logic and arti cial intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · tilburg university...
TRANSCRIPT
![Page 1: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/1.jpg)
Logic and Artificial IntelligenceLecture 5
Eric Pacuit
Currently Visiting the Center for Formal Epistemology, CMU
Center for Logic and Philosophy of ScienceTilburg University
ai.stanford.edu/∼[email protected]
September 14, 2011
Logic and Artificial Intelligence 1/19
![Page 2: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/2.jpg)
Knowledge, belief and (un)awareness
Logic and Artificial Intelligence 2/19
![Page 3: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/3.jpg)
Adding Beliefs
Epistemic Models: M = 〈W , {∼i}i∈A,V 〉
Truth: M,w |= ϕ is defined as follows:
I M,w |= p iff w ∈ V (p) (with p ∈ At)
I M,w |= ¬ϕ if M,w 6|= ϕ
I M,w |= ϕ ∧ ψ if M,w |= ϕ and M,w |= ψ
I M,w |= Kiϕ if for each v ∈W , if w∼iv , then M, v |= ϕ
Logic and Artificial Intelligence 3/19
![Page 4: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/4.jpg)
Adding Beliefs
Epistemic-Doxastic Models: M = 〈W , {∼i}i∈A, {�i}i∈A,V 〉
Truth: M,w |= ϕ is defined as follows:
I M,w |= p iff w ∈ V (p) (with p ∈ At)
I M,w |= ¬ϕ if M,w 6|= ϕ
I M,w |= ϕ ∧ ψ if M,w |= ϕ and M,w |= ψ
I M,w |= Kiϕ if for each v ∈W , if w∼iv , then M, v |= ϕ
Logic and Artificial Intelligence 3/19
![Page 5: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/5.jpg)
Adding BeliefsEpistemic-Doxastic Models: M = 〈W , {∼i}i∈A, {�i}i∈A,V 〉
Plausibility Relation: �i⊆W ×W . w �i v means
“v is at least as plausibile as w .”
Properties of �i : reflexive, transitive, complete and well-founded.
Most Plausible: For X ⊆W , let
Min�i (X ) = {v ∈W | v �i w for all w ∈ X }
Assumptions:
plausibility implies possibility: if w �i v then w ∼i v .locally-connected: if w ∼i v then either w �i v or v �i w .
Logic and Artificial Intelligence 3/19
![Page 6: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/6.jpg)
Adding BeliefsEpistemic-Doxastic Models: M = 〈W , {∼i}i∈A, {�i}i∈A,V 〉
Plausibility Relation: �i⊆W ×W . w �i v means
“v is at least as plausibile as w .”
Properties of �i : reflexive, transitive, and well-founded.
Most Plausible: For X ⊆W , let
Min�i (X ) = {v ∈W | v �i w for all w ∈ X }
Assumptions:
plausibility implies possibility: if w �i v then w ∼i v .locally-connected: if w ∼i v then either w �i v or v �i w .
Logic and Artificial Intelligence 3/19
![Page 7: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/7.jpg)
Adding BeliefsEpistemic-Doxastic Models: M = 〈W , {∼i}i∈A, {�i}i∈A,V 〉
Plausibility Relation: �i⊆W ×W . w �i v means
“v is at least as plausibile as w .”
Properties of �i : reflexive, transitive, and well-founded.
Most Plausible: For X ⊆W , let
Min�i (X ) = {v ∈W | v �i w for all w ∈ X }
Assumptions:
plausibility implies possibility: if w �i v then w ∼i v .locally-connected: if w ∼i v then either w �i v or v �i w .
Logic and Artificial Intelligence 3/19
![Page 8: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/8.jpg)
Adding BeliefsEpistemic-Doxastic Models: M = 〈W , {∼i}i∈A, {�i}i∈A,V 〉
Plausibility Relation: �i⊆W ×W . w �i v means
“v is at least as plausibile as w .”
Properties of �i : reflexive, transitive, and well-founded.
Most Plausible: For X ⊆W , let
Min�i (X ) = {v ∈W | v �i w for all w ∈ X }
Assumptions:
1. plausibility implies possibility: if w �i v then w ∼i v .2. locally-connected: if w ∼i v then either w �i v or v �i w .
