3. THE PROBLEM
Problem: utility of a choice depends on
the salience of alternative choices
i.e., choice set-dependent preferences
e.g., compromise effect & decoy effect
good!
ehh…
ehh… awesome!
awesome!
(2.8 utils)
(2.8 2.2 utils)
(2.8 2.2 utils) (3.6 utils)
(3.6 utils)
Changing utility = don’t represent evolutionary benefits of
consumption (if it did, it would be consistent)
4. THE PROBLEM
Problem: utility of a choice depends on
the salience of alternative choices
i.e., choice set-dependent preferences
e.g., compromise effect & decoy effect
good! awesome!
5. THE PROBLEM
It makes sense to attach utility to outcomes if we
know the perfect relationship between the two.
Sex Offspring
If your goal is to maximize offspring , and you have a
perfect prior statistical understanding of the
world, then you can attach a utility to maximizing
offspring.
typical conception of utility
6. THE MODEL
But, if you don’t know these perfect correlations….
(because it’s impossible to do so) e.g., an opportunity
is faced with an OPTION to eat
You can’t attach utilities to outcomes
ACTOR
You have to attach utilities to actions to (not eating)
(eating) accept reject learn
more about the world 1- p
p q 1- q
success failure success failure
Especially the xpart of the world where you don’t
x
know the consequences= of actions
x success utility
Success: survive long enough to reproduce
7. THE MODEL
Overview If outcomes are perfectly understood in their
effect on success, then actions have no utility.
e.g., an opportunity
1. The probabilities of OPTION to eat
p and q are not
ACTOR
know (e.g., the But, you can’t perfectly process signals about
danger of not eating) the probability of success
2. Actors face noisy
(eating)
You accept reject (not eating)
can’t attach utilities to outcomes, because
signals of p and q you can’t perfectly process signals
p 1- p q 1- q
3. Signals are Instead, you have to attach utilities to accepting
success failure success failure
transformed using a or rejecting an option at the marginal signal
x x
rule (Φ) to estimate
the probability of x = success utility
We incorporate irrelevant factors into our utility
success because they are correlated with quality of
Success: survive information to reproduce
long enough
8. THE MODEL
Overview If outcomes are perfectly understood in their
effect on success, then actions have no utility.
e.g., an opportunity
1. The probabilitiesfaced with an
is of OPTION OPTION
e.g., an opportunity
to eat
p and q are not to eat
ACTOR
know (e.g., the But, you can’t perfectly process signals about
danger of not eating) the probability of success
2. Actors face noisy
(eating) accept (notreject
acceptYou can’t attach utilities to outcomes, because
(not eating)
(eating) reject eating)
signals of p and q you can’t perfectly process signals
p q
p 1- p q1- p 1- q
1- q
3. Signals are Instead, you have to attach utilities to accepting
success failure success failure
success failure success failure
transformed using a or rejecting an option at the marginal signal
x x
rule (Φ) to estimatex+y y x 0
the probability of x = success utility
Wexincorporate irrelevant factors into our utility
= success utility
success y = accepting utility
because they are correlated with quality of
Success: survive information to reproduce
long enough
Success: survive long enough to reproduce
9. THE MODEL
Overview have an accurate statisticalare perfectly understood world, and
We (a) don’t If outcomes representation of the in their
The we can’t accurately process the signals we get about the world eat
(b) probabilities of effect on success, then actions have no utility.
e.g., an opportunity
1. is faced with an OPTION OPTION
e.g., an opportunity
to
p and q are not to eat
ACTOR utility does not involve selecting the evolutionary signals of an
Optimal
know (e.g., the But, you can’t perfectly process value about
danger of not eating) the probability of success
action, but the “information-processing problem” that the individual
faces in choosing between actions (especially when info quality is (not eating)
You accept poor)
(notreject
(eating) can’t attach utilities to outcomes, because
2. Actors face noisy
(eating) accept reject eating)
signals of p and q you can’t perfectly process signals
p q
By making mistakes in
p different contexts,q1- p
1- p our
information processing
1- q
1- q
3. Signals are better tailored to the world, failure attach utilities to acceptingus
abilities are because these contexts allow
Instead, you have to
success success failure
success failure success failure
transformed using a outcome of x rejecting an option at the marginal signal
to understand the orchoices in different circumstances
x
rule (Φ) to estimatex+y y x 0
the probability of x = success utility
Wexincorporate irrelevant factors into our utility
= success utility
success y = accepting utility
because they are correlated with quality of
Success: survive information to reproduce
long enough
Success: survive long enough to reproduce