=O;
PII=5=,3567-*<M>@ ABBQ :
D=<@7-*<
N,<-@7-*<
PR.,<3*0
&*9+9
1,9.<)5.=H,
$2,<*.-<97=3O0,>7.*., 7-*<!>@ AB?B
P&@7-*<
SJ@7-*<
PR.,<3*0>
E,0,*9,%&:7-*<T>@ AB?M
/,3,<*0
J,*<3=3OEA@7-*<:;)3+=3O 7-*<U>@ AB?T
%3=H,<9*0
7)8O-*0=3O
EA@7-*<
162*<@7-*<
$V7!
J=92V<,I,<,35,9 7)8O-*09 7-*<Q>@ AB?U
76G8-0
769.,G9
N,)<=9.=5
7,*<5;
V<-4)5.=-3
769.,G9
%3=H,<9*0
F,*+>#,.;-4
WDV7>Q
J=92F,*+
#,.;-497-*<A>@ AB?Q &-6>&*9+9
V<-80,G>
72*5,9
Soar• Soar has serious engineering.
• “Evolution of Soar” is my favorite paper (Laird & Rosenbloom 1996)
!"
!"#$%&'($&#)*
+,-./
0".%*
1-%/&"#
2.3"%
4-/(5$/
67.895-
0:/$-8/
+895-8-#$.$&"#
#$%&'$&
()*+,'-.
/-*0
1,2,34,356
7)89.*.,
:-;,<,35,1,5=9=-3>:650, 7-*<?>@ ABBB 7/C$
&*5D=<@7-*<
EFD@7-*<
CG2<-H,4
C3.,<I*5,9
&:JK&+
F<*22,<7-*<L>@ ABBM
D=<@7-*<
C39.<)5.-@7-*<
N=O;
PII=5=,3567-*<M>@ ABBQ :
D=<@7-*<
N,<-@7-*<
PR.,<3*0
&*9+9
1,9.<)5.=H,
$2,<*.-<97=3O0,>7.*., 7-*<!>@ AB?B
P&@7-*<
SJ@7-*<
PR.,<3*0>
E,0,*9,%&:7-*<T>@ AB?M
/,3,<*0
J,*<3=3OEA@7-*<:;)3+=3O 7-*<U>@ AB?T
%3=H,<9*0
7)8O-*0=3O
EA@7-*<
162*<@7-*<
$V7!
J=92V<,I,<,35,9 7)8O-*09 7-*<Q>@ AB?U
76G8-0
769.,G9
N,)<=9.=5
7,*<5;
V<-4)5.=-3
769.,G9
%3=H,<9*0
F,*+>#,.;-4
WDV7>Q
J=92F,*+
#,.;-497-*<A>@ AB?Q &-6>&*9+9
V<-80,G>
72*5,9
Soar• Soar has serious engineering.
• “Evolution of Soar” is my favorite paper (Laird & Rosenbloom 1996)
• Admits problems!
!"
!"#$%&'($&#)*
+,-./
0".%*
1-%/&"#
2.3"%
4-/(5$/
67.895-
0:/$-8/
+895-8-#$.$&"#
#$%&'$&
()*+,'-.
/-*0
1,2,34,356
7)89.*.,
:-;,<,35,1,5=9=-3>:650, 7-*<?>@ ABBB 7/C$
&*5D=<@7-*<
EFD@7-*<
CG2<-H,4
C3.,<I*5,9
&:JK&+
F<*22,<7-*<L>@ ABBM
D=<@7-*<
C39.<)5.-@7-*<
N=O;
PII=5=,3567-*<M>@ ABBQ :
D=<@7-*<
N,<-@7-*<
PR.,<3*0
&*9+9
1,9.<)5.=H,
$2,<*.-<97=3O0,>7.*., 7-*<!>@ AB?B
P&@7-*<
SJ@7-*<
PR.,<3*0>
E,0,*9,%&:7-*<T>@ AB?M
/,3,<*0
J,*<3=3OEA@7-*<:;)3+=3O 7-*<U>@ AB?T
%3=H,<9*0
7)8O-*0=3O
EA@7-*<
162*<@7-*<
$V7!
J=92V<,I,<,35,9 7)8O-*09 7-*<Q>@ AB?U
76G8-0
769.,G9
N,)<=9.=5
7,*<5;
V<-4)5.=-3
769.,G9
%3=H,<9*0
F,*+>#,.;-4
WDV7>Q
J=92F,*+
#,.;-497-*<A>@ AB?Q &-6>&*9+9
V<-80,G>
72*5,9
Soar• Soar has serious engineering.
• “Evolution of Soar” is my favorite paper (Laird & Rosenbloom 1996)
• Admits problems!
• Not enough applications for human-like AI
Architecture Lessons(from CMU)
• An architecture needs:
• action from perception, and
• further structure to combat combinatorics.
• Dealing with time is hard.
ACT-R
• Learns (& executes) productions.
• For arbitration, rely on (Bayesian probabalistic) utility.
• Call it implicit knowledge.
ACT-R Research Programme• Replicate lots of
Cognitive Science results.
• See if the brain does what you think it needs to.
• Win Rumelhart Prize (John Anderson, 2000).
Retrieval Buffer(VLPFC)
Goal Buffer(DLPFC)
Manual Motor(Motor)
Intentional Module(not identified)
External World
Matching (Striatum)
Execution (Thalamus)
Selection (Pallidum)
Pro
du
cti
on
s(B
asalG
an
glia)
Declarative Module(Temporal / Hippocampus)
Visual Buffer(Parietal)
Visual Module(Occipital/Parietal)
Manual Module(Motor/Cerebellum)
Architecture Lessons(from CMU)
• Architectures need productions and problem spaces.
