For Applications Of Computational Intelligence In Biology Current Trends

for applications of computational intelligence in biology: current trends and open problems, tomasz g. smolinski, mariofanna m. milanova, abo

For Applications of Computational Intelligence in Biology: Current
Trends and Open Problems, Tomasz G. Smolinski, Mariofanna M. Milanova,
Aboul-Ella Hassanien, editors
To be published by Springer-Verlag in the series "Studies in
Computational Intelligence"
Using broad cognitive models to apply computational intelligence to
animal cognition
===================================================================
Stan Franklin1 and Michael H. Ferkin2
The University of Memphis,
Institute of Intelligent Systems, Fed Ex Institute of Technology1 and
Department of Biology2, Ellington Hall, Memphis, TN 38152 USA
Abstract
--------
The field of animal cognition (comparative cognition, cognitive
ethology), the study of cognitive modules and processes in the domain
of ecologically relevant animal behaviors, has become mainstream in
biology. The field has its own journals, books, organization and
conferences. As do other scientists, cognitive ethologists employ
conceptual models, mathematical models and sometime computational
models. Most of these models, of all three types, are narrow in scope,
modeling only one or a few cognitive processes. This position chapter
advocates, as an additional strategy, studying animal cognition by
means of computational control architectures based on biologically and
psychologically inspired, broad, integrative, hybrid models of
cognition. The LIDA model is one such model. In particular, the LIDA
model fleshes out a theory of animal cognition, and underlies a
proposed ontology for its study. Using the LIDA model, animal
experiments can be replicated in artificial environments by means of
virtual software agents controlled by such architectures. Given
sufficiently capable sensors and effectors, such experiments could be
replicated in real environments using cognitive robots. Here we
explore the possibility of such experiments using a virtual or a
robotic vole to replicate, and to predict, the behavior of live voles,
thus applying computational intelligence to cognitive ethology.
Introduction
------------
The analysis of animal behavior cannot be complete without an
understanding of how behaviors are selected, that is, without the
study of animal cognition (Allen 1997; Kamil 1998; Boysen and Himes
1999; Byrne and Bates 2006). The study of cognitive modules and
processes in the domain of ecologically relevant animal behaviors
(cognitive ethology) has become an exciting research area in biology.
The field has its own journals (e.g., Animal Cognition), books (e.g.,
Bekoff et al., 2002), organization (Comparative Cognition Society) and
conferences (e.g., 2006 Comparative Cognition Society Annual Meeting,
Melbourne, FL).
As do other scientists, cognitive ethologists employ conceptual models
(e.g., Allen 1997), mathematical models (e.g., Alsop 1998; Kruschke
2001) and sometime computational models (e.g., Saksida 1999). Most of
these models, of any of the three types, are narrow in scope, modeling
only one or a few cognitive processes. In contrast to these models,
empirical studies of how animal behaviors are selected should be
guided by comprehensive theories and integrated conceptual models.
While grounded in the underlying neuroscience and consistent with it,
these theories and models must be conceptually at a higher level of
abstraction, dealing with higher-level entities and processes.
This position chapter advocates studying animal cognition by means of
computational control architectures based on biologically and
psychologically-inspired, broad, integrative, hybrid models of
cognition. Using such a model, experiments with animals could be
replicated in artificial environments with virtual software agents
controlled by such architectures. Given sufficiently capable sensors
and effectors, such experiments could be replicated in real
environments using cognitive robots. The LIDA (Learning Intelligent
Distribution Agent) model provides just the kind of broad, integrated,
comprehensive, biologically and psychologically inspired theory that
is needed. In particular, the LIDA model can model animal cognition,
and underlies a proposed ontology for its study (Franklin and Ferkin
2006).
Software agents in robotic simulators as virtual animals
--------------------------------------------------------
Autonomous agents (Franklin and Graesser 1997) are systems embedded
in, and part of, an environment that sense their environment and act
on it, over time, in pursuit of their own agenda. In addition, they
must act so as to potentially influence their future sensing, that is,
they must be structurally coupled to their environment (Maturana 1975;
Maturana and Varela 1980). Biological autonomous agents include
humans, other animals and viruses. As well as computer viruses and
some robots, artificial autonomous agents include software agents,
that is, agents that “live” in computer systems, in databases, or in
networks. The “bots” that autonomously explore the internet indexing
web pages for Google are examples of software agents. Artificial
autonomous agents also include cognitive robots (Clark and Grush 1999;
Asada et al 2001; Franklin 2005b).
Robotic simulators are software tools offering often 3D modeling,
simulation and animation of any physical system. They are particularly
designed as virtual environments for simulations of robots, hence the
name. Examples include ARS MAGNA, RoboWorks, Rossum's Playhouse,
Khepera Simulator, and very many others. Within such an abstract,
virtual world with its own physics, a simulated robot can both sense
and act so that it becomes an autonomous software agent. Such a
simulated robot typically has a simulated body within the robot
simulator.
Modeling such a simulated robot “living” within a robotic simulator
after an animal, say a meadow vole, creates an artificial animal
software agent. The simulator, thought of as the artificial animal’s
environment, can be made to contain objects of various sorts,
including other agents. Such objects can have simulated weight,
rigidity and other realistic physical properties. Such other agents
can be made to behave in a relatively realistic manner, as for example
do the various agents that occur in video games. An artificial animal
(AA) can sense this environment via artificial sights, sounds, odors,
touches, tastes, etc., corresponding to the senses available to the
animal (robot). AA’s artificial effectors can manipulate artificial
objects, including itself, in rather realistic ways as compared to a
real robot, and in a more or less realistic manner as compared to an
animal.
Behavioral experiments with animals typically may involve some sort of
structure, such as a runway with two chambers at its end, or a maze.
This structure can be simulated within the artificial environment, the
robotic simulator. AA’s “body” can then be placed appropriately within
the simulated structure, an artificial run of the experiment carried
out, and data gathered as to how AA responded to the experimental
situation in terms of location, action, timing, etc. Repeated
artificial runs will allow the virtual replication of the experiment
and its dataset. Or, the virtual experiment can be run first. Its data
would then predict the results of carrying out the experiment in vivo.
As any autonomous agent must, AA has a control structure that
interprets its sensory input, selects an action according to its own
agenda, and guides its actions on its environment. The computational
architecture of any such AA control structure gives rise to a
conceptual cognitive model, that is, a theory of how AA and the animal
it simulates interprets its sensory data, and chooses and guides its
actions. Conversely, any cognitive model can be implemented in a
computational architecture that can be used to control AA. Thus, in
principle, any scientific hypothesis that arises from the cognitive
model can be tested using both a virtual experiment and an in vivo
experiment.
Using such an experimental paradigm insures that the conceptual
cognitive model that gives rise to the computational architecture of
AA’s control system will be broad and comprehensive. It must involve
perception, which makes sense of sensory input. It must contain
motivational elements and procedural memory with which to make action
selections. Finally, it must include sensory-motor automatisms with
which to execute actions. Such comprehensive cognitive models allow
for the testing of a broader range of hypotheses than do more
constrained models. They also enable more complete, and therefore more
satisfactory, explanations of the cognitive processes responsible for
the observed behavior. We contend that the adoption of such an
experimental paradigm will result in significant advances in the way
biological theory guides experimentation in animal behavior.
But there’s more to the story. Computational architectures derived
from integrated, comprehensive cognitive models are sure to be rife
with internal parameters whose values must be tuned (discovered)
before the system can perform properly as a control structure for AA.
It is well known that a model with sufficiently many free parameters
can be tuned so as to reproduce essentially any specific dataset. What
is wanted is a tuned set of internal parameters whose values remain
constant while a number of disparate datasets are reproduced. Such a
tuned parameter set offers reassurance as to the accuracy and
usefulness of the model. An inability to find such a tuned parameter
set should warn its designers that something is amiss with the model,
and that it needs revision. The particular parameters that resist such
tuning point researchers to modules and process within the model that
are likely to require revision.
But how are such parameters in a computational architecture to be
tuned. The problem is essentially a search problem. Given the dataset
from one previous in vivo experiment that the model should predict and
explain, one searches for a set of parameter values that, when
implemented, will allow AA to replicate this existing dataset. If
found, this search procedure is iterated on the dataset of a second
previously performed in vivo experiment resulting, hopefully, in a
tuned parameter set that will allow the replication of both datasets.
Further iteration of this procedure should, if the model is correct,
yield a stable set of values for the internal parameters of the
computational architecture that should work for replicating a number
of different existing in vivo experiments. Thus, the replication of
existing data sets from previously performed experiments will allow
the tuning of internal parameters in the theoretical model. Parameters
that resist such tuning over several different data sets indicate
flaws in the model that must be repaired. This parameter tuning
provides something like a metric for assessing the quality of a
cognitive model as a basis for understanding the cognitive processes
responsible for the behavior of AA.
In summary, a tuned version of the computational model will allow AA
to successfully replicate essentially any simulatable experiment with
the animal in question. Successfully accomplishing this goal will
provide substantial evidence of the accuracy and usefulness of the
conceptual cognitive model. Cognitive hypotheses from the model can
then be tested by in vivo experiments with real animals to see if
their data is predicted by running AA in the same experimental
situations. If so, we will have shown the ability of the theoretical
model to predict as well as to explain.
The authors propose that their LIDA cognitive model, to be described
next, is an appropriate example of a broad, integrated, comprehensive
model of the kind we are advocating. We have proposed this model
previously as the source of a useful ontology for the study of animal
cognition (Franklin and Ferkin 2006).
