+ All Categories
Home > Documents > COGNITIVE CONSTRUCTOR: A BIOLOGICALLY-INSPIRED SELF- REGULATED LEARNING PARTNER Alexei Samsonovich,...

COGNITIVE CONSTRUCTOR: A BIOLOGICALLY-INSPIRED SELF- REGULATED LEARNING PARTNER Alexei Samsonovich,...

Date post: 28-Dec-2015
Category:
Upload: phoebe-hampton
View: 218 times
Download: 1 times
Share this document with a friend
Popular Tags:
28
COGNITIVE CONSTRUCTOR: A BIOLOGICALLY-INSPIRED SELF-REGULATED LEARNING PARTNER Alexei Samsonovich, Anastasia Kitsantas, Nada Dabbagh
Transcript

COGNITIVE CONSTRUCTOR: A BIOLOGICALLY-INSPIRED SELF-REGULATED LEARNING PARTNER

Alexei Samsonovich, Anastasia Kitsantas, Nada Dabbagh

Back to 2006:Highlights of the BICA program

Goal: Capture the ‘magic’ of human cognition: Cognitive growth ability (human-like

learning) Metacognition Self-awareness Episodic memory Social cognition Natural language capabilities Emotional intelligence

The focus was on Learning Integration Biological fidelity Test-driven design

The ‘critical mass’ hypothesis underlying GMU approach

1. To implement the magic of human cognition, it is sufficient to create a minimal embedded architecture that is functionally equivalent to the human mind

2. The key building blocks of this architecture are:

a schema

a mental state

a cognitive map

3. The paradigm of virtual embedding allows us to finesse the missing “peripheral” capabilities (sensory, motor, language and general knowledge)

A schema in our framework…

…is an abstract model or a template that represents a cognitive category

provides a uniform format for all symbolic representations

Terminals bind toexternal content asspecified by their attributes

Internal nodes are bound to nodes of the same schema

The ‘head’ The ‘head’ represents the represents the schema itself schema itself (the cognitive (the cognitive category)category)

Internal links define functional relations among components

Each node refersto a schemaor to a primitive

Find S1Find S2

Find theirsharedterms

Schema X Schema Y

State YState X

Create thenew schemahead

Make suresame terminalsare shared

Add internalnodes to thenew schema

Metaschema head

Complexstate of X & Y

Store newschema insemanticmemory

New schema S12

XY

S12

Combinationmetaschema

New schema S12

Se

ma

ntic

me

mo

ryW

orkin

g

me

mo

ry

Schemas can create other schemas (learning)

A mental state is a snapshot of awareness of a Self

I-Now:

•Ideas

•Intent

I-Now:

•Ideas

•Intent

I-Imagine:

•Intermediate goal situation

I-Imagine:

•Intermediate goal situation

I-Goal:

•Stimulus satisfaction

I-Goal:

•Stimulus satisfaction

I-Next:

•Scheduled action

•Expectation

I-Next:

•Scheduled action

•Expectation

I-Previous:

•Ideas

•Visual input

I-Previous:

•Ideas

•Visual input

I-Meta:

•Scenario

•Analysis

I-Meta:

•Scenario

•AnalysisI-Past:

•Past experience

•Prospective memories

I-Past:

•Past experience

•Prospective memories

Working memoryWorking memory

Each box is a mental state. Each bulleted line is an instance of a Each box is a mental state. Each bulleted line is an instance of a schema (a state). The double line shows the current working schema (a state). The double line shows the current working scenario. Red color marks the focus of attention.scenario. Red color marks the focus of attention.The framework allows the system to process each mental state from The framework allows the system to process each mental state from another mental state (perspective), thereby providing a basis for another mental state (perspective), thereby providing a basis for various forms of metacognition.various forms of metacognition.Episodic memory consists of frozen mental states that once were Episodic memory consists of frozen mental states that once were active in working memory. active in working memory.

Mental states “implement” the Self

The Self is an imaginary abstraction that has real representations – tokens to which experiences are attributed.

The Self exists in multiple instances (one for each experienced mental perspective) that are processed in parallel.

Representations of the Self and its experiences are constrained by self axioms, producing an illusion that there is alive subject in the system.

