Information Technology Reference
In-Depth Information
years by then and had deep respect for the maniacal depth of his knowledge
in several domains. During one of the breaks in our deliberations, Terence
went outside and climbed a tree. My videographer friend Rachel Strickland
and I approached him for an ill-formed interview. I asked him about the fi rst
thing that came to mind. “What can you tell me about ants?” He replied, “Oh
Brenda, you have no idea .” He then talked nonstop for over two hours about
the 20,000-plus species of ants, dwelling on those he knew best from his jour-
neys in the Amazon and the Hawaiian islands. Ants do amazing work.
The behaviors of ants as they search for food have been modeled in
computer algorithms to fi nd good paths through graphs, which has become
a way to work on more sophisticated problems. The combined behaviors of
individuals in a colony act as a kind of superorganism. With ants and other
creatures like bees, individual behaviors are governed by a fairly small
number of rules that combine to create self-organizing group behaviors that
are more powerful and successful than a single individual is equipped to
produce. For example, ants searching for food leave pheremone trails. The
more ants that follow a particular trail, the stronger the pheremone signal.
But ants have a neat trick. The pheremone trails dissipate over time. That
serves to discourage dead ends and confusing intersections. Ants also use
pheremones to identify their “task groups” and to discover when it is time
to raise a new queen for the colony.
In interaction design, we can learn from ants and other superorganisms
that changing one of the few rules governing behavior can result in huge
changes in the behavior of the colony as a whole. For example, in algo-
rithms that have been developed to simulate ant behavior, the time scale of
“pheremone” evaporation has large effects:
The time scale must not be too large, otherwise suboptimal premature
convergence behavior can occur. But it must not be too short either, oth-
erwise no cooperative behavior can emerge. Cooperative behavior is the
other important concept here: Ant colony algorithms make use of the si-
multaneous exploration of different solutions by a collection of identical
ants. Ants that perform well at a given iteration infl uence the exploration
of ants in future iterations. Because ants explore different solutions, the
resulting pheremone trail is the consequence of different perspectives on
the space of solutions. Even when only the best performing ant is al-
lowed to reinforce its solution, there is a cooperative effect across time
because ants in the next iteration use the pheremone trail to guide their
exploration (Bonabeau, Dorigo, and Theraulaz 1999).
 
Search WWH ::




Custom Search