Page 160 - Modern Robotics Building Versatile Macines

140 Modern Robotics

danger and protect offspring. Existing robot-controlling computers
are far too feeble to match this massive ultra-optimized perceptual
inheritance.

The retina of the human eye packs together millions of cells that
can detect the edges of objects and react to motion. Making a rough
calculation, Moravec concludes that even the 1,000 MIPS (one bil-
lion calculations per second) capacity of a late 1990s supercomputer
falls far short of the processing occurring in the retina and optic
nerve, let alone the human brain itself, which may perform process-
ing equivalent to about 100 million MIPS (or 100 trillion instruc-
tions per second)! By comparison, Moravec estimates that a 2003
model desktop computer has a processing power equivalent to the
nervous system of an insect or perhaps the brain of a guppy.
Against this formidable processing gap between computer and
brain must be placed Moore’s Law, the well-attested observation (by
pioneer chip-builder Gordon Moore) that the processing power of
the top-of-the-line computer chip roughly doubles every 18 months
to two years. If this trend continues, Moravec believes that comput-
ers (and their associated robots) could reach humanlike processing
capacity by 2040. And because this growth is driven by geometrical
(doubling) functions, humans might be quickly surpassed after that
time.

Robots: The Next Generations

Today’s robots can, at their best, do only a few things well. They
are specialists. Moravec suggests that by around 2020 the first
true “universal” robots may appear. Just as a computer is a uni-
versal machine in that it can perform any kind of calculation for
which it has been given the appropriate instructions, a universal
robot could be given programs enabling it to tidy or clean a house,
wash dishes, mow lawns, take inventory in a warehouse, guard
that warehouse, or even play games with children. Moravec sees
this first generation of universal robots as having about a 10,000
MIPS processing power and “minds” equivalent in complexity to
that of a lizard.