A cognitive map allocates symbolic representations (e.g. schemas) in an abstract space based on their semantics

Cognitive map

contextual conceptual emotional

timetime

locationlocation

real-imagreal-imag

sizesize

specificityspecificity

rationalityrationality

good-badgood-bad

valencevalence

saliencesalience

…Exa

mpl

es o

f co

gniti

vedi

men

sion

s

Episodic Episodic memorymemory

LTM

targ

ets Semantic Semantic

memorymemory Episodic and Episodic and

semantic memorysemantic memory

Functions of cognitive map include

Integration of components Associative memory indexing Cognitive space modeling Generation of ‘intuitive feelings’ and emotions Guidance of the process of thinking Analogical or constrained retrieval Pathfinding during strategic retrieval Memory (re)consolidation Formation of a system of values

Color key: higher-level symbolic, algorithmic, connectionist

The resultant architecture has 8 components mapped onto the brain

Input-output

Proceduralmemory

EngineRewardsystem

Semanticmemory

Episodicmemory

Workingmemory

Cognitivemap

11

Example: Using episodic memory

in analysis of past eventstarget

target

target

detour

10 min before the blast

12 min before the blast

15 min before the blast

me

me

I-Imagine:

•He turns

I-Imagine:

•He turns

I-Goal:

•Report suspicious activity

I-Goal:

•Report suspicious activity

I-Next:

•Report truck

I-Next:

•Report truck

I-Past-2:

•Car accident

•Traffic jam

I-Past-2:

•Car accident

•Traffic jam

I-Meta:

•Scenario

•Analysis

I-Meta:

•Scenario

•Analysis

I-Past-1:

•Seeing suspicious truck

I-Past-1:

•Seeing suspicious truck

He-Past-Previous:

•Intent

•Driving to the target

He-Past-Previous:

•Intent

•Driving to the target

He-Past-Now:

•Traffic jam

•Turning back

He-Past-Now:

•Traffic jam

•Turning back

He-Past-Goal:

•Bomb the target

He-Past-Goal:

•Bomb the target

I-Now:

•bombing

•Recall suspicious activity

I-Now:

•bombing

•Recall suspicious activity

He-Past-Next:

•Take detour

He-Past-Next:

•Take detour

Episodicmemory

Workingmemory

Input-output

Cognitivemap

Semanticmemory

Proceduralmemory

Reward &punishment

Drivingengine

Remembered episode 1

Remembered episode 2

Imagery of the past

I-O:

•Report truck

I-O:

•Report truck

Boss: - Now you may go home, and I will take a train. Don’t forget to fill your tank.

Agent: - OK. Do you need a ride to the train station?

Boss: - No, thanks. I like to walk. Bye.

Agent: - Bye.

BICA BossLegend:

Example:

Dialogue between BICA and its Boss

Boss: - Now you may go home, and I will take a train. Don’t forget to fill your tank.

I-Nowhear Boss

I-Meta-GoalBoss happy

I-Metahelp Boss to achieve his goal

Boss-Nowwant Bica to go homewant Bica to fill the tankplan to take a traincommunicate this to Bica

BICA BossLegend:

Each white box above represents a mental state.

I-Previoushear Boss

Boss: - Now you may go home, and I will take a train. Don’t forget to fill your tank.

I-NowIdeas:Ideas:- go home- fill tank- offer a ride

I-Meta-GoalBoss happy

I-Metahelp Boss to achieve his goal

Boss-Nowwant Bica to go homewant Bica to fill the tankplan to take a traincommunicate this to Bica

BICA BossLegend:

Boss-GoalBICA gets homeBICA fills tankme take a train

Boss-Nexthear OKGet to the train station: ride BICA? walk?

I-Nowhear Boss

Boss: - Now you may go home, and I will take a train. Don’t forget to fill your tank.

I-NowIdeas:- go home- fill tank- offer a ride

I-Meta-GoalBoss happy

I-Metahelp Boss to achieve his goal

Boss-Nowwant BICA go homewant BICA fill tankplan to take a train

BICA BossLegend:

Boss-GoalBICA gets homeBICA fills tankme take a train

Boss-Nexthear OKGet to the train station: ride BICA? walk?

I-Imagined1heading home

I-Imagined2Filling tankGas pumpGas station

I-GoalHomeFull tank

I-Imagined3offer a ride

I-Previoushear Boss

I-Imag4Give a ride to Boss

Boss: - Now you may go home, and I will take a train. Don’t forget to fill your tank.

Agent: - OK. Do you need a ride to the train station?

I-NowHave plan:-fill tank-go homeIntent:- offer a ride

I-Meta-GoalBoss happy

I-Metahelp Boss to achieve his goal

Boss-Nowwant BICA go homewant BICA fill tankplan to take a train

BICA BossLegend:

Boss-GoalBICA gets homeBICA fills tankme take a train

Boss-Nexthear OKGet to the train station: ride BICA? walk?

I-Imag1heading home

I-Imag2Filling tankGas pumpGas station

I-GoalHomeFull tank

I-NextAcknowledgeOffer a rideIntent:go to gas st.

I-Prevhear Boss

I-Imag4Give a ride to Boss

I-Metahelp Boss to achieve his goalHave plan:- fill tank- go home

Boss: - Now you may go home, and I will take a train. Don’t forget to fill your tank.

Agent: - OK. Do you need a ride to the train station?

