Reading What Machines “Think” - Brain Informatics - page 160

We have then a wonderful opportunity. We can think to observe electronic
computers with techniques similar to brainimaging. We have good reasons for
doing this as we can address these two questions: (1) how far we can go with
neuroimaging in understanding human mind? ( foundational perspective ); (2) can
we understand what computers “ think ”? ( applicative perspective ).
This is a fascinating research program and in this introduction we gave only a
taste of it. We will better describe our vision in Sec. 2. We describe the parallelism
between brains and machines that gives the possibility of applying “ brainimag-
ing ” techniques for electronic computers. We better introduce the two reasons
motivating this novel view: a foundational perspective and an applicative per-
spective . The rest of the paper is organized as follows. In Sec. 3, we report on
the studies that are the background for this research. In Sec. 4, we describe the
approximated model for studying the parallelism between brains and machines.
In Sec. 5, we report on the experiments for testing the approximated model. We
here define a test set that can be used for further experiments. Finally, in Sec.
6, we draw some conclusions on this experience and we plan the future work.

2TheV on

Electronic computers nowadays are extremely complex information processing
systems. In some sense, these machines are performing “ cognitive processes ”. As
previously happened in cognitive science and in cognitive psychology studies,
we can imagine the parallelism between computers and minds in the field of
neuroimaging . Computers as well as brains are the physical objects performing
“ cognitive processes ”. We hereafter call them “ cognitive physical objects ”. As
shown in Fig. 1, cognitive tasks activate the “ cognitive physical objects ”. In both
cases, it is possible to observe the activation of these cognitive objects by taking
activation images. We can take these activation images by observing different
physical phenomena, e.g., electric or magnetic. The parallelism is now complete.
On the brain side, we have the brain as the observed cognitive physical object ,a
real cognitive task, and classical brainimaging techniques, i.e., fMRI, as the way
of observing the brain activation. On the electronic computer side, we have the
computer as the observed cognitive physical object , a program as the cognitive
task , and images of the electrical activation of micro-chips as a way of observing
the computer activation.
The parallelism we made between brains and computers in the field of neu-
roimaging opens two possible very interesting research perspectives:

- foundational perspective : how far we can go with neuroimaging in under-

standing human mind?

- applicative perspective : can we understand what computers “ think ”?

Both research questions are extremely fascinating.
The foundational perspective is extremely important. The aim of some studies
in neuroimaging [6,7] is to determine the correlation between high-level cogni-
tive processes and neuroimages. The idea is that processing different conceptual