Page 169 - Deep Learning

152 Creativity

subgoals. In unfamiliar task environments, collectives, like individuals, have
to take tentative steps, evaluate outcomes and try again in the case of failure.
Search is a necessary consequence of uncertainty and fallibility, and those
factors affect collectives no less than individuals.
What we call subgoaling at the level of the individual reappears as divi-
sion of labor at the group level. The analogy between subgoaling and division
of labor is precise from the point of view of task analysis. As the individual
identifies subgoal X as a piece of the overall task that can be attacked in relative
isolation – as the Wright Brothers did with the problem of lift – so an organiza-
tion can package task X and assign it to a group, section or division. Division
of labor does not by itself add anything qualitatively new to problem solving.
If all subtasks had to be solved in temporal succession, a group would be no
faster than an individual. However, from the point of view of task execution,
there is a major difference: Division of labor enables parallelism, the simulta-
neous pursuit of multiple subgoals. This speeds up search. Edison’s lifetime
would not have sufficed to invent the telephone, the electric lightbulb and the
electric battery if he had not been able to assign different members of his staff
to different tasks, to be pursued simultaneously. Parallelism is an emergent
feature that has no counterpart at the level of the individual, and it has major
consequences for the shape of impasses and insights at the collective level.
An equally fundamental difference between an individual and a collec-
tive is that the latter introduces the necessity to communicate. Individuals have
to interact to form a collective. This feature also lacks any counterpart at the
level of the individual mind. Communication is relatively free of friction in
small teams with few individuals who know each other well or work closely
together. The Wright Brothers and the Watson and Crick collaboration are
good examples. In larger collectives, communication acquires characteristics
that have major effects on the nature of impasses and insights.
Collectives no less than individuals can suffer periods of stagnation. The
fundamental reason is that the same mechanism that produces impasses at
high levels of complexity and duration is also operating in collectives: A single
recalcitrant subgoal, no matter how small in scope, can hold up a collective
enterprise, no matter how large. The scaling mechanism is simple in prin-
42
ciple: If each member of a group is at an impasse with respect to some compo-
nent of the shared task, then the group as a whole is at an impasse. If a subgoal
cannot be reached, its superordinate goals are likewise out of reach. Nothing
in the nature of collectives invalidates this law.
For example, the large research organization that created the first atomic
bomb at the Los Alamos research facility during World War II needed not