Logic and Artificial Intelligence 3/19
![Page 9: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/9.jpg)
Adding Beliefs
Epistemic-Doxastic Models: M = 〈W , {∼i}i∈A, {�i}i∈A,V 〉
Truth: M,w |= ϕ is defined as follows:
I M,w |= p iff w ∈ V (p) (with p ∈ At)
I M,w |= ¬ϕ if M,w 6|= ϕ
I M,w |= ϕ ∧ ψ if M,w |= ϕ and M,w |= ψ
I M,w |= Kiϕ if for each v ∈W , if w∼iv , then M, v |= ϕ
I M,w |= Biϕ if for each v ∈ Min�i ([w ]i ), M, v |= ϕ[w ]i = {v | w ∼i v} is the agent’s information cell.
Logic and Artificial Intelligence 3/19
![Page 10: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/10.jpg)
Grades of Doxastic Strength
wv1v0 v2
Suppose that w is the current state.
Knowledge (KP)
Belief (BP)
Safe Belief (�P)
Strong Belief (BsP)
Logic and Artificial Intelligence 4/19
![Page 11: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/11.jpg)
Grades of Doxastic Strength
wv1v0 v2
Suppose that w is the current state.
Knowledge (KP)
Belief (BP)
Robust Belief (�P)
Strong Belief (BsP)
Logic and Artificial Intelligence 4/19
![Page 12: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/12.jpg)
Grades of Doxastic Strength
¬Pw
¬Pv1
P
v0
P
v2
Suppose that w is the current state.
I Belief (BP)
Robust Belief (�P)
Strong Belief (BsP)
Knowledge (KP)
Logic and Artificial Intelligence 4/19
![Page 13: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/13.jpg)
Grades of Doxastic Strength
P
w
¬Pv1
P
v0
P
v2
Suppose that w is the current state.
I Belief (BP)
I Robust Belief (�P)
Strong Belief (BsP)
Knowledge (KP)
Logic and Artificial Intelligence 4/19
![Page 14: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/14.jpg)
Grades of Doxastic Strength
P
w
P
v1
¬Pv0
P
v2
Suppose that w is the current state.
I Belief (BP)
I Robust Belief (�P)
I Strong Belief (BsP)
Knowledge (KP)
Logic and Artificial Intelligence 4/19
![Page 15: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/15.jpg)
Grades of Doxastic Strength
P
w
P
v1
P
v0
P
v2
Suppose that w is the current state.
I Belief (BP)
I Robust Belief (�P)
I Strong Belief (BsP)
I Knowledge (KP)
Logic and Artificial Intelligence 4/19
![Page 16: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/16.jpg)
Conditional Beliefs
I w1 ∼ w2 ∼ w3
w1 � w2 and w2 � w1 (w1 and w2
are equi-plausbile)
w1 ≺ w3 (w1 � w3 and w3 6� w1)
w2 ≺ w3 (w2 � w3 and w3 6� w2)
{w1,w2} ⊆ Min�([wi ])
w3
w2w1
A
B
D
E
ϕ
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Logic and Artificial Intelligence 5/19
![Page 17: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/17.jpg)
Conditional Beliefs
I w1 ∼ w2 ∼ w3
I w1 � w2 and w2 � w1 (w1 and w2
are equi-plausbile)
I w1 ≺ w3 (w1 � w3 and w3 6� w1)
I w2 ≺ w3 (w2 � w3 and w3 6� w2)
{w1,w2} ⊆ Min�([wi ])
w3
w2w1
A
B
D
E
ϕ
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Logic and Artificial Intelligence 5/19
![Page 18: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/18.jpg)
Conditional Beliefs
I w1 ∼ w2 ∼ w3
I w1 � w2 and w2 � w1 (w1 and w2
are equi-plausbile)
I w1 ≺ w3 (w1 � w3 and w3 6� w1)
I w2 ≺ w3 (w2 � w3 and w3 6� w2)
I {w1,w2} ⊆ Min�([wi ])
w3
w2w1
A
B
D
E
ϕ
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Logic and Artificial Intelligence 5/19
![Page 19: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/19.jpg)
Conditional Beliefs
A
B
D
E
ϕ
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Logic and Artificial Intelligence 5/19
![Page 20: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/20.jpg)
Conditional Beliefs
ϕ
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Logic and Artificial Intelligence 5/19
![Page 21: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/21.jpg)
Conditional Beliefs
ϕ
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Logic and Artificial Intelligence 5/19
![Page 22: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/22.jpg)
Conditional Beliefs, continued
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Is Bϕ→ Bψϕ valid? No
What about Bϕ→ Bψϕ ∨ B¬ψϕ? Yes (but need connectedness...)
