(Original Review, 1988-05-30)
Perhaps it is worthwhile clearing up a few
fundamentals here. Specifically, the concept of something complicated being
created as opposed to evolved. Of course, consciousness has evolved and is a
characteristic of the complex arrangement of entities whose properties are
understood by physics. But, going from the basic laws governing the building
blocks to the complex is currently way beyond anything dreamt of in systems
theory, where biological simulation is hovering around the simple swim patterns
of single celled flagellum bacteria. Before attempting to build something we
try and understand it. That is, what aspects of the thing to focus on. What are
the essentials? For example, the planetary orbits are described by their mass
relative orientations and velocities. We don’t bother about their colour, smell
or aesthetic properties. This is a canonical example of the great success of
predictive science. When we come to other things, broadly termed complexes then
we have a completely different situation. This is because of the problem of
defining what we want to predict. To progress we have to get at something
simple. Like a phase transition, a gross change in behaviour. But, what are we
looking for in the human mind? What do we hope to find? This is a perfectly
legitimate area of investigation, but we should be cognisant of the relative
fruitfulness of different approaches. I’m reminded of Nietzsche’s metaphor
where a huntsman sets of in pursuit of this elusive quarry. His dogs get
separated each pursuing a different scent and the hunt has to be abandoned, as
each dog’s lead is equally valid. We end up knowing more and more about less
and less.
I am a little suspicious on the strong
statement about the relation of computation to Physics. I am not aware that computation
is fundamental to physics itself. The statement that given enough time and
space any physical situation can be computed seems overstated. Even at the
mundane levels of chaotic systems Nature 'solves' the physics uniquely and in
real-time - the problems we have arise from the fact that we cannot in
real-time acquire infinitely precise and infinitely covering representations of
the physics and we cannot do the fully parallel calculations required (not even
with more processors - we do not have analytic methods which solve for all
degrees of freedom truly simultaneously). Consider n-body attraction - we have
no analytic solution which emulates Nature - we do have approximations of
course, which are good enough for everyday use.
Of course Penrose has also looked at this from
another angle - specifically whether computation as a formalism can deliver the
measurable/observable behaviour of the human mind. This is still controversial
but for the moment is quite a strong position. More heuristically, Searle's chinese room poses
multiple unresolved issues for the 'mind as computation' camp - precisely
because it is a very good analogue to computation and is self-evidently a
nonsense.
There are plenty of reasons to doubt the
universality of computation and thus our ability to simulate organic and mental
processes beyond the little toy models we have currently.
No objection to the possibility of AGI, nor to
the author's positing of philosophical questions as central, nor to the history
given of AGI attempts so far BUT, Professor Deutsch goes wrong, wrong, wrong
with his very first sentence. The human brain has no capabilities at all. The
human organism however has all the capabilities of thought ascribed to its
brain AND capabilities of movement, action, speech which are in no way optional
accessories. The only intelligence we know of has arisen exclusively in the
context of beings which act and react in and against the world. You can emulate
that in Turing machines all you like; and end up with an simulation.
The computer analogy for the organism, with its
implicit partner of the disembodied brain, has been and remains profoundly
misleading. Following the evidence we have - ourselves - the best prospect for
AGI may be in robotics. There is no reason - other than carbon chauvinism - to
suppose that a machine, with the ability to act and react in the world, cannot
be conscious and intelligent. There is every reason to suppose that a computer
programme cannot.
[2018 EDIT: I suggest looking in to the work of Robert
Rosen rather than Deutsch's; Rosen "revealed" the underlying tacit
assumptions of a state/phase-based physics and showed that biology requires
entailment structures that are not present in this physics. Thus it is that
biological organisms are not a mere subset of current physics, but are
representative of complexities that require physics to be enlarged.]
