Professional Documents
Culture Documents
2
These examples are discussed in lecture The Strange World of Human Decision Making,
\lectures\decisions\lectures\foibles\foibles01.pdf.
lectures\talks\lib\pdox-choice01.tex Severe Uncertainty 196/55
Ignorance is not probabilistic:
Monty Halls 3-door problem.
Pascals wager.
Lewis Carrolls 2-bag riddle.
Keynes new material.
Ignorance is a gap between what we
do know and what we need to know
in order to make a good decision.
lectures\talks\lib\pdox-choice01.tex Robust Satiscing: Response to Severe Uncertainty 196/56
1.5 Response to Severe Uncertainty
lectures\talks\lib\pdox-choice01.tex Robust Satiscing: Response to Severe Uncertainty 196/57
What decision strategy for severe uncertainty?
Best-model optimization.
1-reason (lexicographic).
Min-max (worst case).
Robust satiscing.
Opportune windfalling.
_
Prob
_
_
_
Prob(p
T
S
B
) >
1
2
_
_
_
>
1
2
_
_
_
_
_
>
1
2
= B
r
C (11)
Converse not necessary at any step.
lectures\talks\lib\pdox-choice01.tex Does Robust Satiscing Use Probability? 2nd Question 196/125
Second question was, Is:
p
T
S
B
(12)
necessary to motivate robust preferences:
B
r
C (13)
_
Prob
_
_
_
Prob(p
T
S
B
) >
1
2
_
_
_
>
1
2
_
_
_
_
_
>
1
2
= B
r
C (16)
_
Prob
_
_
_
Prob(p
T
S
B
) >
1
2
_
_
_
>
1
2
_
_
_
_
_
>
1
2
= B
r
C (19)
Higher-order probability statements
provide weaker justication.
_
Prob
_
_
_
Prob(p
T
S
B
) >
1
2
_
_
_
>
1
2
_
_
_
_
_
>
1
2
= B
r
C (22)
Higher-order probability statements
provide weaker justication.
None of any nite sequence of prob
beliefs is necessary to justify
r
.
_
Prob
_
_
_
Prob(p
T
S
B
) >
1
2
_
_
_
>
1
2
_
_
_
_
_
>
1
2
= B
r
C (25)
Higher-order probability statements
provide weaker justication.
None of any nite sequence of prob
beliefs is necessary to justify
r
.
At any step, prob belief can be deferred
to next higher-order belief.
_
Prob
_
_
_
Prob(p
T
S
B
) >
1
2
_
_
_
>
1
2
_
_
_
_
_
>
1
2
= B
r
C (28)
Higher-order probability statements
provide weaker justication.
None of any nite sequence of prob
beliefs is necessary to justify
r
.
At any step, prob belief can be deferred
to next higher-order belief.
Answer to second: Nope.
lectures\talks\lib\pdox-choice01.tex Does Robust Satiscing Use Probability? 3rd Question 196/131
1.7.3 Third Question
lectures\talks\lib\pdox-choice01.tex Does Robust Satiscing Use Probability? 3rd Question 196/132
First question:
Does B
r
C need to assume a uniform pdf on the un-
derlying uncertain events?
Second question:
Does B
r
C assume some probability distribution on
the uncertain events?
_
Prob
_
_
_
Prob(p
T
S
B
) >
1
2
_
_
_
>
1
2
_
_
_
_
_
>
1
2
(29)
necessary in order to justify B
r
C?
_
Prob
_
_
_
Prob(p
T
S
B
) >
1
2
_
_
_
>
1
2
_
_
_
_
_
>
1
2
(30)
necessary in order to justify B
r
C?
Answer: plausibly no.
lectures\talks\lib\pdox-choice01.tex Does Robust Satiscing Use Probability? 3rd Question 196/136
Figure 20: John Locke, 16321704.
John Lockes wingless gentleman.
If we will disbelieve everything, because we cannot cer-
tainly know all things; we shall do muchwhat as wisely
as he, who would not use his legs, but sit still and perish,
because he had no wings to y.
9
9
John Locke, 1706, An Essay Concerning Human Understanding, 5th edition. Roger Woolhouse, editor. Penquin Books, 1997,
p.57, I.i.5
lectures\talks\lib\pdox-choice01.tex Does Robust Satiscing Use Probability? 3rd Question 196/137
Figure 21: John Locke, 16321704.