The LIDA cognitive model and its architecture
---------------------------------------------
The LIDA model is a conceptual (and partially computational) model
covering large portions of human and animal cognition. Based primarily
on global workspace theory (Baars 1988), the model implements and
fleshes out a number of psychological and neuropsychological theories
including situated cognition (Varela et al 1991), perceptual symbol
systems (Barsalou 1999), working memory (Baddeley and Hitch 1974),
memory by affordances (Glenberg 1997), long-term working memory
(Ericsson and Kintsch 1995), and Sloman’s (1999) cognitive
architecture. Viewed abstractly, the LIDA model offers a coherent
ontology for animal cognition (Franklin and Ferkin 2006), and provides
a framework in the sense of Crick and Koch (2003) that can serve to
guide experimental research. Viewed computationally, the model
suggests computational mechanisms that can underlie and explain neural
circuitry.
The LIDA computational architecture, derived from the LIDA cognitive
model, employs several modules motivated by computational mechanisms
drawn from the “new AI.” These include the Copycat Architecture
(Hofstadter and Mitchell 1995), Sparse Distributed Memory (Kanerva
1988), the Schema Mechanism (Drescher 1991), the Behavior Net (Maes
1989), and the Subsumption Architecture (Brooks 1991).
The LIDA model and its ensuing architecture are grounded in the LIDA
cognitive cycle. Every autonomous agent (Franklin and Graesser 1997),
be it human, animal, or artificial, must frequently sample (sense) its
environment, process (make sense of) this input, and select an
appropriate response (action). Every agent’s “life” can be viewed as
pursuing a continual sequence of these cognitive cycles. Each cycle
constitutes a unit of sensing, attending and acting. A cognitive cycle
can be thought of as a moment of cognition, a cognitive “moment.”
Higher-level cognitive processes are composed of many of these
cognitive cycles, each a cognitive “atom.”
During each cognitive cycle (Baars and Franklin 2003; Franklin et al.
2005) the LIDA agent, be it animal or artificial, first makes sense of
its current situation as best as it can. It then decides what portion
of this situation is most in need of attention. Broadcasting this
portion enables the agent to finally choose an appropriate action and
execute it. Please note, that consciousness in the LIDA model refers
to functional consciousness, which is the functional role of the
mechanism as specified by Baars’ (1988) global workspace theory. The
LIDA model takes no position on the issue of phenomenal consciousness
in animals.

Figure 1. The LIDA Cognitive Cycle
The cycle begins with sensory stimuli from the agent’s environment,
both an external and an internal environment. Low-level feature
detectors in sensory memory begin the process of making sense of the
incoming stimuli. These low-level features are passed to perceptual
memory where higher-level features, objects, categories, relations,
situations, etc. are recognized. These recognized entities, comprising
the percept, are passed to the workspace, where a model of the agent’s
current situation is continually being assembled and updated. The
percept serves as a cue to two forms of episodic memory, transitive
and declarative. The response to the cue consists of local
associations, that is, remembered events from these two memories that
were associated with elements of the cue. In addition to the current
percept, the workspace contains recently previous percepts and the
structures assembled from them that haven’t yet decayed away. The
model of the agent’s current situation is assembled from the percept,
the associations and the remaining previous models. This assembling
process will typically require looking back to perceptual memory and
even to sensory memory, to enable the understanding of relations and
situations. This assembled new model constitutes the agent’s
understanding of its current situation within its world. It has made
sense of the incoming stimuli.
For an agent “living” in a complex, dynamically changing environment,
this current model may well be much too much for the agent to deal
with at once. It needs to decide what portion of the model should be
attended to. Which are the most relevant, important, urgent or
insistent structures within the model? Portions of the model compete
for attention. These competing portions take the form of coalitions of
structures from the model. Such coalitions are formed by attention
codelets, small, special purpose processors, each of which has some
particular type of structure it wants to bring to consciousness. One
such coalition wins the competition. The agent has decided on what to
attend.
But, the purpose of all this processing is to help the agent decide
what to do next. To this end, the winning coalition passes to the
global workspace, the namesake of global workspace theory, from which
its contents are broadcast globally. Though the contents of this
conscious broadcast are available globally, the primary recipient is
procedural memory, which stores templates of possible actions
including their contexts and possible results. It also stores an
activation value for each such template that attempts to measure the
likelihood of an action taken within its context producing the
expected result. Templates whose contexts intersect sufficiently with
the contents of the conscious broadcast instantiate copies of
themselves with their variables specified to the current situation.
These instantiations are passed to the action selection mechanism,
which chooses a single action from these instantiations and those
remaining from previous cycles. The chosen action then goes to
sensory-motor memory, where it picks up the appropriate algorithm
(sensory-motor automatism) by which it is then executed. The action
taken affects the environment, and the cycle is complete.
There are neural correlates for each of the modules and processes
included in the LIDA cognitive cycle. For each such module or process,
there is experimental data supporting these correlations.
The LIDA model hypothesizes that in all animals, including humans,
cognitive processing is via a continuing iteration of such cognitive
cycles. These cycles occur asynchronously, with each cognitive cycle
taking roughly 200 ms in humans and closely related primates. The
cycles cascade, that is, several cycles may have different processes
running simultaneously in parallel. This cascading must, however
respect the serial order of consciousness in order to maintain the
stable, coherent image of the world with which consciousness endows us
(Franklin 2005a; Merker 2005). This cascading, together with the
asynchrony, allows a rate of cycling in humans of five to ten cycles
per second. A cognitive “moment” is quite short! There is considerable
empirical evidence from neuroscience suggestive of such cognitive
cycling in humans and closely related primates (Massimini et al. 2005;
Sigman and Dehaene 2006; Uchida et al. 2006; Willis and Todorov 2006).
None of this evidence is conclusive.
Global workspace theory postulates that learning requires only
attention (Baars 1988). In the LIDA model this implies that learning
must occur with each cognitive cycle. More specifically, learning
occurs with the conscious broadcast from the global workspace during
each cycle. Learning in the LIDA model follows the established
artificial intelligence principle of ‘generate and test (Winston 1992;
Kaelbling 1994). New representations are learned in a profligate
manner during each cognitive cycle (the generation). Those that are
not sufficiently reinforced during subsequent cycles (the test) decay
away. Three modes of learning, perceptual, episodic and procedural,
employing distinct mechanisms (Nadel 1992, Franklin et al. 2005) have
been designed and are in various stages of implementation. A fourth
mode of learning is attentional learning, which has been contemplated
but not designed.
Perceptual learning enables an agent to recognize features, objects,
categories relations, situations, etc. It seems to be ubiquitous in
animals. Episodic learning refers to the memory of events, the what,
the where and the when (Tulving 1983, Baddeley et al.2001). In the
LIDA model such learning is stored in transient episodic memory
(Conway 2002; Franklin et al. 2005) and in the longer-term declarative
memory (Franklin et al. 2005). At least episodic-like memory, that is
episodic memory with no assumption of consciousness, has been
demonstrated in many animal species (Dere et al. 2006) including
meadow voles (Ferkin et al. in press). Procedural learning refers to
the learning of new tasks and the improvement of old tasks. In the
LIDA model such learning is accomplished in procedural memory (D'Mello
et al. 2006b). Such procedural learning is widely observed in animal
species (e.g., Foote et al. 2006).
Every autonomous agent must be equipped with primitive motivators,
sometimes called drives that motivate its selection of actions. In
humans, in animals, and in the LIDA model, these drives are
implemented by feelings (Franklin and Ramamurthy 2006). Such feelings
implicitly give rise to values that serve to motivate action
selection. Feelings also act as modulators to learning.
The LIDA theoretical model traverses several levels of biological
complexity within the overall rubric of animal cognition. At the
highest level it models entire organisms by means of software agents
such as a virtual vole. At one step lower, it models various
higher-level cognitive processes such as deliberation (Franklin 2000),
volition (Franklin 2000), metacognition (Zhang et al. 1998),
automization, and non-routine problem solving (D'Mello et al. 2006b).
Yet another step lower one finds cognitive modules and processes that
operate within a single cognitive cycle, that is within a few hundred
milliseconds. These lower-level processes include perception (Franklin
2005), various forms of memory (Franklin et al. 2005), attention
(Baars and Franklin 2003), learning (D'Mello et al. 2006a), and action
selection (Negatu and Franklin 2002). At yet a lower level, the nodes
and links from LIDA’s perceptual memory, implemented via a slipnet
(Hofstadter and Mitchell 1995), provide the common representational
currency throughout the model a la Barsalou’s (1999) perceptual symbol
system. Taking a dynamical systems point of view, each such node may
be thought of as representing a basin of attraction in the state space
of some underlying cell assembly (Skarda and Freeman 1987). By
spanning these various levels of theoretical complexity, the LIDA
model can be expected to contribute to our understanding of several
levels of the dynamics of living systems.
The natural history of the meadow vole, Microtus pennsylvanicus
---------------------------------------------------------------
Meadow voles are small secretive rodents that inhabit ephemeral
grasslands in the northern and eastern portions of the United States
and Canada. Much is known about their life history. Meadow voles also
display striking seasonal differences in behavior. That is, they do
most of their breeding during the spring and summer, when the
photoperiod or day length is relatively long, about 14 hours of light
per 24-hour period. At this time of year, female meadow voles become
sexually receptive to males, producing odors that are attractive to
males as well as displaying behaviors directed towards males (Ferkin
and Seamon 1987; Ferkin et al. 2001; 2004a, b).
During the breeding season, female meadow voles are also territorial
(Madison 1980). They defend their nests and territories often by
behaving aggressively towards intruders. However, fighting is costly
and not frequent (Ferkin and Seamon 1987). Female meadow voles use
other means to defend their territory. Specifically, females scent
mark along the borders of their territories and near their nests, and
over-mark the scent marks of male and female conspecifics that they
encounter. By scent marking and over-marking, female meadow voles are
able to delineate boundaries of their territories and also announce
their residency in an area.
