Consciousness

Chapter

Twelve

The evolution of machines

representation of where elements of the world are
that allows attention to be directed appropriately.
As Aleksander and Morton point out, there are
known to be many cells doing this kind of job in
the human brain. Importantly, KA also includes
depiction of the self in its world. Aleksander con-
cludes that a robot might be said to be conscious
if equipped with KA or in some way manages to
model both itself and the world. He adds that on
his model higher-order thought theories turn out
to be about our ability to attend to different parts
of the architecture that allows this and translate
their activity into language.

Among others who begin from basic principles is
British physicist John Taylor, whose attention-based
CODAM model takes principles of human attention
as a starting point for building a conscious machine
with a sense of self.

A different approach is to start from existing theories of consciousness and build
machines that implement them. For example, according to global workspace the-
ories, the contents of consciousness are whatever is being processed in the GW.
The GW is itself a large network of interconnected neurons, and its contents are
conscious by virtue of the fact that they are made globally available to the rest of
the system, which is unconscious. On these theories, ‘X’ is global availability. So,
presumably, a machine should be conscious if it is designed with a GW whose
contents are made available to the rest of its system.

American mathematician Stan Franklin (2003) built a software agent called
IDA. This ‘Intelligent Distribution Agent’ was developed for the US Navy to help
solve the problem of assigning thousands of sailors to different jobs. To do
this, she has to communicate with the sailors by email in natural language, as
well as satisfying numerous Navy policies and job requirements. IDA is built
on the basis of GW architecture, with coalitions of unconscious processes find-
ing their way into a global workspace from where messages are broadcast to
recruit other processors to help solve the current problem. Franklin describes
IDA as being functionally conscious in the sense that she implements much of
GWT, but not phenomenally conscious or self-conscious, although he argues
that building in a simple kind of self based on Damasio’s ideas of the proto-self
would be quite feasible.

IDA has since been developed into LIDA (Learning IDA), who is capable of per-
ceptual, episodic, and procedural learning (Franklin and Patterson, 2006). Baars
and Franklin argue that the functions of consciousness are produced by adap-
tive, biological algorithms and that ‘machine consciousness may be produced
by similar adaptive algorithms running on the machine’ (2009, p. 23). Since LIDA
implements much of the functionality of GWT, they conclude that she may be
‘functionally conscious’. They also suggest that she could one day be made phe-
nomenally consciousness by adding mechanisms that produce perceptual stabil-
ity or that implement various notions of self in a LIDA-controlled robot (Baars and
Franklin, 2009).

Fly on the left in the world

Muscular

signals:

eyeposition Similar

foveal images

Depiction

of the way that

things areinthe world

Fly on the right in the world

Eye Same

eye

FIGURE 12.12 • An example of a depictive

process which involves muscle

action. A depiction arises in

an area where the foveal

image is ‘positioned’ by the

muscular signals that indicate

eye position (after Aleksander,

2005, p. 39).