BioSystems 56 2000 33 – 42
Autonomous choice in the learning process of a turtle Chinemys ree6esii
Shusaku Nomura
a,
, Yukio-Pegio Gunji
a,b
a
Graduate School of Science and Technology, Kobe Uni6ersity, Nada, Kobe
657
-
8501
, Japan
b
Department of Earth and Planetary Sciences, Faculty of Science, Kobe Uni6ersity, Nada, Kobe
657
-
8501
, Japan Received 2 June 1999; received in revised form 11 January 2000; accepted 21 January 2000
Abstract
We studied animal’s learning of spatial discrimination in an experimental environment. Turtles, Chinemys ree6esii, were employed for the study. We focused on two independent aspects: 1 turtle’s success rate in the task, which is
the most common criterion to estimate the ability of the animals, and 2 the statistical properties of the time interval of the task, which is independent on the spatial criterion. For a statistical analysis, we employed the scheme of power
law distributions which was recently used to estimate animal behaviors in relation to the idea of the fractal. We addressed the basic problem of whether these two criteria, or any other criteria for this matter, could or could not
exclude an observer who studies the animal behavior. To demonstrate inseparability of an observer and the object, we conducted three different learning experiments: 1 complete spatial discrimination, 2 incomplete spatial
discrimination, 3 another, different, complete discrimination, in this order. The incomplete one was taken to mean incomplete only for an observer. Our experiments reveal that the same result success rate was perceived differently
by the animal if the attitude of the observer to the experiments differed. This observation comes to suggest that the notion of autonomous choice on the part of an animal is contingent upon the inseparability between an observer and
the object. © 2000 Elsevier Science Ireland Ltd. All rights reserved.
Keywords
:
Autonomous choice; Discrimination learning; Zipf analysis; Power law www.elsevier.comlocatebiosystems
1. Introduction
An idea currently prevailing in ethology is that animal learning processes are based on instinct,
thus determined by genetic codes, and are subject to stimuli in the animal’s environment Gould and
Marler, 1987. On the other hand, many psycho- logical studies have also been conducted to un-
derstand the ability of an animal. In the latter, the issue has been whether the animal can acquire a
particular behavior to accomplish a task. The animal behavior has generally been evaluated by
referring to the success rate of a given task, such as the decrease of the time interval necessary to
achieve the goal. Especially in discrimination learning experiments, scientists evaluate animal
behaviors from a spatial point of view, for in- stance, recording which way an animal took, ei-
ther left or right.
Corresponding author. 0303-264700 - see front matter © 2000 Elsevier Science Ireland Ltd. All rights reserved.
PII: S 0 3 0 3 - 2 6 4 7 0 0 0 0 0 6 9 - 1
In these studies the learning process is consid- ered to be algorithmic, in which an animal is
supposed to manipulate some specific codes, ei- ther genetic or neuronal. This sort of algorithmic
paradigm upon manipulating specific codes, how- ever, does not take into account, the notion of
animal autonomy or decision-making property. Earlier behaviorist theories attempted to account
for animal behaviors in term of stimulus-response S-R associations Hull, 1943, 1952. Cognitivist
initially appealed to the notion of cognitive maps as differing from behaviorist Tolman, 1948. But
recently, like in the notion of ‘place-cell’ in the hippocampus O’Keefe and Dostrovsky, 1971,
both come to hold basically a common frame- work assuming a one-to-one mapping between an
input and an output.
We now want to examine the notion of animal autonomy and the aspect of learning process, as
shedding some critical remarks upon such a framework. Our issue will be a critical examina-
tion of the role of the observer necessarily intro- duced into the framework.
One implication of behaviorism is that behavior should be observed as a phenomenon, without
making any conjectures about what might be going on within the subject’s mind. It is viewed as
a ‘black box’. However, what does matter more in behaviorism is the notion of ‘external observers’
who could definitely observe, describe and encode animal external behaviors. Consider, for instance,
the following learning experiment: An observer cannot understand animal’s internal processes in
the ‘black box’, but can definitely prepare a prob- lem as an input to an animal and observe the
consequential animal behavior, for instance, in terms of success or failure. It implies that the
observer knows the answer to the problem, while the animal does not. In other words, the observer
can give an animal a problem which is uniquely defined, and ‘the problem’ is not only for the
animal but also for the observer. Note that it remains to be seen whether the animal regards the
problem set by the observer as an appropriate problem to the animal itself. It is just the idea of
anthropomorphism. The external observer who sets the problem and anticipates its a priori
unique solution is an anthropomorphic artifact at best.
Despite that, the role of the external observer cannot be wiped out of the study of animal be-
haviors. Imagine, for instance, an animal put in a maze having no exit. The observer cannot formu-
late a well-defined problem of learning the incom- plete maze to the animal. This observation
suggests that we cannot separate the results of learning from the definition of learning, demon-
strating inseparability of an observer and an ob- ject Mizukami et al., 1999. If an animal behavior
is uniquely encoded by an input-to-output map- ping of a one-to-one, critical examination of the
role of the external observer would be suspended. This situation is quite resembled to the Frame
problem in AI McCarthy and Hayes, 1969. When one tries to construct a robot which can
react to even unknown situations with a one-to- one correspondence scheme, the artifacts originat-
ing in the assumption of external observer would become inevitable. Bickhard and Terveen 1996
deal with the fundamental and issues that are intrinsic to encoding in AI and cognitive science.
According to their view, encodingism restricts learning by limiting it to the combinatoric space
defined by the encoding atoms. It simply means that no new representations are possible, and that
the encoding atoms must successfully anticipate all possible contingencies in their space of combi-
nations. However exactly the very reason for what leaning is functionally intended for. There would
be no need of learning if anticipation succeeds.
To better grasp the notion of animal autonomy or decision-making, it will be required to clarify
the underlying fundamental assumption. Re- cently, the idea of inseparability of an observer
and the object has been attempted in the studies of complex system e.g. Crutchfield, 1994, endo-
physics Rossler, 1987 and internal measurement e.g. Matsuno, 1989; Gunji, 1995; Gunji and Toy-
oda, 1997. In addition, some studies on animal behaviors demonstrated a need of new notions
and concepts, such as behavioral plasticity Gunji, 1996, usage of a tool Kitabayashi et al., 1999,
and duality of an attractive and repulsive force in fish schooling behavior Gunji, 1999. All of these
notions emerged from the inseparability of an observer and the object. In the present study
reported in this article, we experimented with turtles, Chinemys ree6esii.
The animal used in our experiments is com- monly called reeves’ pond turtles Family Emydi-
dae. It
lives in
shallow-freshwater ponds,
marshes, or lakes. Its habitat and ecology are reported, for example, in Ernst and Barbour,
1989 and Parmenter and Avery, 1990. The species belonging to this family is used most often
in
learning experiments
Burghardt, 1977;
Macphail, 1982; Suboski, 1992; Papini and Ishida, 1994.
2. Methods