Male meadow voles are not territorial. Instead, they wander through
large home ranges that encompass the territories of one or more
females. Males often do not display overt behaviors against male
conspecifics. They seldom fight with other males and they do not
target the scent marks of other males and over-mark them (Ferkin and
Seamon 1987; Ferkin et al. 2001, 2004a). However, males scent mark and
over-mark in areas containing the marks of female meadow voles (Ferkin
et al. 2004a, b). This does not mean that male voles do not compete
with one another, they do. However, male-male competition is more
subtle; males compete with one another in two ways. First, males
recall the location and reproductive condition of females that they
encounter during their daily wanderings. That is, they display a
memory for what, when, and where (Ferkin et al. in press). Second,
once they locate females that are willing to mate with them, males
will assess the risk and intensity of sperm competition. They do so,
by determining whether the female has recently encountered other
males. Male meadow voles investigate the area near the female,
attempting to determine if other males have left their scent marks
nearby. If so, the male, when he mates with the female, will increase
his sperm investment 116% relative to his investment if he does not
encounter fresh scent marks of other males nearby (del-Barco-Trillo
and Ferkin 2004, 2006, 2007b).
Although many males may mate with females, the variance in
reproductive success among males is highly skewed, so that only a
relatively small number of males actually sire offspring (Sheridan and
Tamarin 1988; Boonstra et al.1993). What it is that makes these males
more successful is not known. However, studies suggest that the more
successful males 1) produce odors that are more attractive than those
of other males to females (Ferkin and Seamon 1987), 2) have higher
titers of prolactin and gonadal steroids relative to those of other
males, which makes the former more attractive and interesting than the
latter to females (Leonard and Ferkin 2005), 3) display more behaviors
directed at attracting and showing their interest in females (Ferkin
et al. 1996, 2004, 2005); 4) have more copulatory interactions with
females (delBarco-Trillo and Ferkin 2004, 2007a), 5) are better fed
than less successful males (Pierce and Ferkin 2005; Pierce et al.
2005), and 6) are older and more experienced than other male meadow
voles (Ferkin and Leonard in press).
As we mentioned above, meadow voles are seasonal breeders. They
generally do not breed during the late fall and winter, when the day
length is short and the daily photoperiod is less than 10 hours of
light per 24-hour period. During the non-breeding season, females
meadow voles relax their territorial borders, produce odors that are
no longer attractive to males, but are attractive to females, display
behaviors that are directed more at females than at males, and form
communal nests with neighboring females and their last litters.
Aggressive behavior between females is reduced and is replaced with
affiliative and amicable acts (Madison 1980; Ferkin and Seamon 1987).
At this time of year, males generally produce odors that are no longer
attractive to females. Few males direct behaviors towards females as
potential mates. Males appear to be solitary during the winter
(Madison 1980). During the winter, scent marking and over-marking
behavior is no longer directed at opposite-sex conspecifics and with
self-grooming behavior serve a role in maintaining the cohesiveness of
members of that communal nest (Leonard and Ferkin 2005; Ferkin and
Leonard in press).
Meadow voles and cognition – Some case studies
----------------------------------------------
In this section, we summarize the results of some experiments on voles
that imply a strong cognitive component to their behavior. For
example, meadow voles can distinguish between unfamiliar and familiar
conspecifics, littermates and non-littermates, and between sexually
receptive and sexually quiescent opposite-sex conspecifics. Meadow
voles respond preferentially to the odors of littermates relative to
non-littermates by spending more time investigating the odors of the
former as compared to those of the latter (Ferkin 1989; Ferkin et al.
1992). Adult female voles behave amicably towards familiar females but
not towards unfamiliar females, whereas adult male voles behave
agonistically towards familiar males but not unfamiliar males (Ferkin
1988). Male voles over-mark the scent marks of females in heightened
sexual receptivity, during postpartum estrous, as compared to those of
females that in other states of sexual receptivity (Ferkin et al.
2004a, b)
Depending on the social context, the perceptual memory of voles may
last several hours to several days (Ferkin et al. 2005, in press).
Perceptual memory can be fleeting or long term. For instance, a new
person met briefly at a party may not be recognized a few weeks later,
while a friend from childhood who hasn’t been seen for decades may be
recognized in spite of the changes brought by age. Perceptual and
episodic memory (what, when, and where) depend to some extent, and in
different ways, on association. In perceptual memory and object is
associated with its features, a category with its members. Recall from
episodic memory is accomplished in animals (and in at least some
artificial agents) by means of associations with a cue. Improvement of
performance during procedural learning is accomplished in animals by
associating particular actions with desired results. Thus association
plays different roles in the various memory systems and their various
forms of learning, and can be expected to require distinct mechanisms.
First, we asked the question, is it possible for voles to have a sense
of number? To address this question, we determined whether voles
discriminate between two different scent-marking individuals and
identify the individual whose scent marks was on top more often than
the other individual (Ferkin et al. 2005). We tested whether voles
show a preference for the individual whose scent marks was on top most
often. If so, the simplest explanation was that voles can make a
relative size judgment, such as distinguishing an area containing more
of one individual’s over-marks as compared to less of another
individual’s over-marks. We found that voles respond preferentially to
the donor that provided the greater number of over-marks as compared
to the donor that provided the fewer number of over-marks. Thus, we
concluded that voles might display the capacity for relative
numerousness. Interestingly, female voles were better able than male
voles in distinguishing between small differences in the relative
number of over-marks by the two scent donors.
Next, we conducted a series of experiments to determine whether
reproductive condition of female meadow voles affects their scent
marking behavior as well as the scent marking behavior of male
conspecifics (Ferkin et al. 2004b). We did so because, during the
breeding season, the reproductive condition of female mammals changes.
Females may or may not be sexually receptive. In experiment 1, females
in postpartum estrus deposited more scent marks than females that were
neither pregnant nor lactating, reference females or ovariectomized
females (OVX females). In experiment 2, male voles scent marked more
and deposited more over-marks in areas marked by postpartum estrus
females than by reference and OVX females. In experiment 3, postpartum
estrus females deposited more scent marks and over-marks in areas
marked by males than did females in the other reproductive states. The
results of these experiments showed that male and female voles may
vary the number, type, and location of scent marks they deposit in
areas scented by particular conspecifics.
We also tested the hypothesis that male meadow voles posses the
capacity to recall the what, where, and when of a single past event
associated with mate selection in two experiments (Ferkin et al. in
press). Briefly, male voles were allowed to explore an apparatus that
contained two chambers. One chamber contained a day-20 pregnant female
(24 hours prepartum). The other chamber contained a reference female.
Twenty-four hours after the exposure, the males were placed in the
same apparatus, which was empty and clean. At this time, the pregnant
female would have entered postpartum estrus, a period of heightened
sexual receptivity. Males initially chose and spent significantly more
time investigating the chamber that originally housed the pregnant
female (now a postpartum estrus female) than the chamber that
originally housed the reference female. Male voles also explored an
apparatus containing a chamber with a postpartum estrus female and one
chamber containing a reference female. Twenty-four hours later, males
were placed into an empty and clean apparatus. The males did not
display an initial choice and they spent similar amounts of time
investigating the chamber that originally housed the postpartum estrus
female (now a lactating female) and the chamber that originally housed
the reference female. The results of these and additional experiments
suggest that male voles may have the capacity to recall the what,
where, and when of a single past event, which may allow males to
remember the location of females who would currently be in heightened
states of sexual receptivity.
We also examined the effects of winning and losing on over-marking
behavior of mammals, a behavior associated with intrasexual aggression
and competition (Ferkin 2007). We tested the hypothesis that meadow
voles adjust their over-marking behavior according to aggressive
interactions they had experienced with a same-sex conspecific. The
hypothesis was partially supported. That is, female voles that won
their encounter over-marked a greater proportion of their opponent’s
over-marks than did females that either lost their encounter or
females that were evenly matched in their encounter. Females that lost
their encounter and females that were evenly matched over-marked a
similar proportion of their opponent’s over-marks. Male voles,
however, independent of whether they won, lost, or were evenly
matched, over-marked a similar proportion of their opponent’s scent
marks. The present findings suggest over-marking may not play a major
role in male-male competition, but likely plays a large role in
female-female competition among meadow voles.
We also determined to what degree meadow voles display self-cognizance
and use self-referent phenotype matching for self recognition (Ferkin
et al. unpubl. data). We tested animals using
habituation/dishabituation tasks in which they were exposed to their
own current urine scent marks and those of either 1) unfamiliar
same-sex conspecifics, 2) same-sex siblings, 3) their past selves
(post gonadectomy with no steroid-hormone replacement), 4) their past
selves (post gonadectomy with steroid-hormone replacement), and 5)
their past selves (intact gonads). Briefly, we discovered that voles
behaved as if the scent marks of their past and present selves were
the same if the reproductive condition of the voles was not changed
and from the same donor. If, however, the voles were gonadectomized
and their reproductive condition changed, they behaved as if the scent
marks of their past and present selves were the different donors.
Finally, we examined sperm competition in male meadow voles
(delBarco-Trillo and Ferkin 2004, 2006, 2007b), Sperm competition
occurs when a female copulates with two or more males and the sperm of
those males compete within the female’s reproductive tract to
fertilize her eggs. The frequent occurrence of sperm competition has
forced males of many species to develop different strategies to
overcome the sperm of competing males. A prevalent strategy is for
males to increase their sperm investment (total number of sperm
allocated by a male to a particular female) after detecting a risk of
sperm competition. It has been shown that the proportion of sperm that
one male contributes to the sperm pool of a female is correlated with
the proportion of offspring sired by that male. Therefore, by
increasing his sperm investment a male may bias a potential sperm
competition in his favor.