What does it mean if B¬ϕ⊥ is true at a state? The agent knows ϕ
Logic and Artificial Intelligence 6/19
![Page 23: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/23.jpg)
Conditional Beliefs, continued
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Is Bϕ→ Bψϕ valid? No
What about Bϕ→ Bψϕ ∨ B¬ψϕ? Yes (but need connectedness...)
What does it mean if B¬ϕ⊥ is true at a state? The agent knows ϕ
Logic and Artificial Intelligence 6/19
![Page 24: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/24.jpg)
Conditional Beliefs, continued
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Is Bϕ→ Bψϕ valid? No
What about Bϕ→ Bψϕ ∨ B¬ψϕ? Yes (but need connectedness...)
What does it mean if B¬ϕ⊥ is true at a state? The agent knows ϕ
Logic and Artificial Intelligence 6/19
![Page 25: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/25.jpg)
Conditional Beliefs, continued
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Is Bϕ→ Bψϕ valid? No
What about Bϕ→ Bψϕ ∨ B¬ψϕ? Yes (but need connectedness...)
What does it mean if B¬ϕ⊥ is true at a state? The agent knows ϕ
Logic and Artificial Intelligence 6/19
![Page 26: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/26.jpg)
Conditional Beliefs, continued
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Is Bϕ→ Bψϕ valid? No
What about Bϕ→ Bψϕ ∨ B¬ψϕ? Yes (but need connectedness...)
What does it mean if B¬ϕ⊥ is true at a state? The agent knows ϕ
Logic and Artificial Intelligence 6/19
![Page 27: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/27.jpg)
Conditional Beliefs, continued
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Is Bϕ→ Bψϕ valid? No
What about Bϕ→ Bψϕ ∨ B¬ψϕ? Yes (but need connectedness...)
What does it mean if B¬ϕ⊥ is true at a state? The agent knows ϕ
Logic and Artificial Intelligence 6/19
![Page 28: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/28.jpg)
Conditional Beliefs, continued
Bϕi ψ: Agent i believes ψ, given that ϕ is true.
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Is Bϕ→ Bψϕ valid? No
What about Bϕ→ Bψϕ ∨ B¬ψϕ? Yes (but need connectedness...)
What does it mean if B¬ϕ⊥ is true at a state? The agent knows ϕ
Logic and Artificial Intelligence 6/19
![Page 29: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/29.jpg)
Conditional Beliefs, continued
M,w |= Bϕi ψ if for each v ∈ Min�i ([w ]i ∩ [[ϕ]]), M, v |= ϕwhere [[ϕ]] = {w | M,w |= ϕ}
Core Logical Principles:
1. Bϕϕ
2. Bϕψ → Bϕ(ψ ∨ χ)
3. (Bϕψ1 ∧ Bϕψ2)→ Bϕ(ψ1 ∧ ψ2)
4. (Bϕ1ψ ∧ Bϕ2ψ)→ Bϕ1∨ϕ2ψ
5. (Bϕψ ∧ Bψϕ)→ (Bϕχ↔ Bψχ)
J. Burgess. Quick completeness proofs for some logics of conditionals. NotreDame Journal of Formal Logic 22, 76 – 84, 1981.
Logic and Artificial Intelligence 6/19
![Page 30: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/30.jpg)
Types of Beliefs: Logical Characterizations
I M,w |= Kiϕ iff M,w |= Bψi ϕ for all ψi knows ϕ iff i continues to believe ϕ given any new information
I M,w |= �iϕ iff M,w |= Bψi ϕ for all ψ with M,w |= ψ.i robustly believes ϕ iff i continues to believe ϕ given any true
formula.
I M,w |= Bsi ϕ iff M,w |= Biϕ and M,w |= Bψi ϕ for all ψ
with M,w |= ¬Ki (ψ → ¬ϕ).i strongly believes ϕ iff i believes ϕ and continues to believe ϕ given
any evidence (truthful or not) that is not known to contradict ϕ.
Logic and Artificial Intelligence 7/19
![Page 31: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/31.jpg)
Types of Beliefs: Logical Characterizations
I M,w |= Kiϕ iff M,w |= Bψi ϕ for all ψi knows ϕ iff i continues to believe ϕ given any new information
I M,w |= �iϕ iff M,w |= Bψi ϕ for all ψ with M,w |= ψ.i robustly believes ϕ iff i continues to believe ϕ given any true
formula.
