The Commonsense Architecture of Group Intelligence
Edwin Hutchins, Cognition_in_the_Wild
Hearings at European Parliament
It is often
assumed that the best way to improve the performance of a group is to improve
the communication among the members of the group, or, conversely, that what is
lacking in groups is communication.
Our
simulations provide us -with a means to answer this question. They indicate that more communication
is not always in principle better than less. Under some conditions, increasing
the richness of communication may result in undesirable properties at the group
level.
If the
community is restarted again and again, with the persuasiveness increased each
time, the velocity of the networks in conceptual space increases even more. They hurry in groups to the available
interpretations (some to one, some to the other ), and once there, they respond
only a little to additional evidence from the environment.
It is clear
that a megamind such as that described by Asimov would be more prone to
confirmation bias than any individual mind. It might be a mind that would rush
into interpretations and that, once it had lodged in an interpretation, would
manifest an absolutely incorrigible confirmation bias. Is it possible for communication to ever be
rich enough in a real human community to lead to this sort of group pathology? Perhaps.
Buckhout
(1982) asked groups to produce composite descriptions of a suspect in a crime
that all had witnessed
and reported that " the group descriptions were more complete than the
individual reports but gave rise to significantly more errors of commission :
an assortment of incorrect and stereotyped details."
Producing a Diversity of Interpretations
The problem
with confirmation bias is that it prevents an organism from exploring a wider
range of possible interpretations. Although
the first interpretation encountered may well be the best, a search of the
interpretation space may reveal another one that better fits the available
evidence. How can this search be
accomplished ? I have already shown that
in the absence of communication the interpretations formed by the individual
networks - as each exhibits its own confirmation bias- depend on the three
parameters that characterize the individual networks : the underlying
constraint
structure,
the access to environmental evidence, and the initial pattern of activation. If
a community is composed of individuals that differ from one another in terms of
any of these parameters, then various members of the community are likely to
arrive at different interpretations. Thus,
diversity of interpretations is fairly easy to produce as long as the
communication among the members of the community is not too rich.