We showed that male meadow voles increase their sperm investment when
they mate in the presence of another male’s odors. Such an increase in
sperm investment does not occur by augmenting the frequency of
ejaculations, but by increasing the amount of sperm in a similar
number of ejaculations. We also found that sperm investment of males
exposed to the scent marks of five male conspecifics was intermediate
between that of males exposed to the scent marks of one male and that
of males exposed to no scent marks of conspecific males
(delBarco-Trillo and Ferkin 2006). We have recently discovered that
males do not increase their sperm investment if the donors of the
scent marks are males that are in poorer condition than the male
subject, but do so if the male donors are in similar or better
condition than the subject male (Vaughn et al. unpubl. data). Thus,
males can distinguish between different male donors and adjust their
sperm investment accordingly. How they do so and what cognitive
processes are involved in regulating the physiological response of the
vas deferens in the male’s testes is under investigation
(delBarco-Trillo and Ferkin 2007b).
Hypotheses
----------
A previous ontology provides a conceptual framework within which to
conduct empirical research and fashion hypotheses (Franklin and Ferkin
2006). Formulating hypotheses is one of the functions of mathematical,
computational, and conceptual models. Thus, it’s reasonable to
formulate potentially testable hypotheses for the LIDA model. By doing
so, we hope to encourage empirical testing of our hypotheses. Here we
present a few selected testable hypotheses that may be tested with the
current LIDA technology.
1.
The Cognitive Cycle: The very existence of the cognitive cycle in
various species, along with its timing (asynchronously cascading
at a rate of roughly 5-10 hz) is a major hypotheses.
Neuroscientists have provided suggestive evidence for this
hypothesis (Lehmann et al. 1998; Halgren et al. 2002; Freeman
2003).
2.
Perceptual Memory: A perceptual memory, distinct from semantic
memory but storing some of the same contents, exists in humans (Nadel
1992; Franklin et al. 2005), and in many, perhaps most, animal
species.
3.
Transient Episodic-Like Memory: Humans have a content-addressable,
associative, transient episodic memory with a decay rate measured
in hours (Conway 2001). While perceptual memory seems to be almost
ubiquitous across animal species, we hypothesize that this
transient episodic memory is evolutionary younger, and occurs in
many fewer species (Franklin et al. 2005). We refer here to
episodic-like memory instead of to episodic memory, as in humans,
to avoid the controversy over phenomenal consciousness in animals,
about which the LIDA model takes no position (Ferkin et al. in
press). Further reference to episodic memory in non-human animals
should be read as episodic-like.
4.
Consolidation. A corollary to the previous hypothesis says that
events can only be encoded (consolidated) in long-term declarative
memory via transient episodic memory. This issue of memory
consolidation is still controversial among both psychologists and
neuroscientists (e.g. Lisman and Fallon 1999). However, the LIDA
model advocates such consolidation.
5.
Consciousness: Functional consciousness is implemented
computationally by way of a broadcast of contents from a global
workspace, which receives input from the senses and from memory
(Baars 1988, 2002).
6.
Conscious Learning: Significant learning takes place via the
interaction of functional consciousness with the various memory
systems (e.g. Standing 1973; Baddeley 1993). The effect size of
subliminal learning is quite small compared to conscious learning.
Note that significant implicit learning can occur by way of
unconscious inferences based on conscious patterns of input (Reber
et al. 1991). All memory systems represented in the model rely on
attention for their updating, either in the course of a single
cycle or over multiple cycles. (Franklin et al. 2005).
7.
Voluntary and Automatic Memory Retrievals: Associations from
transient episodic and declarative memory are retrieved
automatically and unconsciously during each cognitive cycle.
Voluntary retrieval from these memory systems may occur over
multiple cycles using volitional goals.
8.
Deliberative, volitional decision making: Such functionally
conscious decisions that deliberatively choose between
alternatives are, following Global Workspace Theory (Baars 1988
Chapter 9), are hypothesized in the LIDA model (Franklin 2000) to
follow William James’ ideomotor theory (James 1890). Thus a
decision is reached in favor of a proposed alternative when no
objection to it is raised. Volitional decision making is
inherently a multi-cyclic, higher-order cognitive process.
Connecting the LIDA Model and the behavior of a meadow vole
-----------------------------------------------------------
In what follows we will describe each of the steps in LIDA’s cognitive
cycle, stated in italicized text as if applying to a human, while also
carrying along their application in the mind of a hypothetical male
vole.
Imagine a male vole has turned a corner, and encountered scent marks
from different conspecifics (Ferkin and Johnston 1995). Some of these
scent marks are old and some are fresh, some are overlapping and some
are not. This male vole detects these marks, identifies the donors
that deposited the marks, and spends more time investigating the most
numerous and the freshest marks (Ferkin et al. 1999, 2001, 2004a, b,
2005). The male vole distinguishes between the different scent donors
and responds preferentially to the donors that are of most interest to
him. The most interesting donor may likely be a sexually receptive
female with whom he would attempt to copulate (delBarco-Trillo and
Ferkin 2004). The mechanism that the male voles used to discriminate
between the different scent donors would likely have involved
perceptual learning (Franklin and Ferkin 2006) Keep in mind that the
cognitive cycle to be described takes, in total, only a fifth of a
second or so to complete.
Here are the nine steps of the LIDA cognitive cycle together with an
example interpretation in the mind of our assumed male vole.
1.
Perception. Sensory stimuli, external or internal, are received
and interpreted by perception producing meaning. Note that this
step is preconscious.
In its perceptual memory the male vole categorizes the scent marks as
being from males or females (a category), as known (an individual),
and as sexually receptive (a feature) (Ferkin and Johnston 1995a, b).
During this step our vole scans its perceptual memory and makes
associations between scent marks and scent donors, assessing the
identity, sex, and reproductive condition of the scent donors (Ferkin
et al. 1999, 2004a, b, 2005)
This perceptual memory system identifies pertinent feeling/emotions
along with objects, categories and their relations.
In the male vole, feeling nodes for interest and for sexual arousal
are somewhat activated. If this is a sexually receptive female, for
example, all of these activated nodes are over threshold and become
part of the percept.
2.
Percept to Preconscious Buffer. The percept, including some of the
data plus the meaning, is stored in preconscious buffers of LIDA’s
working memory. In humans, these buffers may involve
visuo-spatial, phonological, and other kinds of information.
Feelings/emotions are part of the preconscious percept.
For the male vole, the percept has identified the freshest scent marks
coming from a female in postpartum estrus, a highly sexually receptive
female. These females readily mate when they encounter males. However,
females are only receptive to males for 12 hours after they deliver
pups (Ferkin et al. 2004a).
3.
Local Associations. Using the incoming percept and the residual
contents of the preconscious buffers (content from precious cycles
not yet decayed away), including emotional content, as cues, local
associations are automatically retrieved from transient episodic
memory (TEM) and from declarative memory.
The contents of the preconscious buffers, together with the retrieved
local associations from TEM and declarative memory, roughly correspond
to Ericsson and Kintsch’s (1995) long-term working memory and to
Baddeley’s (2000) episodic buffer. These local associations include
records of the agent’s past feelings/emotions, and actions, in
associated situations.
Assuming that our male vole possesses declarative memory, the
retrieved local associations may include the memory of a previous
sexual encounter with this particular female and his reaction to her,
a memory for what, when, and where (Ferkin et al. in press) For
example, our male vole may have a memory of this female, when she was
not in postpartum estrus, but simply pregnant and not sexually
receptive (Ferkin and Johnston 1995a, b), which allows our male vole
to anticipate that this female will only be in postpartum estrus for a
few hours, and then she becomes not interested in mating. Although
such expectation may come from either perceptual memory or semantic
memory, anticipating the what (a female is highly sexually receptive
for a relatively narrow window), the when (a female may no longer be
highly sexually receptive), and the where (the location of that female
relative to other female voles in the area), suggest that such
processing may involve an episodic–like memory (Ferkin et al. in
press)
4.
Competition for Attention. Coalitions of perceptual and memory
structures in the workspace compete to bring relevant, important,
urgent, or insistent situations to consciousness. (Consciousness
here is required only in the functional sense as defined in global
workspace theory and as defined by its role in the middle steps of
this cognitive cycle. Phenomenal (subjective) consciousness is not
assumed.) The competition may also include such coalitions from a
recently previous cognitive cycle. Present and past
feelings/emotions influence this competition for consciousness.
Strong affective content strengthens a coalition’s chances of
being attended to (Franklin and McCauley 2004).
In the male vole, one coalition that is on the lookout for sexual
opportunities will carry the other vole’s identity, her reproductive
status and readiness to mate, some details of the previous encounter,
and the feelings associated with the current percept and the previous
encounter. This coalition will compete with other such coalitions for
“consciousness,” but may not win the competition. Suppose our male’s
first encounter with that female’s odor indicated that she has also
attracted the attention of a predator, (fresh weasel scent marks are
present), which has also become part of the percept, along with a
strong fear. In this case, another coalition on the lookout for danger
may well win the competition, and the male vole may not respond by
seeking out this female.
5.
Broadcast of Conscious Contents. A coalition carrying content
gains access to the global workspace. Then, its contents are
broadcast throughout the system.
In humans, this broadcast is hypothesized to correspond to phenomenal
consciousness. No such assumption is made here. The conscious
broadcast contains the entire content of consciousness including the
affective portions.