I M,w |= Bsi ϕ iff M,w |= Biϕ and M,w |= Bψi ϕ for all ψ
with M,w |= ¬Ki (ψ → ¬ϕ).i strongly believes ϕ iff i believes ϕ and continues to believe ϕ given
any evidence (truthful or not) that is not known to contradict ϕ.
Logic and Artificial Intelligence 7/19
![Page 32: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/32.jpg)
Types of Beliefs: Logical Characterizations
I M,w |= Kiϕ iff M,w |= Bψi ϕ for all ψi knows ϕ iff i continues to believe ϕ given any new information
I M,w |= �iϕ iff M,w |= Bψi ϕ for all ψ with M,w |= ψ.i robustly believes ϕ iff i continues to believe ϕ given any true
formula.
I M,w |= Bsi ϕ iff M,w |= Biϕ and M,w |= Bψi ϕ for all ψ
with M,w |= ¬Ki (ψ → ¬ϕ).i strongly believes ϕ iff i believes ϕ and continues to believe ϕ given
any evidence (truthful or not) that is not known to contradict ϕ.
Logic and Artificial Intelligence 7/19
![Page 33: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/33.jpg)
Unawareness
Why would an agent not know some fact ϕ? (i.e., why would¬Kiϕ be true?)
I The agent many or may not believe ϕ, but has not ruled outall the ¬ϕ-worlds
I The agent may believe ϕ and ruled-out the ¬ϕ-worlds, butthis was based on “bad” evidence, or was not justified, or theagent was “epistemically lucky” (eg., Gettier cases),...
I The agent has not yet entertained possibilities relevant to thetruth of ϕ (the agent is unaware of ϕ).
Logic and Artificial Intelligence 8/19
![Page 34: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/34.jpg)
Unawareness
Why would an agent not know some fact ϕ? (i.e., why would¬Kiϕ be true?)
I The agent many or may not believe ϕ, but has not ruled outall the ¬ϕ-worlds
I The agent may believe ϕ and ruled-out the ¬ϕ-worlds, butthis was based on “bad” evidence, or was not justified, or theagent was “epistemically lucky” (eg., Gettier cases),...
I The agent has not yet entertained possibilities relevant to thetruth of ϕ (the agent is unaware of ϕ).
Logic and Artificial Intelligence 8/19
![Page 35: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/35.jpg)
Unawareness
Why would an agent not know some fact ϕ? (i.e., why would¬Kiϕ be true?)
I The agent many or may not believe ϕ, but has not ruled outall the ¬ϕ-worlds
I The agent may believe ϕ and ruled-out the ¬ϕ-worlds, butthis was based on “bad” evidence, or was not justified, or theagent was “epistemically lucky” (eg., Gettier cases),...
I The agent has not yet entertained possibilities relevant to thetruth of ϕ (the agent is unaware of ϕ).
Logic and Artificial Intelligence 8/19
![Page 36: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/36.jpg)
Unawareness
Why would an agent not know some fact ϕ? (i.e., why would¬Kiϕ be true?)
I The agent many or may not believe ϕ, but has not ruled outall the ¬ϕ-worlds
I The agent may believe ϕ and ruled-out the ¬ϕ-worlds, butthis was based on “bad” evidence, or was not justified, or theagent was “epistemically lucky” (eg., Gettier cases),...
I The agent has not yet entertained possibilities relevant to thetruth of ϕ (the agent is unaware of ϕ).
Logic and Artificial Intelligence 8/19
![Page 37: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/37.jpg)
Can we model unawareness in state-space models?
E. Dekel, B. Lipman and A. Rustichini. Standard State-Space Models PrecludeUnawareness. Econometrica, 55:1, pp. 159 - 173 (1998).
Logic and Artificial Intelligence 9/19
![Page 38: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/38.jpg)
Can we model unawareness in state-space models?
E. Dekel, B. Lipman and A. Rustichini. Standard State-Space Models PrecludeUnawareness. Econometrica, 55:1, pp. 159 - 173 (1998).
Logic and Artificial Intelligence 9/19
![Page 39: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/39.jpg)
Sherlock Holmes
While Watson never reports it, Sherlock Holmes oncenoted an even more curious incident, that of the dog thatbarked and the cat that howled in the night.
