Cognitive Architectures Cognitive Architectures
ACT ACT - - R R
Outline Outline
Short glance on the history of ACT
Short glance on the history of ACT - - R R What is ACT
What is ACT - - R ? R ? Mapping ACT
Mapping ACT - - R onto the brain R onto the brain ACT ACT - - R 5.0 Architecture R 5.0 Architecture
Components of ACT
Components of ACT - - R R What is ACT
What is ACT - - R used for? R used for?
General discussion
General discussion
History of the ACT
History of the ACT - - framework framework
1976: first ACT theory came out 1976: first ACT theory came out
1982: first ACT implementation appeared 1982: first ACT implementation appeared
Since then both, the
Since then both, the Theory Theory and the and the Implementations
Implementations were further developed were further developed (ACT, ACT
(ACT, ACT - - R, ACT R, ACT - - R2.0, ACT R2.0, ACT - - R3.0, R3.0, ACT ACT - - R4.0) R4.0)
2001: release of ACT
2001: release of ACT - - R5.0 (Theory and R5.0 (Theory and Implementation) which is since then the Implementation) which is since then the
state of the art ACT
state of the art ACT - - R R
What is ACT
What is ACT - - R R
ACT ACT - - R is a cognitive architecture R is a cognitive architecture Researchers working on ACT
Researchers working on ACT - - R strive to R strive to understand how people organize
understand how people organize knowledge and produce intelligent knowledge and produce intelligent
behaviour
behaviour . .
What is ACT
What is ACT - - R R
ACT ACT - - R is a programming language R is a programming language Models are written in ACT
Models are written in ACT - - R R
During runtime of a model, ACT
During runtime of a model, ACT - - R R provides the runtime environment.
provides the runtime environment.
Due to it’s special design as a cognitive Due to it’s special design as a cognitive
architecture, models in ACT
architecture, models in ACT - - R can mirror R can mirror
human behavior on a cognitive psychology
human behavior on a cognitive psychology
task task
What is ACT
What is ACT - - R R
Based on facts Based on facts
derived from derived from
psychology psychology
experiments, experiments, ACT ACT - - R is a R is a
framework framework
Models in ACT
Models in ACT - - R R reflect a certain reflect a certain
aspect of cognition
aspect of cognition
Framework ? Framework ?
Environment
ACT-R
ACT ACT - - R Architecture R Architecture
ART ART - - R claim: cognition as the interaction R claim: cognition as the interaction between specific units of knowledge:
between specific units of knowledge:
Declarative knowledge Declarative knowledge Unit:
Unit: Chunks Chunks
E.g. facts, goals, … E.g. facts, goals, …
Procedural knowledge Procedural knowledge Unit:
Unit: Production rules Production rules
E.g. action rules, behavior E.g. action rules, behavior
rules, …
rules, …
ACT ACT - - R Architecture R Architecture
Environment
Production rules
Chunk
Chunk Chunk
Chunk
A C T
A C T - - R Architecture R Architecture
Chunks are created by specific
Chunks are created by specific modules modules
visual module produces chunk “Christian is in visual visual module produces chunk “Christian is in visual
field”
field”
Motor module produces “pressure on left hand”
Motor module produces “pressure on left hand”
Chunks set modules to action Chunks set modules to action
“search Christian” said to visual module
“search Christian” said to visual module
Modules transmit and retrieve information Modules transmit and retrieve information
only out of
only out of buffers buffers
Each module has a specific buffer for his Each module has a specific buffer for his
chunks
chunks
ACT ACT - - R Architecture R Architecture
Environment
Modules Modules
Buffers Buffers
Production rules
Chunk
Chunk Chunk
Chunk
Mapping ACT
Mapping ACT - - R onto the brain R onto the brain
Question: How is ACT
Question: How is ACT - - R related to newest R related to newest studies in neurobiology and
studies in neurobiology and neuroimaging
neuroimaging ? ?
Answer: all parts of ACT
Answer: all parts of ACT - - R are designed R are designed to reflect certain brain areas!
to reflect certain brain areas!
