[ Breaking
Speed Barriers ]
I
sets out to design intelligence machines, but exactly what sort of intelligence will
determine what sort of hardware is required to build
such machines. One could develop AI by means of software--in
which case a general-purpose computer may be adequate. Simple AI programs like Eliza
can be runned into such a computer with the appropriate programming language that compiles
the source code into machine code that the computer understands.
More complex programs that perform extensive inferencing and data-manipulation require
faster, more powerful, and even new kinds of hardware to make up the short-comings that
even programming techniques cannot make up for.
In general, the computing industry strives to produce faster, and faster computing, but
AI research and development demands such speeds. A computer can be programmed to
discern an apple from a pear by correlating the processed-image of the mystery fruit with
its large database, but it would take a long time. Capturing the fruit's image into
pixels and then processing every pixel's intensity value to determine what the pixel
means--part of the fruit or part of the background or how the fruit is oriented in respect
to the computer's camera, and then searching its database requires a huge amount of
data-processing and information shuffling. Any system that takes a long time to run
would render it impractical for real, everyday applications where images are constantly
changing.
[ Top ]
More Faster. According to the computing industry's rule of computer
development called Moore's Law, computer speeds will double every 18 months. In the
technical sense, a computer's speed is usually attributed to the number of calculations it
can make in a second. For now, most general purpose computers can perform millions
of instructions per second(MIPS) while supercomputers
like the Cray-II can perform substantially faster. More commonly, the speed of
general-purpose computers are measured by the CPU's clock-speed in terms of
megahertz(MHz). Currently, one of the largest chip makers, Intel Corporation, are
selling CPUs that run at about 450MHz and higher clock-speeds are still to come.
[ Top ]
Parallelism. Despite the speed in which the CPU can calculate,
most computers essentially work the same way as von Neumann envisioned at the beginning of
the 20th century. Computers are serial machines that have to process every piece of
data one at a time. The demands of AI on data manipulation is so great that serial
computers are fundamentally too slow. A radical solution that is almost akin to how
the human brain works is to link many CPUs together and have them process different parts
of the entire data at the same time. Parallel-processing, as the design is
called--allows a computer to break down an entire data like an image of an apple into
regions so that the individual CPUs will just process the pixels in one section of the
picture. With more processors, more of the information can be broken down even more
which would speed up the entire process. Exactly how many CPUs, how complex they
have to be, and how to break up, distribute, and reassemble the information when it is
processed is still a greater challenge.
Other Considerations. How fast a computer's CPU runs is not the
only factor determining the system's overall speed. How fast the memory can deliver
and store its data to the requests of the CPU, the bandwidth of the system bus, and other
components of a computer all contribute to the overall processes of a computer.
Nevertheless, it is the data-handling of the CPU that determines much of the speed of the
computer. However, the increasing the clock-speed is not the only way to increase
the number of calculations the CPU can make in a second. There are already some
developments taking place that will revolutionize how the computer computes that is
described in "The Future".