WhenWatson objected that the dog did not bark and the catdid not Howl, Holmes replied “that is the curious incidentto which I refer.” Holmes knew that this meant that noone, neither man nor dog, had intruded on the premisesthe previous night. For had a man intruded, the dogwould have barked. Had a dog intruded, the cat wouldhave howled. Hence the lack of either of these twosignals means that there could not have been a human orcanine intruder.
Logic and Artificial Intelligence 10/19
![Page 40: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/40.jpg)
Sherlock Holmes
While Watson never reports it, Sherlock Holmes oncenoted an even more curious incident, that of the dog thatbarked and the cat that howled in the night. WhenWatson objected that the dog did not bark and the catdid not Howl, Holmes replied “that is the curious incidentto which I refer.”
Holmes knew that this meant that noone, neither man nor dog, had intruded on the premisesthe previous night. For had a man intruded, the dogwould have barked. Had a dog intruded, the cat wouldhave howled. Hence the lack of either of these twosignals means that there could not have been a human orcanine intruder.
Logic and Artificial Intelligence 10/19
![Page 41: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/41.jpg)
Sherlock Holmes
While Watson never reports it, Sherlock Holmes oncenoted an even more curious incident, that of the dog thatbarked and the cat that howled in the night. WhenWatson objected that the dog did not bark and the catdid not Howl, Holmes replied “that is the curious incidentto which I refer.” Holmes knew that this meant that noone, neither man nor dog, had intruded on the premisesthe previous night.
For had a man intruded, the dogwould have barked. Had a dog intruded, the cat wouldhave howled. Hence the lack of either of these twosignals means that there could not have been a human orcanine intruder.
Logic and Artificial Intelligence 10/19
![Page 42: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/42.jpg)
Sherlock Holmes
While Watson never reports it, Sherlock Holmes oncenoted an even more curious incident, that of the dog thatbarked and the cat that howled in the night. WhenWatson objected that the dog did not bark and the catdid not Howl, Holmes replied “that is the curious incidentto which I refer.” Holmes knew that this meant that noone, neither man nor dog, had intruded on the premisesthe previous night. For had a man intruded, the dogwould have barked. Had a dog intruded, the cat wouldhave howled. Hence the lack of either of these twosignals means that there could not have been a human orcanine intruder.
Logic and Artificial Intelligence 10/19
![Page 43: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/43.jpg)
Modeling Watson’s Unawareness
dw1
dw2 cw3
Logic and Artificial Intelligence 11/19
![Page 44: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/44.jpg)
Modeling Watson’s Unawareness
I Let E = {w2} be the event that there was anintruder.
K (E ) = {w2} (at w2, Watson knows there isa human intruder) and −K (E ) = {w1,w3}K (−K (E )) = {w3} (at w3, Watson knowsthat she does not know E ), and−K (−K (E )) = {w1,w2}−K (E ) ∩ −K (−K (E )) = {w1} and, in fact,⋂∞
i=1(−K )i (E ) = {w1}
dw1
dw2 cw3
Logic and Artificial Intelligence 11/19
![Page 45: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/45.jpg)
Modeling Watson’s Unawareness
I Let E = {w2} be the event that there was anintruder.
I K (E ) = {w2} (at w2, Watson knows there isa human intruder) and −K (E ) = {w1,w3}K (−K (E )) = {w3} (at w3, Watson knowsthat she does not know E ), and−K (−K (E )) = {w1,w2}−K (E ) ∩ −K (−K (E )) = {w1} and, in fact,⋂∞
i=1(−K )i (E ) = {w1}
dw1
dw2 cw3
Logic and Artificial Intelligence 11/19
![Page 46: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/46.jpg)
Modeling Watson’s Unawareness
I Let E = {w2} be the event that there was anintruder.
I K (E ) = {w2} (at w2, Watson knows there isa human intruder) and −K (E ) = {w1,w3}
I K (−K (E )) = {w3} (at w3, Watson knowsthat she does not know E ), and−K (−K (E )) = {w1,w2}−K (E ) ∩ −K (−K (E )) = {w1} and, in fact,⋂∞
i=1(−K )i (E ) = {w1}
dw1
dw2 cw3
Logic and Artificial Intelligence 11/19
![Page 47: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/47.jpg)
Modeling Watson’s Unawareness
I Let E = {w2} be the event that there was anintruder.
