Panksepp posited that the foundation of emotional
feelings is contained in the evolved emotional action
system of mammalian brains (2005). In the LIDA
model (see below), feelings affect the sensory motor
automatisms (SMA) of autonomous agents. When one
is sad, it affects the actions chosen and how the actions
are taken. Similar role of feelings can be observed with
other feelings as well -- how one holds a cup when one
is angry vs. when one is happy. Feelings affect our
facial expressions and our spoken words as well. The
feeling manifests in one's body affecting the SMAs.
Further, we hypothesize that feelings modulate learning
with an inverse U-curve.
Emotions, such as fear, anger, joy, sadness, shame,
embarrassment, resentment, regret, guilt, etc., are taken
to be feelings with cognitive content (Johnston 1999).
One cannot simply feel shame, but shame at having
done something. The something done constitutes the
cognitive content. Similarly, one must be angry at
someone, that someone being the cognitive content.
Feelings, including emotions, are nature’s means of
implementing motivations for actions in humans and
other animals. They have evolved so as to adapt us to
regularities in our environments.
These general preferences derived evolutionarily from
regularities can be viewed as values. Thus feelings
become implementations of values in biological agents,
providing a common currency for quick and flexible
action selection.
Artificial feelings and emotions are beginning to play
an increasingly important role as mechanisms for
primary motivations in software agents and robots, as
well as facilitators of learning in these systems
(Marsella & Gratch 2002, Langley et al. 2003; Franklin
and McCauley 2004, Avila-Garcia and Canamero 2005,
Ahn and Picard 2006). Here we present a case study of
such feelings and emotions playing both roles in an
intelligent software agent capable of performing a
practical, real world task. In this agent they are actively
involved in every instance of action selection, and at
least potentially involved in each learning event. The
pervasive, central role that feelings and emotions play
in the control structure of this software agent mimics
the roles they play in human cognition, and gives rise
to clarifying hypotheses about human decision making
and several forms of human learning.
2. The LIDA Model
LIDA provides a conceptual (and potentially a
computational) model of cognition (Franklin 2000,
2001b) implemented as a software agent (Franklin &
Graesser 1997) or as an epigenetic robot. The
computational IDA “lives” on a computer system with
connections to the Internet and various databases, and
does personnel work for the U.S. Navy, performing all
the specific personnel tasks of a human (Franklin
2001a). In particular, IDA negotiates with sailors in
natural language, deliberates, and makes voluntary
action selections (Frankllin 2000a) in the process of
finding new jobs for sailors at the end of their current
tour of duty. IDA completely automates the work of
certain Navy personnel agents (detailers) (McCauley
and Franklin 2002).
The LIDA (Learning IDA) model implements and
fleshes out Global Workspace theory (Baars 1988,
2002), which suggests that conscious events involve
widespread distribution of focal information needed to
recruit neuronal resources for problem solving. The
LIDA implementation of GW theory yields a fine-
grained functional account of the steps involved in
perception, several kinds of memory, consciousness,
context setting, and action selection. Cognitive
processing in LIDA consists of continually repeated
traversals through the steps of a cognitive cycle (Baars
& Franklin 2003, Franklin et al. 2005), as described
below.
The LIDA architecture (Figure 1) includes modules for
perception (Zhang, et al. 1998), various types of
memory (Anwar and Franklin. 2003, Franklin et al.
2005, D'Mello, Ramamurthy, and Franklin. 2006 in
press), “consciousness” (Bogner, Ramamurthy and
Franklin. 2000), action selection (Negatu and Franklin.
2002), constraint satisfaction (Kelemen, Liang, and
Franklin. 2002), deliberation (Franklin 2000a), and
volition (Franklin 2000a). The mechanisms of these
modules are derived from several different “new AI”
sources (Hofstadter and Mitchell. 1994, Jackson 1987,
Kanerva 1988, Drescher 1991, Maes 1989). Figure 1
provides the current implementation status of the LIDA
model.
The computational IDA senses strings of characters
from email messages and databases, and negotiates
with sailors via email. The computational IDA is a
running software agent that has been tested and
demonstrated to the satisfaction of the U.S. Navy.
Detailers observing the testing commented, “IDA
thinks like I do.”
In addition to the computational model, we will also
speak of the conceptual LIDA (Learning IDA) model,
which includes additional capabilities that have been
designed but not implemented, including mechanisms
for feelings and emotions, and for various forms of
learning.
The LIDA conceptual model contains several different
memory systems. Perceptual memory (often called
perceptual organization) enables identification,
recognition and categorization, including of feelings.