Social Welfare for Automatic Innovation - Multiagent System Technologies - page 32

(a) Agent's Opinion
determinant factors.

(b) Emotional response computed for an individual. Each
of the 11 emotions are computed by a different function.

Fig. 1. Bases for the model

3.2 A KAI Based Emotional Social Agent Model

An agent is situated in an environment, but this environment is not only the
physical world ruled by mechanical changes; it also comprehends the social world.
The environment is modeled as < S,A,δ,r > ,being S the set of states the

environment has; A the analogous set of actions; δ : S, A → S the transition
function, which changes the environment state to another when an action is
performed; and

E

r S, A → R

the observable reward function, that yields a
reward when an action is performed in a state.
The theoretical interest of having emotions is double: On the first hand agents
that are modeled with emotions resemble better human behavior -by not showing
a pure rational behavior-, thus the model can be more accurate when simulating
humans. On the second hand, there is psychological and neurological evidence
[12,13,14] that emotions may be necessary for a person to behave intelligently,
which arises questions about their necessity in agents.
The transformation between emotional stimuli and emotional responses will
be carried out by a set of 11 functions which output the intensity for each one
of the 11 dimensions. Again, the functions' domains can be composed only by
the set of observable and measurable objects and events of the environment and
perhaps some actions of the agents located there. The emotional response is very
important when emitting each agent's opinion. The Social Opinion will be the
aggregation of every agent's opinion taking into account the trust for each of
them, and the reputation which each agent has in this society.
An agent is modeled as < P, S,R Σ ,κ> .

:

R n → R
which map the emotional stimuli to the emotional response, following the OCC
model;

P is a set of 11 functions P i :

S∈ R n , is the set of emotional stimuli (observable data); R Σ :

11

→ R
is the social opinion function, used to expose a public, observable opinion about
the agent's emotions; and

R

R Σ n → R is the KAI index function calculator

that computes how innovative this agent is being given the emotional response.

κ :

4 Social-Welfare RL (SoWelL)

The agent is situated in an open environment, thus it needs to learn the functions
which, at a given time, govern this universe. A good method to do so is RL.