Now imagine that the male vole did not detect a predator’s odor and
that the coalition about the female vole was attended to, that is, it
came to his “consciousness.”
Several types of learning occur. The contents of perceptual memory are
updated in light of the current contents of consciousness, including
feelings/emotions, as well as objects, categories, actions and
relations. The stronger the affect, the stronger the encoding is in
memory.
In the male vole, possibly along with others, representation in
perceptual memory for the particular female vole, for the category of
female voles, for readiness to mate, and for sexual interest would
each be strengthened.
Transient episodic memory is also updated with the current contents of
consciousness, including feelings/emotions, as events. The stronger
the affect, the stronger would be the encoding in memory. (At
recurring times not part of a cognitive cycle, the contents of
transient episodic memory are consolidated into long-term declarative
memory.)
If the male vole possesses a transient episodic memory, and studies
suggest that he may (Ferkin et al. in press), the event of having
again encountered this particular female vole, her condition, and his
reaction to her would be encoded, taking information from the
“conscious” broadcast.
Procedural memory (recent actions) is updated (reinforced) with the
strength of the reinforcement influenced by the strength of the
affect.
For the male vole, the prior acts of turning the corner and sniffing
the encountered scent marks would be reinforced. In this case, both
acts would have been learned and become familiar.
Thus, perceptual, episodic and procedural learning occur with the
broadcast in each cycle.
6.
Recruitment of Resources. Relevant behavior representations
respond to the conscious broadcast. These are typically
representations whose variables can be bound from information in
the conscious broadcast.
The responding representations may be those that can help to deal with
the current situation. Thus consciousness solves the relevancy problem
in recruiting internal resources with which to deal with the current
situation. The affective content (feelings/emotions), together with
the cognitive content, helps to attract relevant behavioral resources.
For the male vole, possibly among others, behavior representations for
turning the head, for turning the body, for sniffing the scent marks
and for moving in the direction that the female vole was traveling,
may respond to the information in the broadcast.
7.
Setting Goal Context Hierarchy. The recruited behavior
representations use the contents of consciousness, including
feelings/emotions, to instantiate new goal context hierarchies,
bind their variables, and increase their activation.
Goal contexts are potential goals, each consisting of a coalition of
behaviors, which, together, could accomplish the goal. Goal context
hierarchies can be thought of as high-level, partial plans of actions.
It is here that feelings and emotions most directly implement
motivations by helping to instantiate and activate goal contexts, and
by determining which terminal goal contexts receive activation. Other,
environmental, conditions determine which of the earlier goal contexts
receive additional activation.
For the male vole, a goal context hierarchy to seek out the female
vole would likely be instantiated in response to information from the
broadcast.
8.
Action Chosen. The action selection mechanism chooses a single
behavior, perhaps from a just instantiated goal context or
possibly from a previously active goal context.
This selection is heavily influenced by the various feelings/emotions.
The choice is also affected by the current situation, external and
internal conditions, by the relationship between the behaviors, and by
the residual strengths of various behaviors.
In the male vole, there may have been a previously instantiated goal
context for avoiding the weasel previously sensed. An appropriate
behavior in avoiding the predator may be chosen in spite of the
presence of the female vole. Alternatively, a beginning step in the
goal context for approaching and exploring the female vole may win
out.
9.
Action Taken. The execution of a behavior results in its action
being performed, which may have external or internal consequences,
or both.
This is LIDA taking an action.
If this particular male that has few opportunities to copulate with a
female, searching for the female would likely have been selected,
resulting in behavior codelets acting to turn the male in the
direction of the female, to sniff, and to begin his approach. If on
the other hand, our vole has frequent opportunities to mate with
females, he may stop his search for this female when he encounters the
odor of a weasel or a male conspecific (delBarco- Trillo and Ferkin
2004).
Sample experiments for tuning a Virtual Vole
--------------------------------------------
The computational LIDA architecture is composed of a number of closely
interconnected modules with their associated processes. Their
implementation is outlined in Table 1 below, which specifies the
conceptual name of the module, the name of its implementation in the
architecture, the source of inspiration for the data structure and
algorithms employed, and references to detailed explanations.
Module
Implementation
Source
References
Perceptual Memory
Slipnet
Copycat Architecture
Hofstadter and Mitchell (1995); Franklin (2005b)
Transient Episodic Memory
Sparse Distributed Memory
Sparse Distributed Memory
Kanerva (1988); D'Mello et al. (2005)
Declarative Memory
Sparse Distributed Memory
Sparse Distributed Memory
Kanerva (1988); D'Mello et al. (2005)
Procedural Memory
Scheme Net
Schema Mechanism
Drescher (1991); D'Mello et al. (2006b)
Action Selection
Behavior Net
Behavior Net
Maes (1989);, Negatu and Franklin (2002)
Table 1. LIDA modules and their implementations
Each of the LIDA modules and their associated processes involve a
number of internal parameters that must be specified before the model
can be used to replicate experimental data. Such specification of
parameters, the tuning of the model, is typically done by trial and
error so as to induce the model to replicate the data from one
specific experiment. This provisionally tuned model is then further
tuned to replicate data from both the original experiment and a second
experiment. The model is then considered tuned, and ready to try on
other, prospective, experiments.
Some change in the model would be needed if it proves difficult or
impossible to successfully tune some parameter. In that case, one must
conclude that the data structure or something in one of its associated
algorithms requires adjustment. Thus, the difficulty of tuning the
model serves as a sort of implicit metric measuring the correctness of
the model.
In the next subsections we describe experiments with voles that might
be expected to serve to tune a Virtual Vole, software agent operating
within a robotic simulator and simulating a live vole.
Odor preference tests for tuning the virtual vole
In a previous experiment, we quantified the olfactory response of
reproductively active voles to the odors of reproductively active
same- and opposite-sex conspecifics. The position of the male or
female donor was varied on the left- or right-side of the Y-maze to
prevent any side bias displayed by the subject (Ferkin and Seamon
1987). We recorded, continuously for 5 minutes, the amount of time
male and female subjects investigated the baskets containing the donor
voles. Thus, we showed that voles discriminate between and respond
preferentially to opposite-sex conspecifics over same-sex conspecifics
(Ferkin and Seamon 1987).
We also performed opposite-sex donor tests in which male subjects were
exposed to scent marks of ovariectomized + blank treated females (not
sexually receptive) and ovariectomized + estradiol treated females
(sexually receptive), and female subjects were exposed to scent marks
of gonadectomized + blank treated males (not sexually receptive) and
gonadectomized + testosterone treated males (sexually receptive). Each
male and female subject was exposed to a unique pair of opposite-sex
odor donors. We found that male and female subjects spent more time
investigating opposite-sex conspecifics given hormone replacement than
opposite-sex conspecifics not given hormone replacement. Thus, voles
prefer opposite-sex conspecifics that are sexually receptive to those
that are not sexually receptive (Ferkin and Johnston 1993, 1995a, b).
The Y-maze apparatus used in Ferkin and Seamon (1987) can be simulated
within a robotic simulator along with the various scent markings. This
would allow a virtual vole to act as subject. Knowing ahead of time
the desired range of results would allow the tuning of the various
parameters in the several modules of the LIDA architecture as
implemented in the virtual vole. As mentioned above, replication of
these and other such experiments would allow the testing of the
implementation of the LIDA model in control of the virtual vole.
Episodic-like memory tests for verifying the tuning of the virtual
vole
As the initial tuning of the internal parameters of the virtual vole
would have been done using odor preference tests, the question remains
of whether these tunings are specific to only those tests. Or, is the
tuning of parameters sufficiently general to make the virtual vole a
good simulation of live meadow voles in a variety of experimental
situations? We suggest testing for the generality of the parameter
tuning by replicating other previously performed experiments with live
meadow voles. One possibility would be the experiments on
episodic-like memory.
Episodic-like memory, the memory for events, allows an animal to
recollect the what, the where and the when of what happened. The LIDA
model asserts that such episodic-like memory comes in two forms,
transient episodic-like memory and declarative memory. Transient
episodic-like memory lasts only a relatively short period of time, a
few hours or a day in humans. In voles with their much shorter life
span, it may be reasonable to assume an even more rapid decay in
transient episodic-like memory. Declarative memory is long-term
episodic-like memory that in humans may last for decades or a
lifetime. This section describes already completed experiments that
may be replicated with virtual voles to further test and adjust the
tuning of their parameters. They also serve as background for possible
future experiments designed to tease out the distinction between
transient episodic-like memory and declarative memory in voles, if the
latter exists.
Despite the controversy swirling around the ability of animals to
recollect specific aspects of past events (Clayton and Griffiths 1998;
Tulving 2005), it is not difficult to imagine that some animals may
use information from such past events to secure a mate. An important
feature that often characterizes most non-human mammals is that
females do not mate with males when they are not in a heightened state
of sexually receptivity, such as estrus or postpartum estrus (Bronson
1989). Thus, for many species of mammals, and particularly the
majority of whom in which opposite-sex conspecifics live separately
during the breeding season, males should be able to discriminate among
females in different states of sexual receptivity. They should be able
to identify females that are in a heightened reproductive state, their
location, and the amount of time that the females are in this
heightened state. Such a capacity would benefit, for example, a male
meadow vole, a microtine rodent.
Adult male and female meadow voles live separately during the breeding
season. At this time of year, female voles tend to occupy territories
that are fixed spatially, but are dispersed widely across the home
range of several males (Madison 1980). Female voles are induced
ovulators and do not undergo estrous cycles (Milligan 1982; Meek and
Lee 1993). Thus, the reproductive condition and sexual receptivity
varies among female voles during the breeding season. That is, female
voles may be pregnant, lactating, both pregnant and lactating, neither
pregnant nor lactating, or in a period of heightened sexual
receptivity during postpartum estrus (Keller 1985). Postpartum estrus
females are more likely to mate with a male than females that are not
pregnant or lactating, or females that are pregnant, lactating or both
(Ferkin et al. 2004; delBarco-Trillo and Ferkin 2007a).
