Measuring Intelligence
Ron Cottam, Willy Ranson & Roger Vounckx
Abstract
The concept of ‘intelligence’ has a long and varied history. Most recently it
has been widely associated with the adaptive capability of a unified ‘system’ to
deal with unexpected contexts. Unfortunately, a problem appears when we
carefully examine the nature of a ‘system’, because it is clear that the
unification of an artificial information-processing assembly into a ‘system’ is
always performed through human intervention. A high-level ‘system’ always
contains life, either through its own living nature or through the inclusion of
human cognitive capabilities.
We suggest that a good metric for intelligence may be derived from the
relationship between a quasi-autonomous information-processing assembly’s
capabilities in dealing with unexpected contexts ‘on its own’ and the ‘help’ we
must provide for it to succeed. To achieve this, we propose formulating the
operation of a quasi-independent entity as a bidirectional communication
channel. The entity must transmit to its human ‘assistant’ sufficient
information about its current context (Ic) to enable the human to return
sufficient help (Ih) for suitable contextual reaction.
We can define an ‘Apparent Intelligence’ for the entity as the simplest
relationship between the two ‘players’ involved:
AI=Ic/Ih
This provides reasonable values for the Apparent Intelligence AI, ranging from
zero (i.e. stupidity - if the human assistant has to do all the work) to
infinity (i.e. genius - if human intervention is never required). The values of
Ic and Ih may be derived from parametric descriptions of the information flows
in the communication channel.
______________________________________________
______________________________________________