I K (E ) = {w2} (at w2, Watson knows there isa human intruder) and −K (E ) = {w1,w3}
I K (−K (E )) = {w3} (at w3, Watson knowsthat she does not know E ), and−K (−K (E )) = {w1,w2}
I −K (E ) ∩ −K (−K (E )) = {w1} and, in fact,⋂∞i=1(−K )i (E ) = {w1}
dw1
dw2 cw3
Logic and Artificial Intelligence 11/19
![Page 48: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/48.jpg)
Modeling Watson’s Unawareness
I E = {w2}I K (E ) = {w2}, −K (E ) = {w1,w3}I K (−K (E )) = {w3} ,−K (−K (E )) = {w1,w2}
I −K (E ) ∩ −K (−K (E )) = {w1},⋂∞i=1(−K )i (E ) = {w1}
dw1
dw2 cw3
Let U(F ) =⋂∞
i=1(−K )i (F ). Then,I U(∅) = U(W ) = U({w1}) = U({w2,w3}) = ∅I U(E ) = U({w3}) = U({w1,w3}) = U({w1,w2} = {w1}
Then, U(E ) = {w1} and U(U(E )) = U({w1}) = ∅Logic and Artificial Intelligence 11/19
![Page 49: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/49.jpg)
Modeling Watson’s Unawareness
I E = {w2}I K (E ) = {w2}, −K (E ) = {w1,w3}I K (−K (E )) = {w3} ,−K (−K (E )) = {w1,w2}
I −K (E ) ∩ −K (−K (E )) = {w1},⋂∞i=1(−K )i (E ) = {w1}
dw1
dw2 cw3
Let U(F ) =⋂∞
i=1(−K )i (F ). Then,I U(∅) = U(W ) = U({w1}) = U({w2,w3}) = ∅I U(E ) = U({w3}) = U({w1,w3}) = U({w1,w2} = {w1}
Then, U(E ) = {w1} and U(U(E )) = U({w1}) = ∅Logic and Artificial Intelligence 11/19
![Page 50: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/50.jpg)
Properties of Unawareness
1. Uϕ→ (¬Kϕ ∧ ¬K¬Kϕ)
2. ¬KUϕ
3. Uϕ→ UUϕ
Theorem. In any logic where U satisfies the above axiomschemes, we have
1. If K satisfies Necessitation (from ϕ infer Kϕ), then for allformulas ϕ, ¬Uϕ is derivable (the agent is aware ofeverything); and
2. If K satisfies Monotonicity (from ϕ→ ψ infer Kϕ→ ψ), thenfor all ϕ and ψ, Uϕ→ ¬Kψ is derivable (if the agent isunaware of something then the agent does not knowanything).
Logic and Artificial Intelligence 12/19
![Page 51: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/51.jpg)
Properties of Unawareness
1. Uϕ→ (¬Kϕ ∧ ¬K¬Kϕ)
2. ¬KUϕ
3. Uϕ→ UUϕ
Theorem. In any logic where U satisfies the above axiomschemes, we have
1. If K satisfies Necessitation (from ϕ infer Kϕ), then for allformulas ϕ, ¬Uϕ is derivable (the agent is aware ofeverything); and
2. If K satisfies Monotonicity (from ϕ→ ψ infer Kϕ→ ψ), thenfor all ϕ and ψ, Uϕ→ ¬Kψ is derivable (if the agent isunaware of something then the agent does not knowanything).
Logic and Artificial Intelligence 12/19
![Page 52: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/52.jpg)
Properties of Unawareness
1. Uϕ→ (¬Kϕ ∧ ¬K¬Kϕ)
2. ¬KUϕ
3. Uϕ→ UUϕ
Theorem. In any logic where U satisfies the above axiomschemes, we have
1. If K satisfies Necessitation (from ϕ infer Kϕ), then for allformulas ϕ, ¬Uϕ is derivable (the agent is aware ofeverything); and
2. If K satisfies Monotonicity (from ϕ→ ψ infer Kϕ→ ψ), thenfor all ϕ and ψ, Uϕ→ ¬Kψ is derivable (if the agent isunaware of something then the agent does not knowanything).
