Computer Classics

From: J. R. Molloy (
Date: Sat Oct 07 2000 - 21:18:32 MDT


Why read cognitive science classics? Part of an answer to the question may be
provided by considering just one of the texts that we selected: the ubiquitously
cited 72 paper by Miller. It came to our attention that contrary to what seems
to be a generally accepted fact about this paper, it is not really about the
capacity of short-term memory. Furthermore, contrary to what is also widely
believed, the empirical evidence in the paper in support of this capacity is
neither strong nor abundant. This is striking considering that the scope of
short-term memory is arguably the most celebrated and most widely disseminated
fact this field has ever produced.
>From this example, there are several things to note. First of all there is what
the researchers claimed to do. Second, what they actually did, and third, the
way in which they have been subsequently portrayed in historical accounts of
cognitive science, and when referred to in contemporary contexts. Clearly, there
are all kinds of discrepancies. There is also merit to reading these original
texts as opposed to only having had contact with them via second-hand sources,
which tend to be sanitised distortions of the original ideas. Distortions which
often exceed those due to brevity and simplification. For instance, as pointed
out in the included paper by Brooks (1991), what was once known to be a
simplified statement, comes to pass as truth for later generations, and
underlying assumptions may be forgotten to be assumptions and become enshrined
as axioms. For these reasons and others, new generations tend to hold the
current position as 'self-evident,' and earlier ones to be so obviously wrong.
'How could anyone believe in that?' Typically, there were good reasons for
believing in just that. The real significance of a new approach may to a great
extent be missed, if the good reasons for believing in earlier ones are

Not so classic:
Trends in Robot Control Architectures

Robot control architectures have changed and become more sophisticated in the
last few years. Early robot control systems attempted to utilize their knowledge
of the world to plan actions to situations. Because of the limited successes of
these systems in the real world, attempts were made to eliminate the
requirements for world knowledge altogether through the use of so-called
Reactive Planning Systems. Such systems demonstrated certain successes
immediately but their lack of sophistication became apparent. Some newer robot
control architectures attempt to combine reactive and planning systems and have
demonstrated sophistication with real-world capabilities.

This archive was generated by hypermail 2b30 : Mon May 28 2001 - 09:50:15 MDT