Mapping ACT
Mapping ACT - - R onto the brain R onto the brain Modules
Modules
In a few examples we will try to give you a In a few examples we will try to give you a
scratch of how ACT
scratch of how ACT - - R is designed. R is designed.
Visual system: there are two build in visual Visual system: there are two build in visual
modules in ACT
modules in ACT - - R referring to: R referring to:
The dorsal “where” pathway (locations) The dorsal “where” pathway (locations)
The ventral “what” pathway The ventral “what” pathway
Mapping ACT
Mapping ACT - - R onto the brain R onto the brain
Environment
Visual Buffer (Parietal) Visual Module (Occipital/etc)
Modules Buffers
Production
rules
Mapping ACT
Mapping ACT - - R onto the brain R onto the brain
As for the visual system, other modules As for the visual system, other modules
have been designed to match specific have been designed to match specific
brain areas:
brain areas:
Manual buffer Manual buffer = motor and = motor and somatosensory somatosensory cortical areas
cortical areas
Goal buffer Goal buffer = = dorsolateral dorsolateral prefrontal cortex prefrontal cortex DLPFC
DLPFC
Retrieval buffer Retrieval buffer = = ventrolateral ventrolateral prefrontal prefrontal cortex VLPFC (long
cortex VLPFC (long - - term declarative memory) term declarative memory)
Mapping ACT
Mapping ACT - - R onto the brain R onto the brain
Environment
Visual Buffer (Parietal) Visual Module (Occipital/etc)
Modules Buffers
Production rules
Manual Buffer (Motor)
Manual Module (Motor/Cerebellum) Retrieval Buffer
(VLPFC) Goal Buffer
(DLPFC) Intentional Module
(not identified) Declarative Module (Temporal/Hippocampus)
Mapping ACT
Mapping ACT - - R onto the brain R onto the brain Production rules
Production rules
The basal ganglia are thought to The basal ganglia are thought to
implement production rules in ACT
implement production rules in ACT - - R: R:
Striatum: corresponding with cortical areas, Striatum: corresponding with cortical areas, responsible for
responsible for patter recognition patter recognition
Palladium: inhibitory component, performs Palladium: inhibitory component, performs conflict
conflict - - resolution function resolution function
Thalamus: projects to all major cortical areas, Thalamus: projects to all major cortical areas, controls
controls execution of production actions execution of production actions
Mapping ACT
Mapping ACT - - R onto the brain R onto the brain Production rules
Production rules
Environment
Productions (Basal Ganglia)
Retrieval Buffer (VLPFC) Matching (Striatum)
Selection (Pallidum) Execution (Thalamus) Goal Buffer
(DLPFC)
Visual Buffer
(Parietal) Manual Buffer (Motor)
Manual Module (Motor/Cerebellum) Visual Module
(Occipital/etc) Intentional Module
(not identified) Declarative Module (Temporal/Hippocampus)
Production
rules
ACT ACT - - R Architecture R Architecture
Environment
Productions (Basal Ganglia)
Retrieval Buffer (VLPFC) Matching (Striatum)
Selection (Pallidum) Execution (Thalamus) Goal Buffer
(DLPFC)
Visual Buffer
(Parietal) Manual Buffer (Motor)
Manual Module (Motor/Cerebellum) Visual Module
(Occipital/etc) Intentional Module
(not identified) Declarative Module
(Temporal/Hippocampus)
Declative
memory procedural
procedural memory
“pattern-
matcher” modules
buffers
Chunk
Chunk Chunk
Chunk
The modules The modules
There are two types of modules:
There are two types of modules:
memory modules memory modules . .
declarative memory declarative memory
procedural memory procedural memory
perceptual
perceptual - - motor modules motor modules
take care of the interface with the simulation of the take care of the interface with the simulation of the real world (visual and the manual modules).
real world (visual and the manual modules).
chunks chunks
chunks = chunks =
units of declarative knowledge units of declarative knowledge
represent things remembered or perceived represent things remembered or perceived
example:
example:
2+3=5 2+3=5
Boston is the capital of Massachusetts Boston is the capital of Massachusetts
there is an attended object in the visual field there is an attended object in the visual field
... ...
