REPORTS ON AI FROM
The Secret Agent Man
DON BARKER
BotSpot's man in the field
Toys Reveal Clues to the Future of AI
By Don Barker
Life as a covert operative is not always full of danger and intrigue (even if I was nearly canceled by rogue mailbot on my last
mission). Once in a while, I take time off to ponder the deeper meaning of life and try to figure out where all this intelligent agent
and bot technology is heading. The answer to this question may well lie in the strangest of all
places, a toy I recently bought for myself...err...I mean for my daughter.
The Danish company Lego has long fascinated child and propeller-head, alike, with snap-together
block sets. Last fall, Lego introduced the MindStorms' Robotics Invention System, which includes
an onboard computer, two motors, lots of neat sensors (pressure, light, etc.), and a relatively
simple programming language called RCX. Programming, which is actually done on your PC by
snapping together colorful block-like icons, is downloaded to the MindStorms' onboard computer
(a bright yellow module, slightly smaller than a paperback novel) via a serial port infrared
transmitter. (For more information visit the Lego MindStorms Web site).
My daughter and I found it quite easy (and fun) to build and program a fairly sophisticated robot,
capable of R2D2-like movements (--the popular bullet-shaped droid in Star Wars). Our next project
is a agile Lego robotic arm called Grabber, which can reach,
bend, and grasp like a human hand. (Click here to see an illustration of Grabber.) Although Grabber looks quite advanced, the easy-to-learn and
easy-to-use nature of this robotics kit makes such a project much simpler than it first appears.
This design edge can readily be attributed to the fact that MindStorms was jointly developed,
over a 13 year period, with the brilliant minds at MIT. So, what do Lego kits have to do with
intelligent agents? I'm glad you asked.
When you think about it, an intelligent agent is really just a "disembodied" robot. Pattie Maes,
the noted Belgian scientist who programmed Rodney Brooks' famous robot Attila to teach itself to
walk, realized this fact when she choose to leave the AI Lab at MIT in 1991 to join their Media
Lab and work on intelligent agents. Intelligent agents offered her a way to push ahead in
artificial intelligence without the then messy and difficult mechanical and electrical
requirements of robotics. Maes went on to found Firefly, a company that specialized in creating
intelligent agents capable of learning users' preferences (Firefly was purchased by
Microsoft Corporation).
However, with the passage of time, robotics systems have become less daunting to construct and
program, as demonstrated by the Lego's Robotics Invention System. What's more, Lego just
announced at the 1999 International Toy Fair, a digital video camera and remote control (due out
in the fall. I...err...my daughter ... can't wait). This duo will enable users to receive a video feed
from a MindStorms robot on their PC, giving them a "robot-eye-view" of the terrain faced by their
creation, and the ability to transmit appropriate maneuvering instructions to circumvent any visible
obstacles. Thus, these add-ons will further reduce the complexity of robotics by letting users
substitute their own intelligence for inflexible onboard programming.
The downsizing and cost-cutting efforts at NASA have also provided powerful incentives to develop
less costly and simpler robotic systems. The 1997 Mars Pathfinder Mission allowed a driver on
earth to tele-operate the Mars robotic rover, Sojourner, by using battery powered goggles and a
unique joystick called a Spaceball. The operator was able to see a three-dimensional
representation of the surface of Mars taken from the lander's camera and use the Spaceball to
plot locations for the rover to visit as you can see in
this illustration of the Mars Rover Control Workstation. Once the locations were uploaded to the rover
(typically, once a day), the onboard software provided the actual intelligence required to
navigate the robotic system from one area to another on the uneven Martian landscape.
Back here on earth, lobster-like robots are being dropped from low-flying aircraft into the
shallows of the ocean to hunt down mines for the U.S. Navy. Joseph Ayers, a biologist,
neuroscientist, and professor at Northeastern University is designing "biomimetic" robots to
mimic the natural movements of lobsters so his creations can move swiftly and deftly in the
water. These tele-operated robots will not only be able to detect enemy mines but also perform
tasks like collecting marine-science data and patrol for pollution. Eventually, Ayers' robotic
lobsters will carry out basic activities on their own and use sonar to alert human controllers
when some action should be taken (for further details on these biomimetic devices, visit
Biomimetic Underwater Robot Program).
These examples of supervised autonomy for robots may very well be pointing to a future where
intelligent agents overcome the current limitations of artificial intelligence by forming a
symbiotic relationship with human operators. Instead of looking at artificial intelligence as
a "all-or-nothing" proposition, telepresence combines the natural brain-power of a human with
varying degrees of intelligence in a remotely operated robot. According to David Mitchell, a
member of the Citizen's Advisory Council on National Space Policy, telepresence makes possible a
sliding spectrum from human to artificial intelligence that will fuel huge leaps forward in
agent/robotic development.
Mitchell is so confident about the potential of telepresence that he has formulated his own
"law." The famous science fiction (and former Byte columnists) Jerry Pournelle's now classic
law "Everyone should have a computer" inspired Mitchell to develop a corollary, "Everyone should
have an Android Agent," which he humbly calls Mitchell's Law. In his new Web column,
Personal Telepresence Journal, Mitchell argues that although right now "...the idea of
everyone having a remote-controlled humanoid robot or Android Agent borders on the impossible,
in a few years it will merely be economically infeasible. Eventually, Hertz and Avis will be
renting telebodies for business teletravelers. Teletravelers will make 10 meetings worldwide
in a day that might take a year in person."
While this may sound a lot more like science fiction than science, it should be noted that David
Mitchell credits include the wildly popular Lunar
Teleoperations Model I click for illustration (LTM1). LTM1, started in 1993, became the longest running
teleoperations exhibit of its kind, finally closing in 1997. The exhibit was then loaned to
SpacePlex, where a slightly updated version ran from February until November of 1998, under the
name LTM2. So this, along with Mitchell's extensive consulting experience with NASA/JPL-related
companies, indicates that he is the kind of "nuts-and-bolts" engineer that should be taken quite
seriously when making predictions about the impact of telepresence.
The National Aeronautics and Space Administration is certainly taking the concept of telepresence
seriously in their upcoming Space Technology 4/Champollion Mission, where they plan to rendezvous with
the comet Tempel 1 and then, after surveying its nucleus, send a lander to the surface for
further studies, as illustrated in an artists vision of the Orbiter
and Lander rendezvousing with the comet. As you can imagine, the challenges
are immense and NASA is counting on the fusion of human and artificial intelligence to pull it off.
(For more information, see the NASA web pages for Space
Technology 4 Champollion). In the light of this daring NASA mission, Android Agents seem a
lot less fantastic. Will telepresence make it possible for the intelligent agents that serve on
our behalf in cyberspace to evolve and move into the real world--a sort of Alice in Wonderland
parallel of into the looking glass and back out again? Time will tell, but I for one would sure
appreciate the ability to send android agents on dangerous operations in my place.
This is BotSpot Agent 007 signing off. You can reach me at don@donbarker.com
or visit me at http://www.donbarker.com. Be
careful what you say or you'll give yourself away.
Don Barker is the senior editor of PCAI Magazine.
|