Sexual receptivity in female varies and they enter PPE asynchronously.
To increase his fitness, male meadow voles should mate with as many
females as possible (Boonstra et al. 1993), particularly those females
that have entered postpartum estrus (Ferkin et al. 2004;
delBarco-Trillo and Ferkin 2007a). Thus, we hypothesize that after a
single visit to a female, male voles would later recollect her
previous reproductive state (what); her location (where), and how long
she would be in that reproductive state (when) (Ferkin et al. in press),
thus demonstrating episodic-like memory. The experimental design of
this experiment (Ferkin et al. in press) is described below.
All female voles were between 125-135 days of age when used in the
tests. Female meadow voles do not undergo estrus cycles (Milligan
1982; Keller 1985). To represent different levels of female
receptivity, we used females that were pregnant for 20 days (day 20
pregnant), in postpartum estrus, females that were not pregnant or
lactating, termed reference females, and day 2 lactating females.
Gestation lasts 21 days in voles, thus day 20 pregnant female voles
deliver their litters within 24 hours (Keller 1985).
Immediately after parturition, these females enter postpartum estrous
(PPE), a period of heightened sexual receptivity, which lasts 8-12
hours (Keller 1985; Ferkin et al. 2004; delBarco-Trillo and Ferkin
2007a). The postpartum estrus females had delivered pups 4-6 hours
prior to testing. Reference females were not currently pregnant or
lactating (Ferkin and Johnston 1995). The reference females had
previously delivered a litter about 3-4 weeks before being used in the
experiment (see below); these females had lived singly for
approximately 21 days before testing began.
In experimental conditions 4 and 5 (see below), we used females that
were in their second day of lactation for each condition. Lactation is
14-16 days in duration, and pups are weaned when they are 16-18 days
old (Keller 1985). The day 2 lactating females were no longer in
postpartum estrus and thus were no longer in a heightened state of
sexual receptivity (Ferkin and Johnston 1995). The postpartum estrus
females and day 2 lactating females had not lived with their mate for
17 and 18 days, respectively, before the testing began.
It is important to note that postpartum estrus female voles are in a
heightened state of reproductive receptivity and readily mate with
males (Keller 1985; Ferkin and Johnston 1995; delBarco-Trillo and
Ferkin 2007a). In contrast, reference females, day 20 females, and day
2 lactating females are not in a heightened state of sexual
receptivity, but they may mate (Ferkin and Johnston 1995;
delBarco-Trillo and Ferkin 2004, 2006). In addition, postpartum estrus
females produce odors that are more attractive to males relative to
those produced by females that are day 20 pregnant, day 2 lactating,
or reference females, who produce odors that are similar in their
attractiveness to males (Ferkin and Johnston 1995; Ferkin et al.
2004).
All behavioral observations were performed on voles placed in a
T-shaped apparatus (Fig. 2). We used two opaque Plexiglas cages with
wired tops for observation purposes. The large boxes served to house
the female donors. There was a transparent divider with small holes
between the females’ living area and the area that males explored.
This divider allowed males to investigate the female’s living area
without coming into direct contact with that female.
Test for Episodic-Like Memory
We conducted an experiment, with five experimental conditions, in
which male subjects were exposed to unique female donors (Ferkin et
al. in press). Each experimental condition contained two phases, an
exposure phase and a test phase. In both phases of the five
experimental conditions, a male meadow vole from one of the above
treatment groups was placed into the starting box located at the base
of the T-shaped arena (Fig. 2) for 30 seconds before the gate was
lifted and the male was allowed to explore the entire apparatus. Each
male underwent a single exposure and single test (see below).
Experimental Condition 1 –During the exposure phase, male voles were
placed into an apparatus that housed a reference female in one box and
a day 20 pregnant female in the other box (Fig. 2). During the
exposure phase, we recorded continuously for 10 minutes, the total
amount of time male voles spent in the arms of the apparatus that
housed each female donor (Fig. 2). We also noted the position of the
home-boxes (left- or right-side of the apparatus) that housed each
particular female donor. The position of a particular female’s
home-box in the left- or right-side of the apparatus was alternated
for each male subject during the exposure phase. After the 10-minute
exposure, the male was returned to its own cage. Then, we disconnected
the two-female home-boxes from the apparatus, and cleaned and
disinfected the apparatus.
The test phase took place 0.5 hour after the exposure phase. During
the test phase, the male voles were re-introduced into the apparatus
that now contained boxes that housed no female donors; the boxes
contained only clean wood chip bedding. We recorded continuously for
10 minutes, the total amount of time that male voles spent
investigating the arm of the apparatus that previously housed the
reference female that they were exposed to and the arm that previously
housed the day 20 pregnant female. During the test phase male voles
spent similar amounts of time investigating the arm of the apparatus
that would have housed the day 20 pregnant female and the arm of the
apparatus that would have housed the reference female (Ferkin et al.
in press).
Experimental Condition 2 - Male voles were exposed to an arena
containing a day 20 pregnant female and a postpartum estrus female.
0.5 hour later, male voles were allowed to investigate an empty arena.
We recorded the initial choice of the male vole and the amount of time
that he spends in both arms of the arena. During the test phase male
voles spent more time investigating the arm of the apparatus that
would have housed the postpartum estrus females than the arm of the
apparatus that would have housed the day-20 pregnant female (Ferkin et
al. in press).
Experimental Condition 3 - Male voles were exposed to an arena
containing a day 2 lactating female and a reference female. 0.5 hour
later, male voles were allowed to investigate an empty arena. We
recorded the initial choice of the male vole and the amount of time
that he spends in both arms of the arena. During the test phase male
voles spent similar amounts of time investigating the arm of the
apparatus that would have housed the day 2 lactating female and the
arm of the apparatus that would have housed the reference female
(Ferkin et al. in press).
Experimental Condition 4 - Male voles were exposed to an arena
containing a day 20 pregnant female and a reference female. 24 hours
later, male voles were allowed to investigate an empty arena. The test
phase took place 24 hours after the exposure phase. At this time, the
day 20 pregnant female had delivered pups and had entered into
postpartum estrus. During the test phase, the male voles were
re-introduced into the apparatus that now contained boxes that housed
no female donors; the boxes contained only clean wood chip bedding
(Fig. 2). During the test phase, which occurred 24 hours after the
exposure phase, males spent more time investigating the arm of the
apparatus that would have contained the postpartum estrus female than
the arm of the apparatus that would contained the reference female
(Ferkin et al. in press)
Experimental Condition 5 - Male voles were exposed to an arena
containing a postpartum estrus female and a reference female.
Twenty-four hours later, male voles were allowed to investigate an
empty arena. We recorded the initial choice of the male vole and the
amount of time that he spends in both arms of the arena. During the
test phase, which occurred 24 hours after the exposure phase, male
voles spent similar amounts of time investigating the arm of the
apparatus that would have housed the day 2 lactating female and the
arm of the apparatus that would have housed the reference female
(Ferkin et al. in press).
The results of these experiments suggest that male voles may have the
capacity to recall the what, where, and when of a single past event,
which may allow males to remember the location of females who would
currently be in heightened states of sexual receptivity. Viewed from
the LIDA model, the outcomes of Experimental conditions 1-3 indicate
recollection of an event after a time interval of 0.5 hour between the
exposure of a subject male vole to female odor and its later testing
can be attributed to transient episodic-like memory (Ferkin et al. in
press).
In that the life span of a meadow vole is only about four months in
the wild and approximately 18 months in captivity (Sheridan and
Tamarin 1988; Ferkin and Leonard in press), we suspect that the bottom
end of the time span for testing for long-term episodic-like memory
would be 24 hours or less. However, experiments can be repeated using
a 48 hour time interval, which would correspond to long-term episodic
in humans,
Replication of Episodic-Like Memory experiments using LIDA model
virtual voles. These tests would involve placing a virtual vole in a
virtual arena that simulates that described for real voles (Fig. 2)
and following experimental methods for virtual voles identical to
those of the episodic-like memory experiments described above.
Specifically, we will use virtual voles and a virtual arena to
replicate the tests described above for a live vole in experimental
conditions 1-5. During the test trials with the virtual voles, we will
identify the initial choice of the virtual male voles and the total
amount of time that they spend investigating the arm of the apparatus
that previously housed the virtual conspecific females. By doing so,
we would be able to compare the response of the virtual male vole with
those of the live male voles and test the efficacy of the LIDA model
for predicting the behavior of voles.
Successful replication of these episodic-like memory experiments with
a virtual vole would demonstrate the efficacy of one aspect of the
LIDA model. Also, replication of these experiments, both in vivo and
virtual, would allow the LIDA model to distinguish transient episodic
memory in voles with its rapid decay rate from declarative (long-term
episodic-like) memory, which can last a lifetime (See hypothesis 3
above).
Figure 2 – The testing arena for Episodic-like memory in voles
--------------------------------------------------------------

A Possible experiment for testing a LIDA hypothesis
With a properly tuned virtual vole in hand it becomes possible to test
the various hypotheses listed above that are derived from the LIDA
model. In this section we suggest one such possible experiment
designed to test Hypothesis 8. The earlier hypotheses involve
processes that are thought to operate within a time frame of a very
few hundred milliseconds, making them difficult, though not
impossible, to test using live animals. A test of Hypothesis 3 was
described in the previous section.