John Lockes wingless gentleman.
If we will disbelieve everything, because we cannot cer-
tainly know all things; we shall do muchwhat as wisely
as he, who would not use his legs, but sit still and perish,
because he had no wings to y.
10
Our situation:
If we disbelieve all props in eq.(30),
rejecting
r
is muchwhat as wise as
Lockes wingless gentleman.
Rejecting
r
is epistemic paralysis.
10
John Locke, 1706, An Essay Concerning Human Understanding, 5th edition. Roger Woolhouse, editor. Penquin Books, 1997,
p.57, I.i.5
lectures\talks\lib\pdox-choice01.tex Does Robust Satiscing Use Probability? 3rd Question 196/138
Causes of epistemic paralysis:
Pdfs not known: info-gaps.
Pdfs dont exist: Wald.
Indeterminism: Shackle-Popper.
u}
Nesting: h < h
= U(h,
u) U(h
,
u)
\lib\proxy02.tex Robustness and Probability: Proxy Theorems 196/179/151
Robustness: max tolerable uncertainty.
h(r, L
c
) = max
_
_
h :
_
_
_
max
uU(h,
u)
L(r, u)
_
_
_
L
c
_
_
(31)
h(r, L
c
) = max
_
_
h :
_
_
_
max
uU(h,
u)
L(r, u)
_
_
_
L
c
_
_
(32)
Robust-satiscing preferences:
r
r
r
if
h(r, L
c
) >
h(r
, L
c
) (33)
h(r, L
c
) = max
_
_
h :
_
_
_
max
uU(h,
u)
L(r, u)
_
_
_
L
c
_
_
(34)
Robust-satiscing preferences:
r
r
r
if
h(r, L
c
) >
h(r
, L
c
) (35)
Probability of survival:
P
s
(r) = Prob[L(r, u) L
c
] =
L(r,u)L
c
p(u) du (36)
h(r, L
c
) = max
_
_
h :
_
_
_
max
uU(h,
u)
L(r, u)
_
_
_
L
c
_
_
(37)
Robust-satiscing preferences:
r
r
r
if
h(r, L
c
) >
h(r
, L
c
) (38)
Probability of survival:
P
s
(r) = Prob[L(r, u) L
c
] =
L(r,u)L
c
p(u) du (39)
Probabilistic preferences:
r
p
r
if P
s
(r) > P
s
(r
) (40)
\lib\proxy02.tex Robustness and Probability: Proxy Theorems 196/179/155
Robustness: max tolerable uncertainty.
h(r, L
c
) = max
_
_
h :
_
_
_
max
uU(h,
u)
L(r, u)
_
_
_
L
c
_
_
(41)
Robust-satiscing preferences:
r
r
r
if
h(r, L
c
) >
h(r
, L
c
) (42)
Probability of survival:
P
s
(r) = Prob[L(r, u) L
c
] =
L(r,u)L
c
p(u) du (43)
Probabilistic preferences:
r
p
r
if P
s
(r) > P
s
(r
) (44)
Do
r
and
p
agree?
h(r, L
c
) proxies for P
s
(r)???
h(r, L
c
) >
h(r
, L
c
) implies P
s
(r) P
s
(r
)???
\lib\proxy02.tex Robustness and Probability: Proxy Theorems 196/179/156
Why
r
and
p
Are
Not Necessarily Equivalent
Two actions, r
1
and r
2
, with robustnesses:
h(r
1
, L
c
) <
h(r
2
, L
c
) (45)
Denote
h
i
=
h(r
i
, L
c
), U
i
= U(
h
i
,
q).
U
i
are nested:
U
1
U
2
because
h
1
<
h
2
(46)
Q
1
Q
2
U
i
reect agents beliefs.
\lib\proxy02.tex Robustness and Probability: Proxy Theorems 196/179/157
Survival set:
(r, L
c
) = {u : L(r, u) L
c
} (47)
Prob of survival:
P
s
(r) = Prob[(r, L
c
)] (48)
Survival sets:
Need not be nested.
Do not reect agents beliefs.
Q
1
Q
2
2
Proxy theorem need not hold.
\lib\proxy02.tex Robustness and Probability: Proxy Theorems 196/179/158
Systems with Proxy Theorems
(Each theorem with its own ne print)
\lib\proxy02.tex Robustness and Probability: Proxy Theorems 196/179/159
Uncertain Bayesian mixing of 2 models.