Logic and Artificial Intelligence 12/19
![Page 53: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/53.jpg)
Properties of Unawareness
1. Uϕ→ (¬Kϕ ∧ ¬K¬Kϕ)
2. ¬KUϕ
3. Uϕ→ UUϕ
Theorem. In any logic where U satisfies the above axiomschemes, we have
1. If K satisfies Necessitation (from ϕ infer Kϕ), then for allformulas ϕ, ¬Uϕ is derivable (the agent is aware ofeverything); and
2. If K satisfies Monotonicity (from ϕ→ ψ infer Kϕ→ ψ), thenfor all ϕ and ψ, Uϕ→ ¬Kψ is derivable (if the agent isunaware of something then the agent does not knowanything).
Logic and Artificial Intelligence 12/19
![Page 54: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/54.jpg)
B. Schipper. Online Bibliography on Models of Unawareness. http://www.
econ.ucdavis.edu/faculty/schipper/unaw.htm.
J. Halpern. Alternative semantics for unawareness. Games and Economic Be-havior, 37, 321-339, 2001.
Logic and Artificial Intelligence 13/19
![Page 55: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/55.jpg)
Epistemic-Probability Models
Logic and Artificial Intelligence 14/19
![Page 56: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/56.jpg)
Adding Probabilities
Epistemic-Probability Model: M = 〈W , {∼i}i∈A, {Pi}i∈A,V 〉where each ∼i is an equivalence relation on W is an epistemicmodel and Pi : W → ∆(W ) assigns to each state a probabilitymeasure over W , and V is a valuation function.
∆(W ) = {p : W → [0, 1] | p is a probability measure })
Write pwi for the i ’s probability measure at state w . We make twonatural assumptions:
1. For all v ∈W , if pwi (v) > 0 then pwi = pvi ; and
2. For all v 6∈ [w ]i , pwi (v) = 0.
Logic and Artificial Intelligence 15/19
![Page 57: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/57.jpg)
Adding Probabilities
Epistemic-Probability Model: M = 〈W , {∼i}i∈A, {Pi}i∈A,V 〉where each ∼i is an equivalence relation on W is an epistemicmodel and Pi : W → ∆(W ) assigns to each state a probabilitymeasure over W , and V is a valuation function.
∆(W ) = {p : W → [0, 1] | p is a probability measure })
Write pwi for the i ’s probability measure at state w . We make twonatural assumptions:
1. For all v ∈W , if pwi (v) > 0 then pwi = pvi ; and
2. For all v 6∈ [w ]i , pwi (v) = 0.
Logic and Artificial Intelligence 15/19
![Page 58: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/58.jpg)
Common Prior
Epistemic-Probabilistic Models: M = 〈W , {∼i}i∈A, p,V 〉
Common Prior: p : W → [0, 1] is a probability measure (assumeW finite)
Truth: M,w |= ϕ is defined as follows:
I M,w |= p iff w ∈ V (p) (with p ∈ At)
I M,w |= ¬ϕ if M,w 6|= ϕ
I M,w |= ϕ ∧ ψ if M,w |= ϕ and M,w |= ψ
I M,w |= Kiϕ if for each v ∈W , if w∼iv , then M, v |= ϕ
I M,w |= B rϕ iff p([[ϕ]] | [w ]i ) = p([[ϕ]]∩[w ]i )p([w ]i )
≥ r
Logic and Artificial Intelligence 16/19
![Page 59: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/59.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
Suppose Ann chooses Hand Bob chooses M
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
H
M
Logic and Artificial Intelligence 17/19
![Page 60: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/60.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
A set of information statesand Bob chooses M
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
H
M
Logic and Artificial Intelligence 17/19
![Page 61: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/61.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
A set of information statesand Bob chooses M
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
H
M
Logic and Artificial Intelligence 17/19
![Page 62: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/62.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
A set of information statesand Bob chooses M
Are these choices rational?Yes.
