MIND MAKING
Artificial Intelligence. It's a science that's now old, yet has
always been new. New because the possibilities have so long seemed
close, and yet keep revealing themselves slightly out of our grasp.
Now at least applications relating to image processing seem to be
bursting out of scientific papers into the real world, though there
still seems to be limited penetration into other applications
[granted that I'm not much exposed to the latest technology].
I'm one of many people with an interest in computers who carry
around niggling facination with AI. Not that I really know any
personally, I know barely any people full stop, but I see it pop up
from time to time on personal websites. One lone page describing
some vague personal research into the topic, or maybe just one link
to a scientific paper, book, or magazine article. I think like me
they see some potential there, but its beyond easy grasp.
No doubt part of the problem is the success of conventional
computing. In fact software today is an incredible achievement,
every process explicitly prescribed by an individual programmer,
all meshing together with layers upon layers of complexity. Like no
creation that can be built but in a computer, and relying on the
complete absence of mistakes in order to function. Artificial
Intelligence will present little competition to most of this
conventional software. It only introduces the uncertainties and
unpredictabilitys of the mind, otherwise eliminated by the clean
mechanical processes of a written program.
The pride that some take in thinking that the human mind is truely
greater than that of a computer is misplaced. In fact the computer
is run by the work of countless human minds invested in the most
direct way into the exact and precise fulfillment of our demands
upon it. It is the cumulation of human thought maintained
indefinitely as a model for the perfect solution to any problem
conceived for it.
You can not do what your computer can do. That is why you use it,
and why it is made. Of course the processes themselves can be
replicated, but with much chance for mistake and no comparison in
terms of speed. AI, like our own minds, offers no promise of a
perfect solution optimised for speed and with no opportunity for
mistakes. As with a person, it might not even be possible to
identify the cause of a specific mistake made by an AI. The
limitation overcome by AI is not in fact that of conventional
programming, but of the conventional programmer. Its promise is to
overcome the limit of even combined human minds to determine a
program to accomplish particular tasks.
This is because at the end of the day humans are not very good
programmers. I in fact find programming quite tedious. Beyond the
conception of a generalised process for accomplishing a task, the
need to fit that in to an exact and faultless sequence of
instructions is quite unnatural. Few indeed are able to, or bother
to, create a program that lives up to the ideals of the science -
where there are no bugs, and the maximum potential of the hardware
is exploited. With the amazing complexity of modern software, the
difficulty of this reaches ever further, and our solutions get ever
worse.
I want programming to be like the design of electronic circuits;
engineered to channel the unpredictable randomity of nature towards
a defined and certain consequence, by what means can be
economically implemented.
The aim then for my concept of AI is beyond a tool that happens by
sequential errors to stumble upon a software process deemed
sufficiently reliable by its human master. It is for an AI to be so
concious of the task presented to it as to fulfil the role of the
conventional programmer and write, as they would, a perfect and
optimal program to achieve it. Therefore the two forms of mind,
that creative one of man and the perfect one of the machine, would
be one. A mind designed for our own modern world, in which
precision is of such extreme importance, yet beyond our own grasp.
The human's task then is just to direct the AI towards the outcome,
so that it can by its own learned process create the code to
achieve it. The uncertainty of the AI's own learned processes for
writing the program are what must be directed and controlled, and
that is the new role of the human programmer.
- The Free Thinker