Chunk
chunks: examples chunks: examples
(CHUNK-TYPE integer value)
(CHUNK-TYPE addition-fact addend1 addend2 sum)
{ATTRIBUTES}
DEFINITION
INSTANCE
TYPE
one way to model the fact:
one way to model the fact: 2+3=5 2+3=5
(CHUNK-TYPE integer value)
(CHUNK-TYPE addition-fact addend1 addend2 sum) (CHUNK-TYPE integer value)
(CHUNK-TYPE addition-fact
addend1 addend2 sum)(three
isa
integer
value 3)
NAME(CHUNK-TYPE integer value)
(CHUNK-TYPE addition-fact addend1 addend2 sum)
(three
isa
integer
value 3)
(CHUNK-TYPE integer value)
(CHUNK-TYPE addition-fact addend1 addend2 sum) (three
isa integer value 3) (four
isa integer value 4) (seven
isa integer value 7) (fact3+4
isa addition-fact addend1 three addend2 four sum seven)
chunks: examples chunks: examples
reference to other chunks
ADDITION-FACT
FACT3+4
ADDEND1 SUM
ADDEND2 THREE
FOUR
SEVEN
isa
isa
INTEGER
isa
VALUE VALUE
3 7
isa
VALUE
4
chunks: examples
chunks: examples
Fact:
Encoding:
(Chunk-Type proposition agent action object)
The cat sits on the mat.
proposition
action cat007
sits_on
mat isa
fact007
agent object
(Add-DM (fact007
isa proposition agent cat007 action sits_on object mat) )
chunks: examples
chunks: examples
Fact:
The black cat with 5 legs sits on the mat.Chunks
(Chunk-Type proposition agent action object) (Chunk-Type cat legs color)
(Add-DM
(fact007 isa proposition agent cat007 action sits_on object mat) (cat007 isa cat
legs 5
color black) )
proposition
action cat007
sits_on
mat isa
fact007
agent object
cat
isa
color 5
black legs
chunks: examples
chunks: examples
chunks: examples chunks: examples
animal
fish
shark salmon canary ostrich
bird moves skin
gills swims
swims
dangerous edible
swims
yellow sings
can‘t fly tall
flies wings
productions productions
Procedural knowledge Procedural knowledge
to achieve a given goal:
to achieve a given goal:
processes processes
skills skills
production = production =
unit of procedural knowledge unit of procedural knowledge
condition- condition - action rule that “fire” when the action rule that “fire” when the conditions are satisfied and execute the conditions are satisfied and execute the
specified actions.
specified actions.
Environment Productions (Basal Ganglia)
Retrieval Buffer (VLPFC)
Matching (Striatum) Selection (Pallidum) Execution (Thalamus) Goal Buffer (DLPFC)
Visual Buffer (Parietal) Manual Buffer (Motor)
Manual Module (Motor/Cerebellum) Visual Module (Occipital/etc)
Intentional Module
(not identified) Declarative Module
(Temporal/Hippocampus)
productions productions
conditions can depend on conditions can depend on
the current goal to be achieved, the current goal to be achieved,
the state of declarative knowledge (i.e. recall of a the state of declarative knowledge (i.e. recall of a chunk)
chunk)
the current sensory input from the external the current sensory input from the external environment.
environment.
actions can:
actions can:
alter the state of declarative memory alter the state of declarative memory
change goals change goals
initiate motor actions in the external environment initiate motor actions in the external environment
Structure of Productions Structure of Productions
( ( P P name name
condition part
condition part Specification of Buffer Tests Specification of Buffer Tests . .
. . delimiter
delimiter ==> ==>
action part
action part Specification of Buffer Specification of Buffer Transformations
Transformations . .
. .
) )
Example of productions
(P increment
=goal>
ISA count-from number =num1
=retrieval>
ISA count-order first =num1
second =num2
==>
=goal>
number =num2 +retrieval>
ISA count-order first =num2
)
Å If the goal is
to count from
=num1
Å and a chunk has been retrieved of type count-order
where the first number is =num1 and it is followed by =num2
Å Then
Å change the goal
to continue counting from =num2 Å and request a retrieval
of a count-order fact
for the number that follows =num2 operation & buffer
Example of productions
(P find-next-word
=goal>
ISA comprehend-sentence word nil
==>
+visual-location>
ISA visual-location screen-x lowest
attended nil
=goal>
word looking )
Å no word currently being processed.
