Page 170 - Deep Learning

Creative Insight Writ Large 153

only to figure out how to enrich the uranium but also how to detonate the
43
bomb. How could they bring subcritical pieces of enriched uranium together
into a single, critical mass in an instant? The group of researchers who worked
directly on this problem had the idea of arranging subcritical pieces of ura-
nium around a hollow core, surrounded by conventional explosives. When the
latter were set off, the hollow sphere imploded, bringing the uranium pieces
into nearly instantaneous contact. Because the entire project rested on being
able to detonate the bomb in a controlled manner, this solution, which was but
one component of a vast enterprise, was essential. Without a reliable method
of detonation, there could have been no atomic bomb.
The role of prior knowledge in causing impasses also scales, in principle,
to collectives. If each member of a collective adheres to a particular constrain-
ing presupposition, then the possibility space searched by the collective will be
constrained accordingly. For example, the assumptions that the Earth is in the
center of the solar system and that the heavenly bodies move in perfect circles
held up progress in astronomy for almost two millennia. 44
It might seem as if this situation ought to be common. If every group
member has the same knowledge as every one else, organized in the same way,
there is no reason to expect one person to say something that causes the others
to see the shared problem differently. But if the members of the group or orga-
nization approach the problem from diverse backgrounds, communication
should preempt impasses at the level of the collective. Even if one team mem-
ber begins by searching the wrong solution space, some other member with
qualitatively different prior knowledge might nevertheless happen to activate
the relevant knowledge and hence work in a more useful search space. The
implication is that a group whose members have widely varied backgrounds
should, on average, be more creative than one in which all members represent
the shared problem in the same way, not because diversity promotes idea gen-
eration per se, but because such a group is less likely to encounter impasses.
This is indeed a commonly claimed advantage for collaborations.
However, research on group problem solving has failed to strongly support
this prediction. Problem-solving groups sometimes perform as well as their
most creative member, but it is rare that a group works better than its most cre-
ative member. What happens to the supposed advantage of bringing multiple
45
perspectives to bear? The phenomenon that the expected advantage of collabo-
ration is seldom realized in practice is known as process loss. There are multiple
sources of process loss lurking in group dynamics, most of them side effects
of the need to communicate. In a particularly elegant analysis, social psychol-
ogist James R. Larson has shown in a series of decisive studies that groups