Information Technology Reference
In-Depth Information
The previous definition of guilt-dependent utility is related with the definition of regret-
dependent utility proposed in regret theory [18,19,15]. Specifically, similarly to Loomes
& Sugden's regret theory, we assume that computation of emotion-dependent utility
consists in adding to player
i
's personal utility the value
δ
i
(
Emotion
(
i
,
s
)) which mea-
sures the intensity of player
i
's current emotion.
3
There are several possible instantia-
tions of the function
δ
i
(
Guilt
(
i
,
s
)). For example, it might be defined as follows:
δ
i
(
Guilt
(
i
,
s
))
=
c
i
×
Guilt
(
i
,
s
)
(3)
where
c
i
∈ R
+
= {
x
∈ R|
x
≥
0
}
is a constant measuring player
i
's degree of guilt
aversion.
2.3
Grounding Moral Values on Personal Utilities
In the preceding definition of normal form game with moral values a player
i
's util-
ity function
U
i
and ideality function
I
i
are taken as independent. Harsanyi's theory of
morality provides support for an utilitarian interpretation of moral motivation which
allows us to reduce a player
i
's ideality function
I
i
to the utility functions of all players
[17,16]. Specifically, Harsanyi argues that an agent's moral motivation coincides with
the goal of maximizing the collective utility represented by the weighted sum of the
individual utilities.
Definition 5 (Normal form game with moral values based on Harsanyi's view).
A normal form game with moral values
Γ
+
=
,{
S
i
}
i
∈
N
,{
U
i
}
i
∈
N
,{
I
i
}
i
∈
N
)
is based on
(
N
∈
Harsanyi's view of morality if and only if for all i
N:
I
i
(
s
)
=
k
i
,
j
×
U
j
(
s
)
(4)
j
∈
N
for some k
i
,
1
,...,
k
i
,
n
∈
[0
,
1]
.
The parameter
k
i
,
j
in the previous equation can be conceived as the agent
i
's
degree of
empathy
towards agent
j
. This means that the higher the degree of empathy of agent
i
towards agent
j
, the higher the influence of agent
j
's personal utility on the degree of
ideality of a given alternative for agent
i
. In certain situations, it is reasonable to suppose
that an agent has a maximal degree of empathy towards all agents,
i.e
.,
k
i
,
j
=
1forall
i
,
j
∈
N
. Under this assumption, the previous equation can be simplified as follows:
=
I
i
(
s
)
U
j
(
s
)
(5)
j
∈
N
An alternative to Harsanyi's utilitarian view of morality is Rawls' view [22]. In response
to Harsanyi, Rawls proposed the
maximin
criterion of making the least happy agent as
happy as possible: for all alternatives
s
and
s
, if the level of well-being in the worst-
off position is strictly higher in
s
than in
s
,then
s
is better than
s
. According to this
3
On the ground of empirical evidence, Loomes & Sugden also suppose that the function
δ
i
should be convex. To keep our model simpler, we do not make this assumption here.