Boss: - No, thanks. I like to walk. Bye.I-PreviousIdeas:- go home- fill tank- offer a ride

I-Meta-GoalBoss happy

Boss-Nowwant BICA go homewant BICA fill tankplan to take a trainlike to walk

BICA BossLegend:

Boss-GoalBICA gets homeBICA fills tankme take a train

Boss-Nexthear OKGet to the train station: ride BICA? walk?

I-Imag1heading home

I-SubgoalFilling tankGas pumpGas station

I-GoalHomeFull tank

I-Nowspeakhear BossHave plan:- fill tank- go home

I-Nextsay ‘Bye’Go to gas station I-Imag4

Give a ride to Boss

Boss: - Now you may go home, and I will take a train. Don’t forget to fill your tank.

Agent: - OK. Do you need a ride to the train station?

Boss: - No, thanks. I like to walk. Bye.

Agent: - Bye.I-PastIdeas:- go home- fill tank- offer a ride

I-Meta-GoalBoss happy

I-Metahelp Boss to achieve his goalHave plan:- fill tank- go home

Boss-Nowwant BICA go homewant BICA fill tankplan walking to trainshear ‘Bye’

BICA BossLegend:

Boss-GoalBICA gets homeBICA fills tankme take a train

Boss-NextWalk to the train station

I-Imag1heading home

I-SubgoalFilling tankGas pumpGas station

I-GoalHomeFull tank

I-Previoushear Bossacknowledge

I-Nowsay ‘Bye’

I-NextGo to gas station

GMU BICA was implemented…

Now, why does it get us closer to human-like learning and cognitive growth?

Hierarchy of intelligent agent architectures

5 meta-cognitive and self-aware

capable of modeling mental states of agents, including own mental states, based on the concept of a self

4 reflective capable of modeling internally the environment and behavior of entities in it

3 proactive, or deliberative

capable of reasoning, planning, exploration and decision making

2 reactive, or adaptive

capable of lower forms of learning and adaptation

1 reflexive based on a set of pre-programmed behavioral responses

Concept of self-regulated learning (SRL)

Bic

a

Input-Output

ProceduralMemory

EngineSelf-

consequating

SemanticMemory

EpisodicMemoryWorking

Memory

CognitiveMap

Student

CB

LE

Des

igne

r

Teacher

Cog

nitiv

e C

onst

ruct

or

Interface

CC~> factorize_

2•3 3•5, 1•15

6x2+19x+15(_x+_)(_x+_)(3x+1)(2x+15)

Given:Goal:

Try:

Our approach to building an SRL partner

An expected snapshot of Cognitive Constructor

Illustrative example

Problem 1: Parents have two children. One child is a boy. What is the probability that the other child is a boy?

Problem 2: In families with two kids, do boys on average have more sisters, as compared to girls?BB BG

GB GG

Recognizing an SRL-naive student working on problem 1

She-Previous Problem 1 – task data:Parents have two children. One child is a boy. Task goal:Find the probability that the other child is a boy.

She-Next (predicted)

Therefore, the probability that the other child is a boy is 50%.

I-NextTest this hypothesis,expect confirmation

I-NowHypothesis: she is spontaneously constructing an intuitive solution

She-Alt-Next (possible)

Therefore, all four possible outcomes for a family with two kids (BB, BG, GB, GG) are equally likely.

I-Imagine-AlternativeShe is planning to construct the sample space and then use the definition of probability

Input-OutputUtter: “Would you say that, therefore, the probability for the second child to be a boy is 50%?”

I-MetaConstruct a model of her mind

I-PrevAssume that she is naive

She-Now (observed)

Each child is equally likely to be born as a boy or as a girl.

The sex of each child is determined independently of other children.

Scaffolding an SRL-naive student working on problem 2

She-PreviousProblem 2: In families with two kids, are boys more likely to have a sister, compared to girls?

She-Now (observed by Bica)

If a family has a boy and a girl, then the boy has a sister, but the girl does not

If one child in a family with two kids is a boy, then the other child is more likely to be a girl

She-Next (anticipated by Bica)Therefore, a boy is more likely to have a sister than a brother

I-NextMake sure the student learns the correct general approach

I-NowShe is constructing an intuitive solution, using the known solution of Problem 1

Input-Output

- Try to select an approach that corresponds to the task instead of intuitive guessing. For example, imagine doing a survey of kids, and construct a representative sample.

She-Past

Problem 1 solution

Conclusions

Having GMU BICA as a model of the student mind as the core of Cognitive Constructor will allow us to select the right level of SRL feedback in each given case.

Creating and using a higher-level model of student learner in education is a step toward creating a computational equivalent of student learner

Scalability of this approach should be the main criterion for success

Credits:

Ken De Jong Giorgio Ascoli Anastasia Kitsantas Nada Dabbagh Mark Coletti Deepankar Sharma Robert Lakatos Tashfeen Bhimdi

DARPA IPTO BICA Grant “A Biologically Inspired Self-Aware Cognitive Architecture”


Recommended