Professional Documents
Culture Documents
Alvin Seminar Handout
Alvin Seminar Handout
such that
b
accuracy-dominates B?
Branden Fitelson Individual Coherence & Group Coherence 12
The Traditional Stories A New Approach to A, C & E Results & Open Questions Refs Proofs & Illustrations
For simplicity, we restrict our discussion to nite, logically
omniscient, opinionated agents who make denite judgments
regarding each proposition in some Boolean algebra 1.
I.e., for each p 1, S either believes p [B(p)] or disbelieves
p [D(p)], and not both (and no suspension of judgment).
First step: say what it means for beliefs/disbeliefs to be
(in)accurate. This is easy and uncontroversial. Of course:
B(p) is (in)accurate at w i p is true (false) at w.
D(p) is (in)accurate at w i p is false (true) at w.
Second step: say what it means for one judgment set B
(over an algebra 1) to be accuracy-dominated by another B
.
The analogous way to think about accuracy-dominance for
sets of qualitative judgments involves comparing numbers
of inaccurate judgments in each of the judgment sets, i.e.,
One set of judgments B
(strictly) accuracy-dominates
another B (over a full algebra 1) i B
.
What is the precise content of (CB
)? Is it non-trivial? And,
how does it relate to our evidential norm (EB)?
It turns out that (CB
)!]
On most accounts of propositional justication [? 20, 7], an
agent S with evidence E is justied in B(p) [D(p)] only if
there is some Pr() such that Pr(p E) >
1
2
_
Pr(p E) <
1
2
_
.
)
}
Local/Narrow Evidential Norm: (EB)
The full picture of logical relationships between norms:
(TB)
(CB)
(CB
)
(EB)
L
B
X & Y B D
11
32
X & Y B D
7
32
X & Y B D
13
32
X & Y D D
1
32
Y B B
18
32
X Y B B
24
32
X D D
12
32
X B B
20
32
(X Y) D D
8
32
Y D D
14
32
X Y B B
31
32
X Y B B
19
32
X Y B B
25
32
X Y B B
21
32
X X B B 1
X & X D D 0
Here is a concrete example, which illustrates how
the coherence norm (CB
) works.
Consider a very simple agent S with a language
containing two atomic sentences X and Y. This
gives rise to an algebra 1 with 16 propositions.
Let Ss judgment set B be as depicted in the
table. B violates (CB
.
Note: B
satises (CB
is to note
that it has a probabilistic representation [L
B
]
with a strict
1
2
-threshold (of course, B does not).
Branden Fitelson Individual Coherence & Group Coherence 19
The Traditional Stories A New Approach to A, C & E Results & Open Questions Refs Proofs & Illustrations
On this slide, well prove Theorem (I). First, some notation.
Let (B, w) = the # of inaccurate judgments in B at w.
(I) Theorem. S violates (CB
disagree. Since (B
agree with
B on all judgments outside of and disagree with B on all
judgments in . Then, B
must dominate B.
On the next slide, we prove Theorem (II). Let 1(B, Pr) be the
expected score of B, relative to Pr(). 1(B, Pr) is dened in
terms of the proposition-wise score of B at w, (B, p, w).
(B, p, w) =
_
_
_
1 if Bs judgment re p is inaccurate at w,
0 if Bs judgment re p is accurate at w.
Branden Fitelson Individual Coherence & Group Coherence 20
The Traditional Stories A New Approach to A, C & E Results & Open Questions Refs Proofs & Illustrations
Note: the score of B at w is just (B, w) =
p
(B, p, w).
Thus, the expected score 1(B, Pr) is the expectation of the
sum
p
(B, p, w). Because expectation is linear, 1(B, Pr) is
therefore the sum of the expectations of the (B, p, w). So:
1(B, Pr) =
p
E((B, p, w), Pr) =
w
Pr(w)(B, p, w)
=
_
p
_
_
_
w=p
Pr(w)(B, p, w) +
_
w=p
Pr(w)(B, p, w)
_
_
=
p
_
_
_
_
_
_
_
_
_
w=p
Pr(w) 1 +
_
w=p
Pr(w) 0 if D(p) B,
_
w=p
Pr(w) 0 +
_
w=p
Pr(w) 1 if B(p) B.
_
_
_
_
_
_
=
p
_
_
_
_
_
Pr(p) if D(p) B,
Pr(p) if B(p) B.
_
_
1(B, Pr) is minimized if, p, Pr(p) >
1
2
B(p) B and
Pr(p) <
1
2
D(p) B. This implies Theorem (II).
Branden Fitelson Individual Coherence & Group Coherence 21
The Traditional Stories A New Approach to A, C & E Results & Open Questions Refs Proofs & Illustrations
Consider a language w/16 state descriptions s
1
, ..., s
16
. Let:
p = s
1
s
2
s
3
s
4
q = s
1
s
5
s
6
s
7
r = s
2
s
5
s
8
s
9
s = s
3
s
6
s
8
s
10
t = s
4
s
7
s
9
s
10
= p, q, r, s, t
Here are four key facts about the set .
(i) Any two sentences in are logically consistent.
because any pair shares a state description.
(ii) Any three sentences in are logically inconsistent.
because every state description occurs exactly twice.
(iii) Any four sentences in are coherent (if jointly believed).
Theorem (I) + the fact that it is not guaranteed that such a
judgment set will contain a subset such that, at every
world, a majority of s members are inaccurate.
(iv) is not coherent (if jointly believed).
Theorem (I) + the fact that, at every world, a majority of
members of any such judgment set must be inaccurate.
Branden Fitelson Individual Coherence & Group Coherence 22
The Traditional Stories A New Approach to A, C & E Results & Open Questions Refs Proofs & Illustrations
p q r s t
J
1
B B B B D
J
2
B B B D B
J
3
B B D B B
J
4
B D B B B
J
5
D B B B B
Majority B B B B B
Each judge can be coherent because judgment sets with 4/5
beliefs (and 1/5 disbeliefs) over can be non-dominated.
This is because there will be worlds in which a majority of
such judgments are accurate. (For example: in worlds that
make state description s
1
true, p, q and t are all true.)
However the (80%!) majority believes all members of . And,
any judgment set containing these judgments must be
dominated. So, majority rule doesnt preserve coherence.
On the next slide, well sketch a proof of our positive
Theorem (III). The key will be to use Theorem (II).
Branden Fitelson Individual Coherence & Group Coherence 23
The Traditional Stories A New Approach to A, C & E Results & Open Questions Refs Proofs & Illustrations
Recall, Theorem (II) ensures that that if a judgement set B is
representable by some probability function via a strict
1
2
-threshold, then that judgment set B must be coherent.
For majority acceptance on individually consistent and
complete inputs this is clearly true. The probability
function in question is just the pattern of individual votes:
For all p, Pr
(p) =
# of judges for p
# of total judges
.
To verify this, note that Pr
).
Branden Fitelson Individual Coherence & Group Coherence 24