Input systems can be very fast because of the limited source of their
information. They do not incur the computational costs involved in taking
account of background knowledge. But central processes do take back-
ground knowledge into account. However much it may look to me as if
there is a pig in my garden, this is something I am going toWnd hard to
believe, just because it is so surprising in relation to my background beliefs
- for example, since there are no farms nearby it is diYcult to explain how a
pig could get to be there. Input systems may be designed to take the world
at face-value, but central systems need to be at least somewhat dogmatic in
order to avoid jumping straight from appearances to conclusions.
Fodor sometimes expresses this point by suggesting that whereas input
systems have limited informational resources, a person’s central cognitive
processes only operate in a properly rational way if they take account of
everything that the person knows. This seems to us a mistake. At least, it is a
mistake if one takes potentiality for actuality. What we mean by this is that
almost anything a person knowsmightbe relevant toWxing upon a belief or
making an inference. But clearly we cannot repeatedly be carrying out
exhaustive surveys of our prior stock of beliefs. Even if this were part of an
ideally rational, fail-safe procedure for belief acceptance, it clearly is not
something that human beings, with limited processing resources and
limited time, could possibly go in for. As we will be emphasising in chapter
5, it is important to distinguish between abstract ideals of rationality, and
the sort of rationality appropriate to the human condition.
Fodor does have what seems like a better argument for supposing that
central systems are unencapsulated, however, if one buys the idea that
these systems are engaged in a sort of non-demonstrativeWxation of belief
which is analogous to the way in which scientiWc theories are conWrmed.
The inferences involved in belief-Wxation are surely going to be non-
demonstrative: in other words, they are not simply going to follow deduc-
tively valid rules. Inferences which do follow deductively valid rules can be
as blinkeredly encapsulated as you like, because all that is needed to
implement them is a system which advances in a reliable way from a list of
premises to some of the conclusions which can be derived from those
premises. Thus ‘All robecks are thwarg’, ‘Omega-1 is a robeck’, hence
‘Omega-1 is thwarg’ is an inference which can be drawn with absolute
deductive security against a background of no matter what degree of
ignorance on the topicsrobecksandthwargness. Some of our inferential
capacities may rely on a topic-neutral logic module which works this way.
But clearly that cannot be the general story, since it ultimately cannot
explain where we obtain the premises, from which to run demonstrative
inferences.
So how does non-demonstrative inference work? Well, if only we knew.
70 Modularity and nativism