PowerPoint Presentation
Meta-Reasoning
Metacognition
Learning by Correcting Mistakes
Meta-Reasoning
Ethics in Artificial Intelligence
Lesson Preview
Mistakes in knowledge, reasoning, and learning
Gaps in knowledge and reasoning
Strategy selection and integration
Meta-meta-reasoning?
Goal-based autonomy
Object
Cup
is
Object
Bottom
has
Bottom
Flat
is
Object
Concavity
has
Object
Light
is
Object
Handle
has
Object
Porcelain
made-of
Object
Decoration
has
Object
Liquids
carries
Object
Liftable
is
Object
Drinking
enables
Object
Stable
is
Cognitive System
Reaction
Deliberation
Metacognition
Reasoning
Learning
Memory
Input
Output
C
B
A
A
C
B
Current State
Goal State
A on B
B on C
C on D
D on Table
D
D
A on B
B on C
C on Table
D on Table
= 1
Cognitive System
Reaction
Deliberation
Metacognition
Reasoning
Learning
Memory
Input
Output
Object
Cup
is
Object
Concavity
has
Object
Light
is
Object
Handle
has
Object
Liquids
carries
Object
Liftable
is
Object
Drinking
enables
Handle
Fixed
is
Object
Manipulable
is
Object
Orientable
is
Object
is
Object
Drinking
enables
Object
Stable
is
Object
Heat
protects
against
Object
Heat transfer
limits
Object
Thick sides
has
Object
Thick sides
made-of
Mug
Object
is
Object
Drinking
enables
Object
Stable
is
Object
Heat
protects
against
Object
Heat transfer
limits
Object
Thick sides
has
Object
Thick sides
made-of
Oven Mitt
Mug
C
B
A
A
C
B
Current State
Goal State
A on B
B on C
C on D
D on Table
D
D
A on B
B on C
C on Table
D on Table
= 1
Cognitive System
Reaction
Deliberation
Metacognition
Reasoning
Learning
Memory
Input
Output
Object
Cup
is
Object
Bottom
has
Bottom
Flat
is
Object
Concavity
has
Object
Light
is
Object
Handle
has
Object
Porcelain
made-of
Object
Decoration
has
Object
Liquids
carries
Object
Liftable
is
Object
Drinking
enables
Object
Stable
is
Cognitive System
Reaction
Deliberation
Metacognition
Reasoning
Learning
Memory
Input
Output
Means-Ends Analysis
Problem Reduction
Generate & Test
Analogical Reasoning
Planning
Case-Based Reasoning
Constraint Propagation
Means-Ends Analysis
Problem Reduction
Planning
Metacognition
Constraint Propagation
Generate & Test
Analogical Reasoning
Case-Based Reasoning
Version Spaces
Explanation-Based
Learning
Incremental
Concept Learning
Metacognition
Learning by
Recording Cases
Classification
Means-Ends Analysis
Problem Reduction
Planning
Metacognition
Constraint Propagation
Generate & Test
Analogical Reasoning
Case-Based Reasoning
Metacognition
Case-Based Reasoning
Metacognition
Case-Based Reasoning
Retrieval
Adaptation
Evaluation
Storage
Metacognition
Case-Based Reasoning
Retrieval
Adaptation
Evaluation
Storage
Metacognition
Case-Based Reasoning
Retrieval
Adaptation
Evaluation
Storage
By rules?
By model?
Recursively?
Metacognition
Case-Based Reasoning
Retrieval
Adaptation
Evaluation
Storage
By rules?
Rule 2?
Rule 1?
Rule 3?
Means-Ends Analysis
Problem Reduction
Planning
Metacognition
Constraint Propagation
Generate & Test
Analogical Reasoning
Case-Based Reasoning
Metacognition
Deliberation
Meta-Metacognition
N-Metacognition
Is this a good way to think about levels of metacognition?
Yes, because it is possible to think about every successive level.
No, because there is a maximum level of metacognition possible.
No, because each level of metacognition is conceptually identical, so they are better represented as self-referential.
No, because there is no need to distinguish between metacognition and deliberation.
Metacognition
Deliberation
Is this a good way to think about levels of metacognition?
Yes, because it is possible to think about every successive level.
No, because there is a maximum level of metacognition possible.
No, because each level of metacognition is conceptually identical, so they are better represented as self-referential.
No, because there is no need to distinguish between metacognition and deliberation.
Reaction
Deliberation
Metacognition
Reasoning
Learning
Memory
Input
Output
Object
Cup
is
Object
Concavity
has
Object
Light
is
Object
Handle
has
Object
Liquids
carries
Object
Liftable
is
Object
Drinking
enables
Handle
Fixed
is
Object
Manipulable
is
Object
Orientable
is
Goal: Painted(Ladder)
paint-ladder
On(Robot, Floor)
Dry(Ladder) Dry(Ceiling)
On(Robot, Floor)
Dry(Ladder) Dry(Ceiling) Painted(Ladder)
Goal: Painted(Ceiling)
On(Robot, Floor)
Dry(Ladder) Dry(Ceiling)
On(Robot, Ladder)
Dry(Ladder) Dry(Ceiling)
climb-ladder
paint-ceiling
On(Robot, Ladder)
Dry(Ladder) Dry(Ceiling) Painted(Ceiling)
climb-ladder:
Precondition:
On(Robot, Floor)
Dry(Ladder)
Postcondition:
On(Robot, Ladder)
inning : 5th
portion : bottom
game : 131
weather : windy
runners : 1st, 3rd
outs : 1
batter : Pierzynski
average : .283
bats : left-handed
score : 1-4
goal : pitch
pitch : throw-fast-ball
result : homerun
(r8) If two operators selected and one has an episode with result homerun
then prefer other operator
chunking
Version Spaces
Specific
General
Hypothesis Space
H1
H2
H3
H4
H5
H6
H7
H8
H9
HN
Data Space
D1
D2
D3
D4
D5
D6
D7
D8
D9
DN
Abstract
Map
Refine
Treatment Space
T1
T2
T3
T4
T5
T6
T7
T8
T9
TN
Is this a foo?
Yes
No
This is not a foo.
Modify the concept on the right to specialize based on this example.
Brick
Brick
Brick
Brick
supports
supports
supports
supports
touches
touches
Current Concept
Assignment
How would you use meta-reasoning to design an agent that could answer Ravens progressive matrices?
To recap
Resolving mistakes and gaps
Strategy selection and integration
Meta-meta-reasoning?
Goal-based autonomy
/docProps/thumbnail.jpeg
Reviews
There are no reviews yet.