Whatever area or application domain of AI we work in, when we choose an approach to solving a prob- lem we face, we willy-nilly make a few high-level
methodological choices. One such choice is between developing systems that aim to supplant humans — cognitive
prostheses — and systems that aim to enhance human performance — cognitive orthotics. The distinction between the
two is clear on the example of machine translation (MT).
Although prosthetic systems aim to supplant humans by
independently matching human performance on a task,
most prosthetic systems still have to rely on people to yield
a high-quality final result. Thus, results of Google Translate
must be edited by a person to yield a high-quality translation.
The practice of postediting the results of machine translation
has been employed for over half a century. It is clear that
today’s fully automatic MT systems yield much better raw
translations than systems of yore, thus making the job of a
; The term cognitive systems may
mean different things to different people. This article argues that the core
desiderata of an artificial intelligent
system are properties that make it more
humanlike in its abilities to understand, learn, and explain.