Testing for Volitional Decision Making in Meadow Voles
Hypothesis 8 predicts that some animals are capable of deliberative,
volitional decision making. Humans deliberate and make volitional
decisions. Do other animals such as meadow voles have this ability? In
many animal experiments the subject is faced with a forced choice of
response to a stimulus, say push this lever or that. Such experimental
situations almost always have involved learning on the part of the
subject. In this case, the subject’s action selection is likely to
have resulted from perceptual recognition and learned action
selection, all within a single cognitive cycle, rather than from
deliberative decision making. On the contrary, studies of searching
for live prey suggest that the jumping spider, Portia labiata, may
engage in deliberate decision making. In the field studies (Wilcox and
Jackson. 2002) these spiders were observed to spend a number of tens
of minutes out of sensory contact, circling behind and above a prey
spider, before lowering itself on a thread and ambushing the prey,
which has appeared in a location that was “anticipated” by the spider.
Such ambush behavior would seem to require deliberation, and even
planning. This behavior on the part of these jumping spiders has also
been tested experimentally by Tarsitano (2006). Here we suggest a
version of Tarsitano’s (2006) experiments, adapted to test the
hypothesis that meadow voles are capable of making decisions
deliberatively. This section will briefly describe such an experiment.
The experimental apparatus consists of a relatively simple maze
together with a platform above the maze from which the entire maze can
be viewed through a transparent floor. The maze has two disjoint
zigzagged arms that interleave with one another in three dimensions in
some complex way, with the ends of the two arms separated.
In the exposure phase of the experiment the subject male vole has the
run of the platform from one end of which he can see and smell a
postpartum estrus female vole positioned at the end of one of the arms
of the maze. The subject male vole can sense but not approach the
female vole, and can inspect the maze below the platform through its
transparent floor. In the test phase of the experiment the subject
male vole is positioned at the beginning of the maze where he is faced
with a choice of the two entrances of the two arms, and where he is
unable to detect the postpartum estrus female vole. Having no
procedural learning on which to depend, but only the perceptual
learning from his inspection of the maze from his earlier vantage
point on the platform, the subject vole, faced with the entrances to
the two arms, must carry out a deliberative selection of which of the
two arms to explore to encounter the postpartum estrus female. Based
on the preferences of male voles, for postpartum estrus females
(Ferkin et al. 2005), a male would demonstrate deliberative decision
making by initially choosing and exploring the arm that will bring him
into contact with the postpartum estrus female. Such a response by the
male voles would have been the result of prior deliberative planning
and a prior volitional decision to seek the postpartum estrus female
along the appropriate arm. Such a decision will have likely occurred
while the subject vole was exploring the platform, and discovered that
one arm of the maze led to the postpartum estrus female and the other
arm did not lead to the postpartum estrus female. This choice cannot
be successfully made with the sensory information available to the
subject vole positioned at the beginning of the maze. The capacity to
make the appropriate choice and choose the direct path to the female
would provide support for the hypothesis that meadow voles make
deliberative decisions.
Experimenting with a Cognitive Robot
------------------------------------
In principle, it should be possible to perform real world experiments
using an artificial animal, say an artificial vole, in the form of a
cognitive robot controlled by some cognitive architecture based on,
for example, the LIDA model. Using such a cognitive robot would retain
all the benefits described above for the use of software agent
simulations of animals, say virtual voles. In addition, the use of
such artificial animals/cognitive robots might be expected to reveal
real world issue or difficulties that could be obscured by the use of
virtual animals in a simulated environment.
The major problem with designing cognitive robots for such a purpose
would seem to be sensing. It is difficult to imagine an artificial
vole with the acute sense of smell of a real vole. With the advent of
nanotechnology and other new techniques, artificial olfaction is
becoming a reality (Pearce et al. 2002). Replicating experiments using
cognitive robots as artificial animals may someday become a reality.
Conclusion
----------
We conclude that it is in principle possible to employ virtual animals
in the form of software agent simulations to benefit biological
theory. Controlled by cognitive architectures such as LIDA, such
virtual animals allow biologists to test their theories directly by
replicating experiments within a virtual environment. To do so
requires that the controlling cognitive architecture, like LIDA, be
sufficiently broad and comprehensive to serve to control a software
agent. Thus broad, comprehensive theories of animal cognition should
prove themselves of value to biologists.
References
----------
Allen C (1997): Animal cognition and animal minds. In Machamer P,
Carrier M (eds), Philosophy and the Sciences of the Mind: Pittsburgh
University Press and the Universitätsverlag Konstanz, pp 227-243.
Alsop B (1998) Receiver operating characteristics from non-human
animals: Some implications and directions for research with humans.
Psychonomic Bulletin & Review 5:239-252.
Asada M, MacDorman KF, Ishiguro H, Kuniyoshi Y (2001) Cognitive
developmental robotics as a new paradigm for the design of humanoid
robots. Robotics and Autonomous Systems 37:185–193.
Baars BJ (1988) A Cognitive Theory of Consciousness. Cambridge:
Cambridge University Press.
Baars BJ (2002) The conscious access hypothesis: origins and recent
evidence. Trends in Cognitive Science 6:47–52.
Baars BJ, Franklin S (2003) How conscious experience and working
memory interact. Trends in Cognitive Science 7:166–172.
Baddeley A, Conway M, Aggleton J (2001) Episodic Memory. Oxford:
Oxford University Press.
Baddeley AD, Hitch GJ (1974) Working memory. In Bower GA (ed), The
Psychology of Learning and Motivation. New York: Academic Press, pp
47–89.
Barsalou LW (1999) Perceptual symbol systems. Behavioral and Brain
Sciences 22:577–609.
Bekoff M, Allen C, Burghardt GM (2002) The Cognitive Animal.
Cambridge, MA: MIT Press.
Boonstra R, Xia X, Pavone L (1993) Mating system of the meadow vole,
Microtus pennsylvanicus. Behavioral Ecology 4: 83-89.
Boysen ST, Himes GT (1999) Current issues and emerging theories in
animal cognition. Ann. Rev. Psych. 50:683–705.
Brooks RA (1991) How to build complete creatures rather than isolated
cognitive simulators. In VanLehn K (ed), Architectures for
Intelligence. Hillsdale, NJ: Lawrence Erlbaum Associates, pp 225–239.
Bronson FH (1989) Mammalian Reproductive Biology. University of
Chicago Press, Chicago
Byrne RW, Bates LA (2006) Why are animals cognitive? Current Biology
16:R445–R448.
Clark A, Grush R (1999): Towards a cognitive robotics. Adaptive
Behavior 7:5-16.
Clayton NS, Griffiths, DP (1998). Episodic-like memory during cache
recovery by scrub jays. Nature 395: 272-274
Conway MA (2002): Sensory-perceptual episodic memory and its context:
autobiographical memory. In Baddeley A, Conway M, Aggleton J (eds),
Episodic Memory. Oxford: Oxford University Press, pp 53–70.
Crick F, Koch C (2003) A framework for consciousness. Nature
Neuroscience 6:119–126.
D'Mello SK, Ramamurthy U, Franklin S (2005) Encoding and Retrieval
Efficiency of Episodic Data in a Modified Sparse Distributed Memory
System, Proceedings of the 27th Annual Meeting of the Cognitive
Science Society. Stresa, Italy.
D'Mello SK, Franklin S, Ramamurthy U, Baars BJ (2006a) A Cognitive
Science Based Machine Learning Architecture, AAAI 2006 Spring
Symposium Series Sponsor: American Association for Artificial
Intelligence. Stanford University, Palo Alto, California, USA.
D'Mello SK, Ramamurthy U, Negatu A, Franklin S (2006b) A Procedural
Learning Mechanism for Novel Skill Acquisition. In Kovacs T, Marshall
JAR (eds), Proceeding of Adaptation in Artificial and Biological
Systems, AISB'06, Vol 1. Bristol, England: Society for the Study of
Artificial Intelligence and the Simulation of Behaviour, pp 184–185.
delBarco-Trillo, J. & Ferkin, M. H. (2004) Male mammals respond to a
risk of sperm competition conveyed by odours of conspecific males.
Nature 431: 446-449.
delBarco-Trillo J, Ferkin MH (2006) Male meadow voles respond
differently to risk and intensity of sperm competition. Behavioral
Ecology 17: 581-585.
delBarco-Trillo J, Ferkin MH (2007a) Female meadow voles, Microtus
pennsylvanicus, experience a reduction in copulatory behavior during
postpartum estrus. Ethology 113: 466-473.
delBarco-Trillo J. Ferkin MH (2007b). Increased sperm numbers in the
vas deferens of meadow voles, Microtus pennsylvanicus, in response to
odors of conspecific males. Behavioral Ecology and Sociobiology. 61:
1759-1764.
Dere E, Kart-Teke E, Huston JP, De Souza Silva MA (2006) The case for
episodic memory in animals. Neuroscience and Biobehavioral Reviews
30:1206–1224.
Drescher GL (1991) Made-Up Minds: A Constructivist Approach to
Artificial Intelligence. Cambridge, MA: MIT Press.
Ericsson KA, Kintsch W (1995) Long-term working memory. Psychological
Review 102: 211–245.
Ferkin MH (1988) The effect of familiarity on social interactions in
meadow voles, Microtus pennsylvanicus: a laboratory and field study.
Animal Behaviour 36: 1816-1822.
Ferkin MH (1989) Adult-weanling recognition among captive meadow voles
(Microtus pennsylvanicus). Behaviour 118: 114-124.
Ferkin MH (2007) Effects of previous interactions and sex on
over-marking in meadow voles. Behaviour. 144: 1297-1313.
