Minds and Computers : An Introduction to the Philosophy of Artificial Intelligence

algorithms. We will call an algorithm a learning algorithm if, as a result
of employing it, the system is conferred with greater capacities.
There are at least two ways in which a mind can be conferred with
greater capacities – it can gain new information or it can attain, or
improve, skills or abilities. In the former case, the mind is learning that
things are the case. In the latter, the mind is learning how things are
done.
In the system [MIND] learning that things are the case corresponds
with storing new content – i.e. storing new values in registers (where
these values code content). Learning how things are done corresponds
with storing new algorithms or with optimising existing algorithms.
Now, given our understanding of [MIND] as a kind of [OS] which is
in continual iterated operation and which is expandable – i.e. has algo-
rithms which store content and which generate or optimise further
algorithms, both of which confer on the system greater capacities – we

105

---

can help ourselves to an explanation available to the computationalist
of variation among [MIND]s with respect to capacities.
Given that, as already established, any two [MIND]s are highly
likely to be in distinct states, and given that, depending on the amount
of time it has been operating for and the nature of the inputs it has
received, a [MIND] will contain more or less stored content and
greater or fewer available stored algorithms (optimised better or
worse), it should come as no surprise that any two [MIND]s will vary
significantly with respect to capacities.
This explanation responds to one of the two challenges presented
to the computationalist by the learning objection – it explains how
two isomorphic formal systems can have di
fferent capacities and,
hence, how minds can be held to be isomorphisms of [MIND] despite
immense variation among minds with respect to what they can do.
The challenge remains, however, for the computationalist to specify
algorithms which govern learning. This is a challenge to which we will
return at various points in the following chapters, particularly in
Chapter 13 when we discuss automated reasoning systems, and in
Chapter 19 when we examine learning in artificial neural networks. We
will also be investigating the way in which humans learn languages in
Chapter 16 and considering evidence that this learning is rule governed.
We have now considered two objections one might mount against
computationalism. These were essentially stronger and weaker ver-
sions of the same objection – minds vary. In both cases, we have seen
how a computationalist might reasonably respond. Let’s consider one
further objection against the theory that minds are computational
devices.
10.5 CREATIVITY
Another standard prima facie objection appeals to the human creative
capacity, as follows. The operations of formal systems are entirely
mechanical but minds are creative. Minds create great works of art,
music, architecture and literature, and have an enormous capacity to
innovate. This characteristic creativity of human minds seems to be
compelling evidence against computationalism which seeks to account
for mentality in terms of purely mechanical operations.
It is certainly the case that it seems that nothing could be further
from an algorithmic process than painting an artwork or composing
an orchestral symphony. As we saw in Chapter 2 however, the way
things seem is no reliable indicator of the way things are.
106
  

---

The challenge here for the computationalist is to explain how the
mental functions we cite as paradigmatically ‘creative’ can be algo-
rithmically delivered, contra-intuition. For that explanation, we need
an understanding of this notion of creativity.
The opponent of computationalism might endorse a definition of
creativity along the lines of: an activity is creative if its result is the pro-
duction of a work (an artwork, composition, etc.) which could not have
been produced by simply following a rule-governed procedure.
Although this definition is somewhat intuitive – after all we’re all
fairly certain da Vinci wasn’t painting by numbers when he painted
the Mona Lisa – it begs the question rather straightforwardly against
the computationalist. Whether or not creativity can be accounted for
algorithmically is precisely what is at issue.
So what is it about creativity such that our initial intuition is to con-
trast creative behaviour with rule-governed behaviour?
Well, firstly, not everyone is equally creative. People have di
fferent
capacities for engaging in creative enterprises. We have already seen,
though, how a computationalist can account for variation with
respect to capacities so this is not su
fficient as an objection, but it
points us in the right direction.
It seems that what disposes us initially against the notion of cre-
ative behaviour being rule-governed is an intuition that were it rule-
governed, it would be more readily teachable. It is characteristic of
those we laud as creative masters – artists, artisans, composers, etc. –
that there is something mysterious to others about their talent.
Further, it seems that when it comes to creative endeavour, one either
‘has it’ or not. Certainly one can learn various techniques and
methods for working with materials to generate certain e
ffects;
however, it is not clear how one could learn to ‘be creative’ per se.
This does not, though, speak against the possibility that this kind
of behaviour is indeed underwritten by computational processes.
Certainly the opponent of computationalism may well demand an
account of how such behaviour could be computationally delivered;
however, there are certain responses available.
For instance, the computationalist might tell a story about certain
algorithms requiring certain computational resources for their imple-
mentation, such that variation with respect to the ability to acquire
certain algorithms is to be explained in terms of variation in the sub-
strate (brains) in which [MIND]s are realised.
In any case, the computationalist is well within their rights to
dissent from answering the question – how is creative behaviour com-
putationally delivered? – until this notion of ‘creative behaviour’ is

107

---

more rigorously specified. In the absence of an independently plausi-
ble account of creativity which speaks directly against the possibility
of such behaviour being rule-governed, the computationalist is no
worse o
ff in this respect than other theorists.
We have now considered a number of preliminary objections against
computationalism and have seen how, in each case, the objection fails.
We have yet to consider more sophisticated arguments against the
theory. We shall hold o
ff on these until after we have seen some arti-
ficial intelligence applications which will provide context for phil-
osophical objections.
To recapitulate, we have discussed some common misunderstand-
ings of computationalism and some bad arguments against the
theory which trade on these misunderstandings. We have advanced a