Measuring Behavior 96: Looking Back...
Berry Spruijt, Rudolf Magnus Institute, Utrecht University, Utrecht, The Netherlands
Sequential organization of behavior
Over the past decades we have observed an accelerating development of scientific tools for
measuring biological responses. The study of behavior under variable circumstances is special in
that it requires a very flexible and non-invasive approach. Technology first appeared at the level
of data acquisition. Behaviors were no longer handwritten but registered with a computer
keyboard. This was at first an extension of the human "subjective observation" ability. This
ability to estimate what is essential in behavior remains indispensable for defining new
parameters and guiding new developments in behavioral measurements. Next, novel analysis
methods yield new parameters which may so far have escaped our intuitive understanding of
behavior. The sequential organization of behavior, which can be expressed in rigidity or
plasticity, is not always noticed by our eyes. Nice examples of this phenomenon were given by
Dr. M.S. Magnusson (Reykjavik) in his talk on pattern detection and by Dr. M.R. Kruk (Leiden)
in his thought-provoking closing lecture.
Implicit or explicit intentions
What we measure is no longer solely determined by what we can measure, but rather by what we
want to measure. For instance, behavioral physiologists and neuro-ethologists want to apply their
knowledge of the underlying neural substrate on the output of the brain, i.e., behavior. Their goal
is not so much an understanding of the biological function of the behavior itself, but they use
behavior as an indicator, e.g., of the efficacy of a drug. However, we must always be aware of the
danger of incorporating the assumed intentions of a particular behavior in the definitions of the
ethogram and now - even more dangerous - in the algorithms of computer programs
automatically recording behavioral patterns. Behaviors such as 'approach' and 'avoidance'
implicitly presume an intention of the animal, in contrast to spatial displacements and directions
measured by an automated system. Assumed intentions of behavior implied in their definition
become explicit for the designer of automatic behavior recognition software, but may remain
hidden for the user of such a program! Whether that is acceptable, depends on the arguments one
may have to presume an intention. This potential danger in the automation of behavioral
observations was recognized at the workshop, but it should receive more attention.
At the workshop, several exciting methodological and technical innovations were presented. To
mention just a few:
Implications of automation which deserve special attention
- The latest video tracking technology allows prolonged automated observation of identified
animals in a group, and new quantitative analysis methods allow the interactions between all
animals to be scored. As Prof. J.M. Koolhaas (Groningen) outlined, such long-lasting
observations have the (statistical) power to reveal subtle individual differences in the dynamics
of group-housed animals over time which would otherwise remain undiscovered.
- Digital imaging systems can measure much more than just spatial coordinates. New parameters
extracted from video images, such as distance between animals, speed of movement, bouts of
activity, body shape and orientation, position of specific body parts, etc., lead to a more precise
description of the animal's behavior and its position in space and time. Prof. I. Golani (Tel Aviv)
and Prof. A.L. Cools (Nijmegen) use this approach to define the position of animal's "home
base" and to measure "excursions" into an open field.
- There is a strong interest in the synchronization of behavioral and physiological data streams
(heart rate, blood pressure, etc.). Mr. J. Kerl (Göttingen) demonstrated how multiple heart rate
signals and behavioral channels can be measured with a single device.
The validity of parameters measured by automatic systems depends on the scientific and
experimental context of the parameters, and this context must be delineated clearly. The
bottleneck is not so much anymore what you would like to measure, but rather what would you
like to know? There is a danger that we completely rely on computer algorithms and forget what
is actually measured. Since different researchers will rarely want to measure the same behaviors
in exactly the same way, a solution is to have trainable' systems which can be adapted to special
research questions without losing the advantage of objectivity. In fact one would like to be able
to tell the program which behaviors have to be recognized in a particular context. The Eureka
project in which the Rudolf Magnus Institute, Noldus Information Technology and several
European drug companies work together, is aimed at developing such a system.
It is evident that new research tools can only be developed in close collaboration between
behavioral and technical disciplines. The more sophisticated tools for measuring behavior are
desired, the more discussion on the input and output of such tools is required. And we should not
hesitate to borrow ideas from other research areas. Measuring Behavior '96 was intended to
provide a forum for a cross-fertilization between different research fields and the stimulation of
new forms of collaboration. I have the impression that this goal was accomplished. The
workshop organizers hope that this stimulating meeting was the first of a series of regular events!
This article is based on the closing remarks presented at Measuring Behavior '96, International Workshop on Methods and Techniques in Behavioral Research, 16-18 October 1996, Utrecht, The Netherlands