Å find left-most unattended location
Å update state
Example of productions
(P attend-next-word
=goal>
ISA comprehend-sentence word looking
=visual-location>
ISA visual-location
==>
=goal>
word attending +visual>
ISA visual-object
screen-pos =visual-location )
Å looking for a word
Å visual location has been identified
Å update state
Å attend to object in that location
Discussion Discussion
The atomic components of thought?
a Is declarative knowledge (=chunk) available in every cognitive module?
a semantic be modeled arbitrary a chunks be of any granularity
•
“pixel in visual field” vs. “Chris is standing in front of me”
Î Can timing of the computation be compared with humans?
Î Does the division of symbolic and subsymbolic processing make sense?
Î Is ACT-R just a strange kind of programming language?
Environment
Productions (Basal Ganglia)
Retrieval Buffer (VLPFC)
Matching (Striatum) Selection (Pallidum) Execution (Thalamus)
Goal Buffer (DLPFC)
Visual Buffer
(Parietal) Manual Buffer
(Motor)
Manual Module (Motor/Cerebellum) Visual Module
(Occipital/etc)
Declarative Module (Temporal/Hippocampus) Intentional Module
(not identified)
the perceptual
the perceptual - - motor modules motor modules
no real sensors and effectors no real sensors and effectors
the output of the visual and the input to the the output of the visual and the input to the
motor system are just modeled motor system are just modeled
the visual and manual module are most the visual and manual module are most
important ( because of many computer important ( because of many computer
tasks, involving scanning the screen, tasks, involving scanning the screen,
typing, moving the mouse...)
typing, moving the mouse...)
the Act
the Act - - R visual system R visual system
visual location module
“where”
-> dorsal stream
visual object module
“what”
->ventral stream
visual system
Production P Request:
•constraint a
•constraint b
•...
Response:
location meeting those constraints
e.g. - screen x lowest - color: red
- screen-y-greater-than 153 ....
Æ leftmost word
Æ red object ( among green ones; supports experimental data from visual “pop-out- effects”
chunks
the Act
the Act - - R visual system R visual system
visual location module
“where”
-> dorsal stream
visual object module
“what”
->ventral stream
visual system
Production P chunk:
representation of the visual location
attention-shift to that location
Environment
Productions (Basal Ganglia)
Retrieval Buffer (VLPFC)
Matching (Striatum) Selection (Pallidum) Execution (Thalamus)
Goal Buffer (DLPFC)
Visual Buffer
(Parietal) Manual Buffer
(Motor)
Manual Module (Motor/Cerebellum) Visual Module
(Occipital/etc)
Declarative Module (Temporal/Hippocampus) Intentional Module
(not identified)
the goal module the goal module
In humans: Suppose the goal is to add 64 + 36 In humans: Suppose the goal is to add 64 + 36 Assumption: the sum is not already stored, but Assumption: the sum is not already stored, but
one has to go through a series of
one has to go through a series of substeps substeps to to come up with the answer and keep track of the come up with the answer and keep track of the
various partial results ( e.g. sum of the ten digits) various partial results ( e.g. sum of the ten digits)
The goal module has this responsibility of The goal module has this responsibility of
keeping track of what these intentions are so keeping track of what these intentions are so
that behavior will serve that goal
that behavior will serve that goal
How could the goal buffer be How could the goal buffer be
organized ? organized ?