• Real-time is hard.
• Being easy to use can be a win.
Spreading Activation Networks
• “Maes Nets” (Adaptive Neural Arch.; Maes 1989)
• Activation spreads from senses and from goals through net of actions.
• Highest activated
Spreading Activation Networks
Spreading Activation Networks
• Sound good:
• easy
• brain-like (priming, action potential).
• Still influential (Franklin 2000, Shanahan 2006).
Spreading Activation Networks
• Sound good:
• easy
• brain-like (priming, action potential).
• Still influential (Franklin 2000, Shanahan 2006).
• Canʼt do full action selection:
• Donʼt scale; donʼt converge on comsumatory acts (Tyrrell 1993).
Tyrrell (1993)
Extended Rosenblatt and Payton Free-Flow HierarchyN NE E SE S SW W NW
UTReproduce
1.4
T U
Move ActionsMate
-0.08
Court
P. Mate Rand. Dir P. Den R. Den All Dirs
Clean Leavethis Sq
CleanSleep
Mate Court
ApproachMate
Explore For Mates
Explore
Sleep
ApproachP. Den
Approach R. Den
Sleepin Den Clean
Keep
DirtinessLow HealthNight Proxfrom DenDistance
-0.10
-0.05
-0.01
-0.05 -0.05-0.15
Courted Mate in Sq
Mate in SqReceptive
No Denin Sq
Den in Sq
No Den in Sq
in SqDen
-0.02
-0.02-0.25
-0.30-0.04
= small negative activation
= positive activation
= small positive activation
= zero activation
= large positive activation(1.0)
Subsumption (Brooks 1986)
Subsumption (Brooks 1986)• Emphasis on
sensing to action (via Augmented FSM).
Subsumption (Brooks 1986)• Emphasis on
sensing to action (via Augmented FSM).
• Very complicated, distributed arbitration.
Subsumption (Brooks 1986)• Emphasis on
sensing to action (via Augmented FSM).
• Very complicated, distributed arbitration.
• No learning.
Subsumption (Brooks 1986)• Emphasis on
sensing to action (via Augmented FSM).
• Very complicated, distributed arbitration.
• No learning.
• Worked.
Architecture Lessons(Subsumption)
Architecture Lessons(Subsumption)
• Action from perception can provide the further structure -- modules (behaviors).
• Modules also support iterative development / continuous integration.
Architecture Lessons(Subsumption)
• Action from perception can provide the further structure -- modules (behaviors).
• Modules also support iterative development / continuous integration.
• Real time should be a core organizing principle -- start in the real world.
Architecture Lessons(Subsumption)
• Action from perception can provide the further structure -- modules (behaviors).
• Modules also support iterative development / continuous integration.
• Real time should be a core organizing principle -- start in the real world.
• Good ideas can carry bad ideas a long way (no learning, hard action selection).
Architecture Lesson?• Goals ordering
needs to be flexible.
Architecture Lesson?• Goals ordering
needs to be flexible.
• Maybe spreading activation is good for this.
SA: Layers vs. Behaviours• Relationship not
evident except in development!
SA: Layers vs. Behaviours• Relationship not
evident except in development!
SA: Layers vs. Behaviours• Relationship not
evident except in development!
SA: Layers vs. Behaviours• Relationship not
evident except in development!
SA: Layers vs. Behaviours• Relationship not
evident except in development!
Layered or Hybrid Architectures
1. Incorporate behaviors/modules (action from sensing) as “smart” primitives.
2. Use hierarchical dynamic plans for behavior sequencing.
3. (Allegedly) some have automated planner to make plans for layer 2.
• Examples: Firby/RAPS/3T (ʻ97); PRS (1992-2000); Hexmoore ʻ95; Gat ʻ91-98
Belief, Desires, Intentions (BDI)• Beliefs:
Predicates
• Desires: goals & related dynamic plans
• Intentions:current goal
Procedural Reasoning System
Procedural Reasoning System• BDI
Procedural Reasoning System• BDI
• And reactive (responds to emergencies by changing intentions.)
Procedural Reasoning System• BDI
• And reactive (responds to emergencies by changing intentions.)
• Er... once or twice (Bryson ATAL 2000).
Architecture Lessons
• Structured dynamic plans make it easier to get your robot to do complicated stuff.
• Automated planning (or for Soar, chunking/learning) is seldom actually used.
• To facilitate that automated planning, modularity is often compromised.
Soar as a 3LA• J. Laird & P.
Rosenbloom, “The Evolution of the Soar Cognitive Architecture”, Mind Matters, D. Steier and T. Mitchell eds., 1996.
CogAff
• Reflection on Top.
• Sense & Action separated!
• (Davis & Sloman 1995)
• Reflection on Top.
• Sense & Action separated!
• Hierarchy in AS; Goal Swapping (Alarms).
• (Sloman 2000)
CogAff
• Reflection on Top.
• Sense & Action separated!
• Hierarchy in AS, Goal Swapping (now reactive).
• Current Web
CogAff
Architecture Lessons (CogAff)
Architecture Lessons (CogAff)
• Maybe you don’t really want productions as your basic representation -- you may want to come between a sense and an act sometimes.
Architecture Lessons (CogAff)
• Maybe you don’t really want productions as your basic representation -- you may want to come between a sense and an act sometimes.
• Aaron Sloman thinks about a lot more human cognitive traits than I do.