Consciousness

• seCtIon FoUR: eVoLUtIon
  Note that IDA and LIDA are software agents and so, like KA,
  are not permanently tied to any particular physical machine,
  raising the question of just what it is that we think might be
  conscious. Could a complicated mass of software or a virtual
  machine without a single material body be conscious? Could
  it be that self-representations rather than people (or bats, or
  octupuses) are the subjects of experience? If so, there might
  already be conscious entities living in cyberspace and sup-
  ported by multiple machines in different locations. Come to
  think of it, when we refer to ourselves as conscious, are we
  referring to our bodies, our brains, our inner selves, or some-
  thing else entirely? This is another interesting conundrum that
  MMC may throw light on.
  A strongly embodied example is Owen Holland’s CRONOS,
  an anthropomimetic upper-body robot designed to include
  internal models of itself and the world (Holland, 2007; Holland,
  Knight, and Newcombe, 2007). The idea of internal models
  began over sixty years ago when Craik suggested that intel-
  ligent organisms might need small-scale models of external
  reality and of their own possible actions. For a long time Hol-
  land rejected this idea and developed purely behaviour-based
  robotics with no internal models, but he later returned to this
  principle in creating CRONOS and its successor ECCE Robot.
  With human-like elastic muscles and tendons, and a bone-like
  skeleton, CRONOS has a single colour camera for an eye, an
  elongated neck to help it inspect objects, and complex moving
  arms. It builds models of the world around it by moving its eye
  and looking at and interacting with objects, and uses a model
  of its own body and capabilities to plan its possible actions. It
  does not interact with people, however, and has no language
  or emotions.
  How does this relate to consciousness? The robot has a model
  of itself, the Internal Agent Model (IAM), and a model of the
  world, the Internal World Model (IWM), and it uses these to
  act and to track changes in its own body and beyond. These
  internal simulations form the basis for what its creators call ‘functional imagina-
  tion’: the ability to manipulate information that is not directly available to sensors
  (Marques and Holland, 2009). These models may not be completely detailed and
  accurate, but they are the only self and world that the robot knows. ‘Unless it
  somehow had access to the information that it was only a model, the IAM would
  operate as if it were the agent itself ’ (Holland, 2007, p. 101). Critically, this means
  that the IAM would be transparent in the sense used by Metzinger (2003, 2009)
  when describing his phenomenal self-model: the robot depends on a model
  which doesn’t include the fact that it’s a model. This model would in some sense
  describe itself as an embodied agent – rather as we do.

Gauges in the head, Szpindel had called them. But there were other

things in there too. There was a model of the world, and we didn’t

look outward at all; our conscious selves saw only the

simulation in our heads, an interpretation of reality,

endlessly refreshed by input from the senses. What

happens when those senses go dark, but the model –

thrown off-kilter by some trauma or tumor – fails to

refresh? How long do we stare in at that obsolete

rendering, recycling and massaging the same old data

in a desperate, subconscious act of utterly honest denial?

How long before it dawns on us that the world we see no

longer reflects the world we inhabit, that we are blind?

(Peter Watts, Blindsight, 2006, p. 193)

Some other examples of theory-led robotics might include

Giulio Tononi’s (2015) integrated information theory (if you

build integration into a machine in ways that increase Φ, the cor-

responding degree of consciousness should follow) or Michael

Graziano’s attention schema theory (if a system can model its

own attention, it can lay claim to consciousness) (Webb and

Graziano, 2015).

According to quantum theories, none of these implementations

would produce real consciousness because that needs quantum processes. For

example, in Penrose and Hameroff ’s version, consciousness emerges from quan-

tum coherence in the microtubules, so one would need to build a quantum com-

puter that achieved this kind of integration across its system. One might then

conclude that it was really conscious.

None of this avoids the two big problems mentioned at the start of this section.

First, we do not know what consciousness is. Each of these theories (and many

others) says something about what consciousness is or what it emerges from, but

if the appropriate machine were built, critics could still argue that this particular

theory was wrong and therefore the machine was not conscious after all. Second,

we have no test for proving whether a machine is conscious or not, so even if

one of these machines claimed to be conscious, stayed awake all night worrying

about consciousness, and passed the Turing test, we could still not convince scep-

tics that it was really conscious, even though we might have learned a lot from the

machine.

DELUDED MACHINES

There is a completely different way of thinking about X. Perhaps consciousness

is not what it seems to be, and we are in some fundamental way deluded about

the nature of consciousness. According to this view, we may believe we are con-

scious observers, experiencing a continuous stream of contents passing through

our conscious minds, but we are wrong because there is no Cartesian Theatre,

no audience, no ‘actual phenomenology’, and no continuous stream of conscious

experiences (Dennett, 1991; Blackmore, 2002, 2012). We humans certainly seem

to be conscious, and that requires explaining, but the right kind of explanation

‘Engineering will step

from the mere design of

complex artefacts to the

design of subjects’

(Chella and Manzotti, 2007,
p. 11)

PRoFILe 12.3

Owen Holland (b. 1947)

Owen Holland is best known

for his work on machine con-

sciousness and for building

biologically inspired robots,

but he only started robotics as

a hobby in 1988 after work-

ing as a production engineer,

boatbuilder, transport man-

ager, insurance salesman, and chef in a steak bar. He

had a croft in Orkney for eight years where he built his

own house and tended cows, goats, ducks, and chickens,

grew oats, and made hay. Just as eclectically, he has had

academic positions in psychology, electrical engineering,

computer science, and cognitive robotics at universities in

England, Scotland, Germany, Switzerland, and the USA.

He worked on two robot projects at Caltech and helped

set up the robotics lab at the University of the West of

England, Bristol. Holland used the biologically inspired

robot CRONOS to ask whether it could be phenomenally

conscious according to various theories of consciousness.

CRONOS has since developed into the anthropomimetic

ECCE robot which, with its human-like structures, may

lead to human-like cognition. He is now Emeritus Pro-

fessor of Cognitive Robotics in the Sackler Centre for Con-

sciousness Science at the University of Sussex.