good example for goal
good example for goal - - subgoal subgoal structures structures in problem solving: Tower of Hanoi
in problem solving: Tower of Hanoi Problem
Problem
naive human response: move the disks to naive human response: move the disks to
their ultimate location ( greedy) their ultimate location ( greedy)
but: goal
but: goal - - subgoal subgoal strategy is often strategy is often discovered during practice
discovered during practice
Tower of Hanoi
Tower of Hanoi - - problem problem
Interest of Anderson et al. (Tower of Hanoi:
Interest of Anderson et al. (Tower of Hanoi:
Evidence for the Cost of Goal Retrieval1, Evidence for the Cost of Goal Retrieval1,
2002):
2002):
not which strategy is adopted, but what not which strategy is adopted, but what
does it tell about goal
does it tell about goal - - subgoal subgoal interaction interaction what are the “cognitive costs” for
what are the “cognitive costs” for implementing a
implementing a subgoaling subgoaling strategy strategy
Experimental setup:
Experimental setup:
the strategy to solve the problem was given and the strategy to solve the problem was given and
trained trained
task: solve the problem as fast as possible task: solve the problem as fast as possible
formulate a goal by clicking the disk on the source formulate a goal by clicking the disk on the source peg , then the destination peg
peg , then the destination peg
do action or post goal do action or post goal
time between actions and accuracy of moves time between actions and accuracy of moves
were measured; also eye movements were were measured; also eye movements were
recorded
recorded
user interface
user interface
Strategy
Strategy - - Algorithm Algorithm
1. formulate a goal 1. formulate a goal
2. decision 2. decision
if a legal move to achieve your goal is possible, do it and skip if a legal move to achieve your goal is possible, do it and skip next step (3), otherwise post it on the
next step (3), otherwise post it on the goal stack
goal stack3. formulate a prerequisite goal 3. formulate a prerequisite goal
if you cannot move a disk D, find the if you cannot move a disk D, find the largest disk
largest diskthat’s blocking that’s blocking the move and move it to a peg which is neither the source, nor the move and move it to a peg which is neither the source, nor
the destination peg of D the destination peg of D
4.Try again 4.Try again
go back to #2 to see whether you can achieve your last goal go back to #2 to see whether you can achieve your last goal posted
posted
5. Repeat the process 5. Repeat the process
go back to step one, until all disks are at their final position go back to step one, until all disks are at their final position
example demo example demo
A B C
Do it ! Post it !
GOAL STACK
C B C
results:
results results
participants are slower at those points where participants are slower at those points where
they must retrieve a goal, and are more slower they must retrieve a goal, and are more slower
the longer ago it was posted the longer ago it was posted
the accuracy data suggests that participants are the accuracy data suggests that participants are
forgetting their goals forgetting their goals
The tendency to inspect the goal stack increases The tendency to inspect the goal stack increases
dramatically at those retrieval points dramatically at those retrieval points
Æ Æ goal retrieval seems to be the major factor goal retrieval seems to be the major factor limiting performance in this task
limiting performance in this task
But what has Act-R to do
with this
experiment?
Act Act - - R R
Act Act - - R was used to model this task R was used to model this task
Act Act - - R 4.0 had a “perfect memory” goal R 4.0 had a “perfect memory” goal stack on which all goals can be stored stack on which all goals can be stored
perfectly and accessed without any perfectly and accessed without any
retrieval time costs retrieval time costs
BUT: data shows clear goal limitations!
BUT: data shows clear goal limitations!
Altman &
Altman & Trafton Trafton : memory for goals might : memory for goals might behave like any other memory and be
behave like any other memory and be subject to forgetting
subject to forgetting
new Act
new Act - - R model R model
getting rid of the goal stack ! getting rid of the goal stack !
relies on
relies on ACt ACt - - Rs Rs general declarative general declarative memory to store goals
memory to store goals
in Act in Act - - R each chunk has a base R each chunk has a base - - level level
activation that increases each time the chunk activation that increases each time the chunk
is used and decreases with lack of use is used and decreases with lack of use
Gaussian retrieval Gaussian retrieval - - probability function over probability function over the base
the base -level activation - level activation
results results
nice fit !
nice fit ! nice fit !
nice fit !
