The Evolution
21
have replaced the whole-to-whole links of "eggs," a small modification
that makes necessary a new mathematical analysis.
It is clear from the above "blade" profile that if memory perceived
only the presence of some sort of "blade," but did not know what kind
it was, "breadknife" would, by a factor of 12.5, be a better guess
than "scalpel." The key question is, however, what memory would do in
disambiguating "blade" if the concept "hospital," with all of the
statistical information it contains, were present as well?
To answer this, consider that "blade" indicates that "breadknife"
(probability .5) is 12.5 times as likely as "scalpel" (probability
.04). Accordingly, if the addition of "hospital" increased the proba-
bility of "scalpel" by a factor greater than 12.5, then "scalpel"
would become more likely than "breadknife."
The probability of "scalpel" given the presence of neither "hospital"
nor "blade" is P(blade) x P(scalpel∣blade), which is (.0001)(.04), or
.000004. According to the "hospital" profile the probability of "scalpel"
jumps to .01 with the occurrence of "hospital," a 2500-fold increase.
It follows that the 12.5:1 edge that "breadknife" has over "scalpel"
in the absence of "hospital" becomes a 1:200 edge in favor of "scalpel"
when "hospital" occurs.
All of this can be explained in a way that is perhaps somewhat
easier to understand. "Breadknife" occurs with a probability of P(blade)
x P(breadknifeI blade), which is (.0001)(.5), or .00005; "scalpel"
occurs with a probability of P(blade) x P(scalpe11blade), which is
(.0001)(.04), or .000004. This means that given the conditional occurrence of