Computers
The computer revolution is upon us. In the next 25 years, all aspects
of computing -- from input to output -- will change radically. The
greatest change will occur in the way computers are put together: there
will be three fundamental modifications to our thoughts on how we
compute. One change will involve neither new hardware nor new
philosophy, but will come about simply through the realization of the
potential of existing technology. Another change will be made through
the development of new hardware, more powerful than anything currently
constructed, but will remain within our philosophical paradigm of
computing. The third change will be a completely new approach to
artificial intelligence, and will require the abandonment of many
current "truths".
The least radical of the three new approaches will make computers
ubiquitous. Microcomputers with the power of current minicomputers can
be made easily and cheaply. The basis will be a distributed
multiprocessor architecture. As soon as computer designers realize that
for many tasks it is already cheaper to use a dedicated microprocessor
than it is to use special purpose hardware and a central processor, we
will see a new breed of machine. Imagine a microcomputer based on a
16-bit central processor with dedicated 8-bit processors for
input-output control, printer control, mass storage control, and
telecommunications. Such a machine would have the full power of the CPU
available for computing, would not be tremendously expensive, and
could, with good software, rival many minicomputers. If this design
philosophy is used, and is coupled with software designed on the
principle of maximizing computer productivity, we will soon see many
very powerful, cheap computers. By 2008, these machines will be
everywhere.
We will also overcome the 11.8" barrier. Supercomputing has almost
reached the point where the speed of light (11.8" per nanosecond) is a
serious design limitation. The next 25 years will see the creation of
new machines able to circumvent the problem through parallel
processing. The gradual development of the technology of vector and
array architectures, and the concurrent development of the techniques
and algorithms needed to program such computers, will provide the basis
for scientific calculation in the future. These machines will supplant
current mainframes, and will provide more power for the tasks computers
already perform. Some problems which are computationally infeasible
(astrophysical and weather models, for example, or any other problem
based on partial differential equations) will become tractable. But the
basic idea of computers as number crunchers will remain.
The advances in artificial intelligence will be based on one simple
fact: that while digital computers can perform millions of calculations
in a flash, the hallmark of intelligence is precisely the opposite. The
difference between a human being and a computer is that the human seeks
to avoid solving problems by repetitive methods. As an example,
consider the game of chess. If chessmen are placed on a chessboard in a
completely random fashion and the layout shown to a novice and a Grand
Master, they will each be able to remember the positions of only a few
pieces. But if the pieces are placed in a position which could arise in
play, the Grand Master will be able to recreate the setup almost
perfectly. A human being understands the game as a "gestalt" rather
than simply as an arrangement of figurines. While chess-playing
computers are gradually getting better, their approach is still the
simple one. In the next 25 years, artificial intelligence researchers
will realize that you don't mimic the brain by doing ever better what
the brain does badly to begin with. The result will be the creation of
new theories. The computers which will be built will be unlike anything
ever imagined. They may be non-deterministic, the will probably not be
based on boolean logic, they may even be bioelectronic. But by 2008,
they will think.