Ann considers it possibleBob is irrational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
H
M
Logic and Artificial Intelligence 17/19
![Page 63: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/63.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
A set of information statesand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 64: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/64.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
116
18
116
12
A common prior and Bobchooses M
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 65: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/65.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
116
18
116
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 66: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/66.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
116
18
116
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 67: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/67.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
116
18
116
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 68: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/68.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
12
18
116
14
18
116
14
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 69: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/69.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 U
D 0,0 1,1 H
18
12
18
116
14
18
116
14
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
3 ·PA(L) + 0 ·PA(R) ≥ 0 ·PA(L) + 1 ·PA(R)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 70: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/70.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
12
18
116
14
18
116
14
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
3 ·PA(L) + 0 ·PA(R) ≥ 0 ·PA(L) + 1 ·PA(R)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 71: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/71.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
12
18
116
14
18
116
14
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
3 · 12 + 0 · PA(R) ≥ 0 · 1
2 + 1 · PA(R)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 72: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/72.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
12
18
116
14
18
116
14
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
3 · 12 + 0 · 1
2 ≥ 0 · 12 + 1 · 1
2
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 73: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/73.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
116
112
212
18
116
112
812
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
0·PB(U)+1·PB(D) ≥ 3·PB(U)+0·PB(D)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 74: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/74.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
116
112
212
18
116
112
812
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Ann (Bob) knows thatBob (Ann) is rational
0 · 212 + 1 · 10
12 ≥ 3 · 212 + 0 · 10
12
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 75: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/75.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
116
18
116
12
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Bob (Ann) knows thatAnn (Bob) is rational
1·PA(L)+0·PA(R) ≥ 0·PA(L)+2·PA(R)
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 76: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/76.jpg)
An Example
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
16
116
18
16
116
12
46
Suppose Ann chooses Uand Bob chooses R
Are these choices rational?Yes.
Bob (Ann) knows thatAnn (Bob) is rational
0 · 16 + 1 · 5
6 ≥ 3 · 16 + 0 · 5
6
U
D
L R
Logic and Artificial Intelligence 17/19
![Page 77: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/77.jpg)
Two Issues
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
116
18
116
12
Zero probability 6= “impossible”
Different “types” of players canmake the same choice
Are Ann and Bob rational? Yes.
Do they know that each other isrational? No.(though PrBob(Irrat(Ann)) = 0)
U
D
L R
Logic and Artificial Intelligence 18/19
![Page 78: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/78.jpg)
Two Issues
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
18
18
0
12
1. Zero probability 6= “impossible”
Different “types” of players canmake the same choice
Are Ann and Bob rational? Yes.
Do they know that each other isrational? No.(though PrBob(Irrat(Ann)) = 0)
U
D
L R
Logic and Artificial Intelligence 18/19
![Page 79: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/79.jpg)
Two Issues
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
18
18
0
12
1. Zero probability 6= “impossible”
2. Different “types” of players canmake the same choice
Are Ann and Bob rational? Yes.
Do they know that each other isrational? No.(though PrBob(Irrat(Ann)) = 0)
U
D
L R
Logic and Artificial Intelligence 18/19
![Page 80: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/80.jpg)
Two Issues
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
18
18
0
12
1. Zero probability 6= “impossible”
2. Different “types” of players canmake the same choice
I Are Ann and Bob rational? Yes.
Do they know that each other isrational? No.(though PrBob(Irrat(Ann)) = 0)
U
D
L R
Logic and Artificial Intelligence 18/19
![Page 81: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/81.jpg)
Two Issues
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
18
18
0
12
1. Zero probability 6= “impossible”
2. Different “types” of players canmake the same choice
I Are Ann and Bob rational? Yes.
Do they know that each other isrational? No.(though PrBob(Irrat(Ann)) = 0)
U
D
L R
Logic and Artificial Intelligence 18/19
![Page 82: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/82.jpg)
Two Issues
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
18
18
0
12
1. Zero probability 6= “impossible”
2. Different “types” of players canmake the same choice
I Are Ann and Bob rational? Yes.
I Do they know that each other isrational? No.(though PrBob(Irrat(Ann)) = 0)
U
D
L R
Logic and Artificial Intelligence 18/19
![Page 83: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/83.jpg)
Two Issues
Bob
Ann
U L R
U 3,3 0,0 H
D 0,0 1,1 H
18
18
18
18
0
12
1. Zero probability 6= “impossible”
2. Different “types” of players canmake the same choice
I Are Ann and Bob rational? Yes.
I Do they know that each other isrational? No.(though PrBob(Irrat(Ann)) = 0)
U
D
L R
Logic and Artificial Intelligence 18/19
![Page 84: Logic and Arti cial Intelligenceepacuit/classes/logicai-cmu/logai-lec5.pdf · Tilburg University ai.stanford.edu/˘epacuit e.j.pacuit@uvt.nl September 14, 2011 Logic and Arti cial](https://reader036.vdocuments.mx/reader036/viewer/2022071101/5fdadc5e04b85c7097713fe2/html5/thumbnails/84.jpg)
Next: Common Knowledge
Logic and Artificial Intelligence 19/19