conclusions of this research:
conclusions of this research:
cognitive architectures like Act
cognitive architectures like Act - - R(4.0) or SOAR R(4.0) or SOAR are wrong in their assumption of a special goal are wrong in their assumption of a special goal
stack stack
goals in the
goals in the subgoaling subgoaling task are probably no task are probably no different than other sort of intentions people set different than other sort of intentions people set
goals appear to behave like other more common goals appear to behave like other more common
kinds of declarative memory and shows the kinds of declarative memory and shows the
same effects in practice and retention interval
same effects in practice and retention interval
Environment
Productions (Basal Ganglia)
Retrieval Buffer (VLPFC)
Matching (Striatum) Selection (Pallidum) Execution (Thalamus)
Goal Buffer (DLPFC)
Visual Buffer
(Parietal) Manual Buffer
(Motor)
Manual Module (Motor/Cerebellum) Visual Module
(Occipital/etc)
Declarative Module (Temporal/Hippocampus) Intentional Module
(not identified)
the buffers the buffers
ACT ACT - - R accesses its modules (except for R accesses its modules (except for the procedural
the procedural - - memory module) through memory module) through buffers
buffers
For each module, a dedicated buffer For each module, a dedicated buffer
serves as the interface with that module serves as the interface with that module
The contents of the buffers at a given
The contents of the buffers at a given
moment in time represents the state of
moment in time represents the state of
ACT ACT - - R at that moment. R at that moment.
the buffers the buffers
each buffer can hold a relatively small amount of each buffer can hold a relatively small amount of
information (
information ( Î Î chunk) chunk)
chunks that were former buffer contents are now chunks that were former buffer contents are now
stored in the declarative memory module stored in the declarative memory module
buffers are conceptual similar to
buffers are conceptual similar to Baddley’s Baddley’s working memory “slave systems”
working memory “slave systems”
the central cognitive system can only sense the the central cognitive system can only sense the content of the buffers
content of the buffers
the content of the chunks can only be accessed by the content of the chunks can only be accessed by the highly specialized modules
the highly specialized modules
the buffers the buffers
the most important buffers in Act
the most important buffers in Act- -R are: R are:
Goal Buffer
Goal Bufferkeeps track of one’s internal state in solving a problem keeps track of one’s internal state in solving a problem preserves information across production cycles
preserves information across production cycles
Retrieval Buffer
Retrieval Bufferholds information retrieved from long
holds information retrieved from long--term declarative memoryterm declarative memory seat of chunk activation calculations
seat of chunk activation calculations
Manual Buffer
Manual Bufferresponsible for control of hands responsible for control of hands
Visual “where” Buffer
Visual “where” Buffer locationlocation
Visual “what” Buffer
Visual “what” Buffer visual objectsvisual objects
attention shifts correspond to buffer transformations attention shifts correspond to buffer transformations
Environment
Productions (Basal Ganglia)
Retrieval Buffer (VLPFC)
Matching (Striatum) Selection (Pallidum) Execution (Thalamus)
Goal Buffer (DLPFC)
Visual Buffer
(Parietal) Manual Buffer
(Motor)
Manual Module (Motor/Cerebellum) Visual Module
(Occipital/etc)
Declarative Module (Temporal/Hippocampus) Intentional Module
(not identified)
Pattern matcher Pattern matcher
The pattern matcher searches for a The pattern matcher searches for a
production that matches the current state production that matches the current state
of the buffers of the buffers
Only one such production can be Only one such production can be executed at a given moment
executed at a given moment
That production, when executed, can That production, when executed, can
modify the buffers and thus change the modify the buffers and thus change the
state of the system state of the system
Thus, in ACT
Thus, in ACT - - R cognition unfolds as a R cognition unfolds as a succession of production firings
succession of production firings
Making Choices: Conflict Resolution Expected Gain = E = PG-C
Probability of choosing i =
∑
jj t E
i t E
e e
//
P =
Successes = α + m Failures = β + n
Successes
Successes + Failures
P is expected probability of success G is value of goal
C is expected cost
t reflects noise in evaluation and is like temperature in the Boltzman equation
α is prior successes
m is experienced successes β is prior failures
n is experienced failures
Production selection
Production selection
Outlook Outlook
What is ACT
What is ACT - - R used for? R used for?
What is ACT
What is ACT - - R used for? R used for?
ACT ACT - - R has been used successfully to R has been used successfully to create models in domains such as:
create models in domains such as:
learning and memory, learning and memory,
problem solving and decision making, problem solving and decision making,
language and communication, language and communication,
perception and attention, perception and attention,
cognitive development, or cognitive development, or