Questioning Turing test
View/ Open
Date
26/07/2020Author
Damassino, Nicola Michele
Metadata
Abstract
The Turing Test (TT) is an experimental paradigm to test for
intelligence, where an entity’s intelligence is inferred from its ability,
during a text-based conversation, to be recognized as a human by the
human judge. The advantage of this paradigm is that it encourages
alternative versions of the test to be designed; and it can include any
field of human endeavour. However, it has two major problems: (i) it
can be passed by an entity that produces uncooperative but human-like
responses (Artificial Stupidity); and (ii) it is not sensitive to how the
entity produces the conversation (Blockhead).
In light of these two problems, I propose a new version of the TT, the
Questioning Turing Test (QTT). In the QTT, the task of the entity is not
to hold a conversation, but to accomplish an enquiry with as few
human-like questions as possible. The job of the human judge is to
provide the answers and, like in the TT, to decide whether the entity is
human or machine.
The QTT has the advantage of parametrising the entity along two
further dimensions in addition to ‘human-likeness’: ‘correctness’,
evaluating if the entity accomplishes the enquiry; and ‘strategicness’,
evaluating how well the entity carries out the enquiry, in terms of the
number of questions asked – the fewer, the better. Moreover, in the
experimental design of the QTT, the test is not the enquiry per se, but
rather the comparison between the performances of humans and
machines. The results gained from the QTT show that its experimental
design minimises false positives and negatives; and avoids both
Artificial Stupidity and Blockhead.