Ferkin MH, Johnston RE (1993) Roles of gonadal hormones on controlling
sex-specific odors in meadow voles (Microtus pennsylvanicus). Hormones
and Behavior 27: 523-538.
Ferkin,MH, Johnston RE (1995a) Meadow voles, Microtus pennsylvanicus,
use multiple sources of scent for sexual recognition. Animal Behaviour
49: 37-44.
Ferkin MH, Johnston RE (1995b) Effects of pregnancy, lactation, and
postpartum oestrous on odour signals and the attraction to odours in
female meadow voles, Microtus pennsylvanicus. Animal Behaviour. 49:
1211-1217.
Ferkin MH, Leonard ST (2005) Self-grooming by rodents in social and
sexual contexts. Acta Zool. Sinica. 51: 772-779.
Ferkin MH, Leonard ST Age of the subject and scent donor affects the
amount of time that voles self-groom when they are exposed to odors of
opposite-sex conspecifics. in Beynon R, Hurst J, Roberts C, Wyatt T
(eds), Chemical Signals in Vertebrates 11. Springer Press. in press.
Ferkin MH, Li HZ (2005) A battery of olfactory-based screens for
phenotyping the social and sexual behaviors of mice. Physiology and
Behavior 85: 489-499.
Ferkin MH, Seamon JO (1987) Odor preferences and social behavior in
meadow voles, Microtus pennsylvanicus: seasonal differences. Canadian
Journal of Zoology 65: 2931-2937.
Ferkin MH, Lee DN, Leonard ST (2004a) The reproductive state of female
voles affects their scent marking behavior and the responses of male
conspecifics to such marks. Ethology 110: 257-272.
Ferkin MH, Li HZ, Leonard ST (2004b) Meadow voles and prairie voles
differ in the percentage of conspecific marks that they over-mark.
Acta Ethologica 7: 1-7.
Ferkin MH, Mech SG, Paz-y-Mino C (2001). Scent marking in meadow voles
and prairie voles: a test of three hypotheses. Behaviour 138:
1319-1336.
Ferkin MH, Tamarin RH, Pugh SR (1992) Cryptic relatedness and the
opportunity for kin recognition in microtine rodents. Oikos 63:
328-332.
Ferkin MH, Combs A, delBarco-Trillo J, Pierce AA, Franklin S. Meadow
voles display a capacity for what, where, and when. Animal Cognition.
in press.
Ferkin MH, Leonard ST, Bartos K, Schmick MK (2001a) Meadow voles and
prairie voles differ in the length of time they prefer the top-scent
donor of an over-mark. Ethology 107: 1099-1114.
Ferkin MH, Pierce A A, Sealand RO, delBarco-Trillo J (2005) Meadow
voles, Microtus pennsylvanicus, can distinguish more over-marks from
fewer over-marks. Animal Cognition 8: 82-89.
Foote AD, Griffin RM, Howitt D, Larsson L, Miller PJO, Hoelzel AR
(2006) Killer whales are capable of vocal learning. Biology Letters
2:509–512.
Franklin S (2000) Deliberation and Voluntary Action in ‘Conscious’
Software Agents. Neural Network World 10:505–521.
Franklin S (2005) A "Consciousness" Based Architecture for a
Functioning Mind. In Davis DN (ed), Visions of Mind. Hershey, PA:
Information Science Publishing, pp 149–175.
Franklin S (2005a) Evolutionary Pressures and a Stable World for
Animals and Robots: A Commentary on Merker. Consciousness and
Cognition 14:115–118.
Franklin S (2005b) Cognitive Robots: Perceptual associative memory and
learning, Proceedings of the 14th Annual International Workshop on
Robot and Human Interactive Communication (RO-MAN 2005), pp 427-433.
Franklin S, Baars BJ, Ramamurthy U, Ventura M (2005) The role of
consciousness in Memory. Brains, Minds and Media 1:1–38.
Franklin S, Ferkin MH (2006) An Ontology for Comparative Cognition: a
Functional Approach. Comparative Cognition & Behavior Reviews 1:36–52.
Franklin S, Graesser AC (1997) Is it an Agent, or just a Program? A
Taxonomy for Autonomous Agents, Intelligent Agents III. Berlin:
Springer Verlag, pp 21–35.
Franklin S, Ramamurthy U (2006) Motivations, Values and Emotions:
Three sides of the same coin, Proceedings of the Sixth International
Workshop on Epigenetic Robotics, Vol 128. Paris, France: Lund
University Cognitive Studies, pp 41–48.
Gibson JJ (1979) The Ecological Approach to Visual Perception. Mahwah,
New Jersey: Lawrence Erlbaum Associates.
Glenberg AM (1997) What memory is for. Behavioral and Brain Sciences
20:1–19.
Hofstadter DR, Mitchell M (1995) The Copycat Project: A model of
mental fluidity and analogy-making. In Holyoak KJ, Barnden JAORoEe
(eds), Advances in connectionist and neural computation theory, Vol.
2: logical connections. Norwood N.J.: Ablex, pp 205–267.
James W (1890) The Principles of Psychology. Cambridge, MA: Harvard
University Press.
Kaelbling LP, Littman ML, Moore AW (1996) Reinforcement Learning: A
Survey. Journal of Artificial Intelligence Research 4:237–285.
Kamil AC (1998) On the proper definition of cognitive ethology. In
Balda R, Pepperberg I, Kamil AO (eds), Animal cognition in nature: The
convergence of psychology and biology in laboratory and field. New
York: Academic Press, pp 1–28.
Kanerva P (1988) Sparse Distributed Memory. Cambridge MA: The MIT
Press.
Keller, B. L. (1985) Reproductive patterns. in Tamarin RH, (ed),
Biology of new world Microtus. 8th edn. (American Society of
Mammalogists, Special Publication, Lawrence, KS, pp 725-778.
Kruschke JK (2001) Toward a unified model of attention in associative
learning. Journal of Mathematical Psychology 45:812-863.
Leonard ST, Ferkin MH (2005) Seasonal differences in self-grooming in
meadow voles, Microtus pennsylvanicus. Acta Ethologica. 8: 86-91.
Madison DM (1980) An integrated view of the social biology of meadow
voles, Microtus pennsylvanicus. The Biologist 62:20-33.
Maes P (1989) How to do the right thing. Connection Science 1:291–323.
Maturana HR (1975) The Organization of the Living: A Theory of the
Living Organization. International Journal of Man-Machine Studies
7:313–332.
Maturana, H R, Varela FJ (1980) Autopoiesis and Cognition: The
Realization of the Living, Dordrecht. Netherlands: Reidel.
Meek LR, Lee TM (1993) Prediction of fertility by mating latency and
photoperiod in nulliparous and primiparous meadow voles (Microtus
pennsylvanicus). Journal of Reproduction and Fertility 97:353-357.
Merker B (2005) The liabilities of mobility: A selection pressure for
the transition to consciousness in animal evolution. Consciousness and
Cognition 14:89–114.
Milligan SR (1982) Induced ovulation in mammals. Oxford Reviews of
Reproduction 4:1-46.
Negatu A, Franklin S (2002) An action selection mechanism for
'conscious' software agents. Cognitive Science Quarterly 2:363–386.
Negatu A, McCauley TL, Franklin S (in review) Automatization for
Software Agents.
Pearce TC, Schiffman SS, Nagle HT, Gardner JW (2002) Handbook of
Machine Olfaction: Electronic Nose Technology. Weinheim: Wiley-VCH.
Pierce AA, Ferkin MH (2005) Re-feeding and restoration of odor
attractivity, odor preference, and sexual receptivity in food-deprived
female meadow voles. Physiology and Behavior. 84:553-561.
Pierce AA, Ferkin MH, Williams TK (2005) Food-deprivation-induced
changes in sexual behavior of meadow voles, Microtus pennsylvanicus.
Animal Behaviour 70:339-348.
Saksida LM (1999) Effects of similarity and experience on
discrimination learning: A non associative connectionist model of
perceptual learning. Journal of Experimental Psychology: animal
Behavior Processes 25:308-323.
Sheridan M,Tamarin RH (1988) Space use, longevity, and reproductive
success in meadow voles. Behavioral Ecology and Sociobiology 22:
85-90.
Skarda C, Freeman WJ (1987) How Brains Make Chaos in Order to Make
Sense of the World. Behavioral and Brain Sciences 10:161–195.
Sloman A (1999: What Sort of Architecture is Required for a Human-like
Agent? In Wooldridge M, Rao AS (eds), Foundations of Rational Agency.
Dordrecht, Netherlands: Kluwer Academic Publishers, pp 35–52.
Tarsitano M (2006) Route selection by a jumping spider (Portia labiata)
during the locomotory phase of a detour. Animal Behavior 72:1437–1442.
Tulving E (1983) Elements of episodic memory. Oxford: Clarendon Press.
Tulving E (2005) Episodic memory and autonoesis: uniquely human? in
Terrace HS, Metcalfe J (eds), The missing link in cognition, New York,
Oxford University Press, pp 3-56.
Varela FJ, Thompson E, Rosch E (1991) The Embodied Mind. Cambridge,
MA: MIT Press.
Wilcox S, Jackson R (2002) Jumping Spider Tricksters: Deceit,
Predation, and Cognition. In Bekoff M, Allen C, Burghardt GM (eds),
The Cognitive Animal. Cambridge, MA: MIT Press, pp 27–33.
Winston PH (1992): Artificial Intelligence, 3rd ed. Boston: Addison
Wesley.
Zhang Z, Dasgupta D, Franklin S (1998) Metacognition in Software
Agents using Classifier Systems, Proceedings of the Fifteenth National
Conference on Artificial Intelligence. Madison, Wisconsin: MIT Press,
pp 83–88.