Information Technology Reference
In-Depth Information
Answer
2
An alternative hypothesis could also be:
=
s
i
=
1
/
2
s
i
−
1
+
1
10
=
18
/
2
+
1
6
=
10
/
2
+
1
4
=
6
/
2
+
1
?
=
4
/
2
+
1
Answer
3
where
s
i
is the ith value in the series.
A more human answer would be:
=
s
i
=
s
i
−
1
−
1
/
2(
s
i
−
2
−
s
i
−
1
)
6
=
10
−
(18
−
10)
/
2
4
=
6
−
(10
−
6)
/
2
?
=
4
−
(6
−
4)
/
2
=
Answer
3
However, it is the second and third hypotheses that give the correct answer from
the human testing of IQ. It is the last hypothesis, which is human because it uses all
the given data within its structure; the same criteria for cryptic clues in crossword
puzzles. Yet the first answer is not wrong given the problem.
To get the right answer in the right way indicates that the selection of a hypothesis
generator is important, and must depend upon some abstract features of the series;
abstract features such as rising, falling or fluctuating of the numbers in the series.
Mohammed Zakaria tried four different learning strategies, where the program
modified its selection of a hypothesis to fit with a human choice (see Chaps. 11
and 12 for more detail). He compared the results of the strategies with a control (no
learning and no bias). Sometimes the hypothesis chosen would give the right answer,
but for the wrong (inhuman) reason. The answers were counted as correct in these
cases (see Fig.
2.10
). A human quotient (HQ) was introduced that was defined as:
Number of correct Answers with Human Hypotesis
Number of correct Answers
HQ
=
*100
Notice that in general the number of 'correct' answers increases with the HQ.
What we have here is an illustration of
learning
that improves the
acceptability
of
the range of potentially correct solutions. It is an acceptability that goes beyond the
criteria of validation since it involves all possible styles of solution; it is an illustration
of
machine wisdom
.
