reactive behaviours, which depend strongly of an
external stimulus, or a set or sequence of external
stimuli (McFarland, 1981). Examples of these can be
locomotion patterns. These behaviours (and the ones
which follow) require an action selection process,
whereas reflex behaviours are executed whenever the
triggering stimulus is present. Motivated behaviours do
not only depend on external stimuli (or the absence of
a specific stimulus), but also on internal motivations.
For example, “exploration for food” can be performed
when there is the internal motivation “hunger”. The
previous types of behaviour have been modelled with
behaviour-based systems (BBS) (e.g. Brooks, 1986;
Beer, 1990; Maes, 1990; 1993; Hallam, Halperin and
Hallam, 1994; Gonzalez, 2000; Gershenson, 2001).
Reasoned behaviours are the ones which are
determined by manipulations of abstract concepts or
representations. Preparing yourself for a trip would be
an example. You would like to make plans, for which
you would need to have abstract representations, and
very probably a language (Clark, 1998), and to
manipulate these representations. This manipulation
can be considered as the use of a logic. This level has
been modelled with knowledge-based systems (KBS)
(e.g. Newell and Simon, 1972; Lenat and Feigenbaum,
1992). We could speculate about conscious behaviours,
without entering the debate of the definition
consciousness, just saying that they are behaviours that
are determined by the individual’s consciousness. We
do not believe that there is an “ultimate” level of
behaviour. We could, in theory, always find behaviours
produced by mechanisms more and more complex. But
for now we have enough trying to model behaviours
less complex than reasoned ones. If we cannot clearly
identify further levels, there is no sense in trying to
model them. Figure 1 shows a diagram of the types of
behaviours described above.
Figure 1. Abstraction levels in animal behaviour
(Gershenson, 2001).
We believe that the behaviours in the higher levels
evolved and developed from the behaviours in the
lower levels, since in animals you cannot find higher
levels of behaviour without the lower ones. Thus,
higher levels of behaviour require the lower ones, in a
similar way as children need to develop first lower
stages in order to reach higher ones (Piaget, 1968).
Also the higher types of behaviour in many cases can
be seen as complex variants of the lower ones.
Therefore, it is sensible to attempt to build artificial
cognitive systems exhibiting adaptive behaviour of
higher levels incrementally: in a bottom-up fashion
(Gershenson, 2001:3). This does not mean that we
cannot model any level separately. But the more levels
we consider, the less-incomplete our models will be.
Historically, KBS were used first trying to model
and simulate the intelligence found at the level of
reasoned behaviours in a synthetic way (Steels, 1995;
Verschure, 1998; Castelfranchi, 1998). This means that
we build an artificial system in order to test our model,
instead of contrasting our model directly with
observations on the modelled system. The synthetic
method allows us to contrast our theories with artificial
systems, and in the case of intelligence and mind,
theories are very hard to contrast with the natural
systems. KBS have proven to be acceptable models of
the processes of reasoned behaviours. Not only they
help us understand reasoned behaviours, but are able
to simulate these behaviours themselves. But when
people tried to model the lower levels ofbehaviour, the
artificial systems which were built failed to reproduce
the behaviour observed in natural systems, mainly
animals (Brooks, 1995). This was one of the strong
reasons that motivated the development of BBS on the
first place, but the fact is that BBS have modelled
acceptably animal adaptive behaviour. BBS help us
understand adaptive behaviour (e.g. Webb, 1996;
2001), but also we can build artificial systems which
show this adaptiveness (Maes, 1991).
Observed reality
Adaptive
Behaviourj
Cognitive
Processes
Natural
Exhibitions of
Intelligence
Synthetic theories
KBS
BBS
I Computer
JK. ∖ Simulations of
'"⅛ Intelligence
Figure 2. Simulating exhibitions of intelligence
(Gershenson, 2001).
But if we believe that reasoned behaviours evolved
and developed from lower levels of behaviour, we