- seCtIon FoUR: eVoLUtIon
into imagining that a very simple table-lookup program could do the job, when
really ‘no such program could produce the sorts of results that would pass the
Turing test, as advertised’ (Dennett, 1991, p. 439). Complexity does matter – so
even if a hand calculator does not understand what it is doing, a more complex
system, like one that passes the Turing test, could. He suggests that we should
think of understanding as a property that emerges from lots of distributed qua-
si-understandings in a large system (p. 439).
We might go even further and reject Searle’s thought experiment (like the zombie
argument or Mary the colour scientist we considered in Chapter 2) on the grounds
that it instructs us to imagine something impossible. Searle claims that with only
the Chinese symbols and his rule book (or even with the rules memorised and
inside his head), he really could pass the Turing test without understanding a
word of Chinese. But what if he couldn’t? It might turn out that symbol ground-
ing, or learning by interactions with the real world, or something else again, is
necessary for passing the test as well as for ‘really understanding’ a language. In
this case, there are only two options. Either he does not have these necessities,
and his symbol manipulations fail to convince the Chinese people outside, or
he does, and that means he comes to understand Chinese in the process. Either
way, the scenario Searle described in the original thought experiment might be
impossible.
Just as with Mary and zombies, there is no final consensus on what, if anything,
the Chinese Room shows. Some people think it shows nothing. Some people
think it demonstrates that you cannot get semantics from syntax alone, and that
a machine could not be conscious simply by virtue of running the right program.
Some (perhaps a minority) agree with Searle that it demonstrates a fundamental
difference between the real, conscious intentionality that we humans have, and
merely as-if intentionality. In this case, machines could be conscious only if they
had the same causal properties as living human brains, whatever those proper-
ties are.
HOW TO BUILD A CONSCIOUS MACHINE
Many roboticists and computer engineers ignore all the arguments and simply
get on with pursuing their ‘Holy Grail’: ‘the artificial consciousness quest – nothing
less than the design of an artificial subject’ (Chella and Manzotti, 2007, p. 10).
There are two main ways of setting about the task. The first asks how to build a
machine that seems to be conscious; the second asks how to build a machine that
really is conscious (whatever that means).
But some say there is no need for a grand quest, for conscious artificial machines
are all around us already.
THEY’RE ALREADY CONSCIOUS
In 1979, John McCarthy, one of the founders of AI, claimed that machines as
simple as thermostats can be said to have beliefs. John Searle was quick to
challenge him, asking ‘John, what beliefs does your thermostat have?’ Searle
IS THIS MACHINE
CONSCIOUS?
‘My thermostat has
three beliefs – it’s too
hot in here, it’s too cold
in here, and it’s just right
in here.’
(McCarthy, in Searle, 1984,
p. 30)