Robot Consciousnessin a way that opens windows onto events that would otherwise remain closed. Again, plausibly
we produce these external models because our own offline (i.e., decoupled from external stim-
uli) internal modeling ability is so limited. At the same time, as external modeling complexity
increases, the more limited becomes our ability to understand the models themselves. And so
we must find ways of compressing what the models do into digestible form. ANNs are one case in
point. They often display uncanny resemblances to human cognition (Bechtel and Abrahamsen
2002), but from the standpoint of understanding cognition their creation is typically only half the
battle. The rest is figuring out how the models work, and to do so researchers invent techniques
such as principal component analysis (the emphasis being on ‘principal’) that help us to tease
out some of the major factors underlying a model’s performance.^6 Thus, even with the aid of
computer models, there can be many details that remain closed to us.
But taken in its extreme form, the cognitive closure thesis involves a dubious assumption—
namely, that humans as we currently exist are the final word in terrestrial intelligence. Is it not
likely that, as Kurzweil (2005) maintains, we are climbing a mountain, but getting stronger and
faster as we go, toward the production of successors to ourselves, whether GMOs, cyborgs, or,
straight to the present point, robotic contrivances. Surely some of these successors will utilize
internal models that are not superficial and gappy like ours, but internal models with a degree of
depth and detail far surpassing that of even our best present-day computer simulations. Whereas
you and I might, in our mind’s eye, see in a general, hazy way how floating plates and mantle
currents can produce mountains and rift valleys, they might see in theirs how vast conglomera-
tions of atoms do this, or how subatomic particles realize life, or how quantum events undergird
general relativity, and yes, even how vast conglomerations of low-level processes, neural or elec-
tronic, conspire to realize qualia.^7
Many phenomena that are closed to us should thus be wide open to our hyper-intelligent
offspring, and qualia may be among them. If so, then we may find that the ‘robot consciousness’
question was on the wrong track from the get-go. Instead of asking whether synthetic contriv-
ances can have qualia, we should have been asking when those contrivances will attain the level
of sophistication needed to explain qualia, or at least our attestations thereof.
To take things a step further, we have just assumed that the problem of qualia is largely quan-
titative and can be solved with much brute force computation, plus, one imagines, significant
theoretical insight. But what if the problem is not just quantitative but also qualitative? What if
qualia can’t be modeled under the constraints of a fundamentally Newtonian physical frame-
work? Indeed, given their poor apparent fit with the mechanical view of things, they might be a
different beast altogether. They might really be quantum in nature, or perhaps property dualism
is correct, or panpsychism. Or maybe qualia reside in a realm of dark matter or energy. Indeed,
it has become ever clearer that these are the dominant modes of existence in this universe, and
some have suggested that the dark realm could be at least as variegated as that of ‘normal’ matter
and energy. It might have its own dark chemistry, dark biology, or even dark cognition. Descartes
could be more right than most give him credit for. Minds might be a kind of halo around cer-
tain machines in the way dark matter is an invisible halo around visible galaxies. Where extreme
ignorance persists, anything seems possible.
But even in this case our robotic successors will have a real shot at understanding and con-
firming the truth. For though our own biologically evolved mental models have been shack-
led through inheritance and learned interactions with organism-level reality, the mathematical
constructs we have devised, and which figure into the implementation of computer models,
are significantly freer. We add new dimensions by adding variables, and we explore unthinkable
non-Euclidean spaces by altering basic axioms. We even explore possible universes by tweaking
fundamental constants and laws and modeling out their creation and evolution on computers.