Decision r.
Outcome depends on which of 2 models,
A or B, is true.
u is uncertain probability that A is true.
h(r, L
c
) is robustness for satiscing at L
c
.
P
s
(r) is probability for satiscing at L
c
.
h(r, L
c
) is robustness for satiscing at L
c
.
P
s
(r) is probability for satiscing at L
c
.
h(r, L
c
) is a proxy for P
s
(r):
Any change in r that increases
h cannot decrease P
s
.
\lib\proxy02.tex Robustness and Probability: Proxy Theorems 196/179/163
Risky and risk-free assets.
r is fraction of budget invested in risky assets.
u is uncertain return from risky assets.
h(r, L
c
) is robustness for satiscing at L
c
.
P
s
(r) is probability for satiscing at L
c
.
h(r, L
c
) is robustness for satiscing at L
c
.
P
s
(r) is probability for satiscing at L
c
.
h(r, L
c
) is a proxy for P
s
(r):
Any change in r that increases
h cannot decrease P
s
.
\lib\proxy02.tex Robustness and Probability: Proxy Theorems 196/179/167
Foraging in an uncertain environment:
Is the grass really greener?
r is duration on current patch.
u is relative productivity of other patch.
h(r, L
c
) is robustness for satiscing at L
c
.
P
s
(r) is probability for satiscing at L
c
.
h(r, L
c
) is robustness for satiscing at L
c
.
P
s
(r) is probability for satiscing at L
c
.
h(r, L
c
) is a proxy for P
s
(r):
Any change in r that increases
h cannot decrease P
s
.
\lib\proxy02.tex Robustness and Probability: Proxy Theorems 196/179/171
49 W
51 B
R
100
W, B
H
Figure 24: Ellsbergs Urns.
Ellsbergs paradox.
2 lotteries: 1 risky, 1 ambiguous.
2 experiments: win on black or win on white.
h(r, L
c
) is robustness for satiscing at L
c
.
P
s
(r) is probability for satiscing at L
c
.
h(r, L
c
) is robustness for satiscing at L
c
.
P
s
(r) is probability for satiscing at L
c
.
h(r, L
c
) is a proxy for P
s
(r):
Any change in r that increases
h cannot decrease P
s
.
\lib\proxy02.tex Robustness and Probability: Proxy Theorems 196/179/179
Recap:
Many systems have proxy property.
Many systems dont have proxy prop.
How prevalent is the proxy property?
Human: economic competition.
Biological: foraging.
Physical: quantum uncertainty.
lectures\talks\lib\pdox-choice01.tex Robust Satiscing: Normative or Prescriptive? 196/180
4 Robust Satiscing: Normative or Prescriptive?
Main Source:
Barry Schwartz, Yakov Ben-Haim, and Cli Dacso, 2011,
What Makes a Good Decision? Robust Satiscing as a
Normative Standard of Rational Behaviour, The Jour-
nal for the Theory of Social Behaviour, 41(2): 209-227.
Pre-print to be found on:
http://info-gap.com/content.php?id=23
lectures\talks\lib\pdox-choice01.tex Robust Satiscing: Normative or Prescriptive? 196/181
3 types of decision theories:
Descriptive.
Normative.
Prescriptive.
lectures\talks\lib\pdox-choice01.tex Robust Satiscing: Normative or Prescriptive? 196/182
3 types of decision theories:
Descriptive:
Describe how DMs decide.
Normative.
Prescriptive.
lectures\talks\lib\pdox-choice01.tex Robust Satiscing: Normative or Prescriptive? 196/183
3 types of decision theories:
Descriptive:
Describe how DMs decide.
Normative:
Establish norm, gold standard for decision making.
Prescriptive.
lectures\talks\lib\pdox-choice01.tex Robust Satiscing: Normative or Prescriptive? 196/184
3 types of decision theories:
Descriptive:
Describe how DMs decide.
Normative:
Establish norm, gold standard for decision making.
Prescriptive:
Specify implementable strategies
given human and epistemic limitations.
lectures\talks\lib\pdox-choice01.tex Robust Satiscing: Normative or Prescriptive? 196/185
Simon introduced satiscing (1950s)
as a prescriptive compromise.