The
Robots Are Getting Closer - 5
1/7/2003
| Home |
Can
Computers Become Fast Enough and Cheap Enough Soon Enough?
Dr. Marovec's forecast of computers that become smarter than
we are between 2030 and 2050 is based upon the assumption that by the year 2030,
we will be able to build low-cost computers that are 333,000 times as powerful
as present-day 3 GHz (Gigahertz) computers. That may not happen
on that schedule.
Some Key Points:
IBM/Sony/Toshiba's 1,000,000,000,000 floating-point-operations-per-second (one teraflops) Cell microprocessor chip is real, and is scheduled for marketing as early as the end of next year. It is designed to be networked with other Cell chips, and may greatly accelerate the timetable for robotics. (It's also apparently intended for the Sony Playstation 3... a supercomputer in a Playstation!)
By 2010, clock speeds should increase by a factor of 5, and chip transistor counts might rise by a factor of 5 to 8, conceivably leading to chip speeds of 25 to 40 teraflops. Such a development, if it occurred, could jump-start advanced robotics much sooner than is presently anticipated.
Other computing components such as SDRAM and disks aren't on accelerated development schedules and may constitute a bottleneck to near-term robotics development. However, IBM/Sony/Toshiba must have some plans to provide memory and bandwidth that can support "The Cell".
My private suspicion is that human-class AI could occur with computer resources somewhat below those currently said to be required. If so, we might begin the development of human-level robots within the next ten years.
There seems to be general agreement that a Moore's-Law rate of improvement can be continued at least through 2010, and probably through 2012. Consequently, whatever can be achieved with a 2012 technology level is of particular interest because we can be fairly sure of reaching that level of technology on schedule.
Computer
technology technology will be a pacing item in robotics
realizations.
Robotics May Arrive Much Sooner Than I Thought
What's particularly exciting is what will be available within
the next few years... soon enough to begin planning and setting up the
laboratory environment for higher-level artificial intelligence.
Given cheap teraflops chips by 2005, we might see 10-teraflops chips by 2010,
and 50 teraflops chips by 2015. As indicated at the bottom of this page, RAM and
disk memory may be sufficiently low-priced that within 10 years (2013), a
private individual might be able to afford adequate computing resources (at a
total cost of the order of $6,000) to conduct his/her own 2025-level research
program into robotics. It may mean
that ten years from now, robots will be farther along than has been projected.
It's also clear that the big money is going to be in
industrial applications in which human "service representatives" can
be replaced with artificially intelligent computer programs, as described below.
