M. Ian Graham, David Kitchin
A hard AI problem can be defined as a problem in artificial intelligence
that has gone unsolved for a long period of time, even under intense
research. In other words, a hard AI problem is a problem which
is solvable by humans but not readily solvable by computers. A
CAPTCHA is a program which generates and grades hard AI problems
in order to automatically verify that it is communicating with
a human and not another program. (www.captcha.net)
A program can pass a CAPTCHA test (thereby thwarting the CAPTCHA)
through one of two methods. The first is solving the relevant
hard AI problem, something which is unlikely to occur. The second
is having access to a "human oracle"--that is, the program must
be able to query a human resource and use the answer in its evaluation.
A human oracle is not only a tool for breaking CAPTCHA tests;
it provides an extremely powerful augmentation to the computational
abilities of any program which may query it. With access to a
human oracle, hard AI problems become solvable. A computer will
suddenly be able to handle issues such as emotion, symbolism,
and general recognition much more easily.
We explore the problem of creating a human oracle system, under
the assumption that paying participants for their time is financially
prohibitive. We propose that an economical human oracle may be
fashioned by embedding a hard AI problem in an online game. Specifically,
we create a human oracle system for general image description
and classification, with associative, symbolic, and emotional
qualification. The game in which the problem is embedded is entertaining
for players, who might not even know their answers are used as
resources. The system receives an image as input and returns an
intelligent description as output.
To our knowledge, this is the first implementation of a practical
and economical human oracle system.