sci.lang
[Top] [All Lists]

Re: Universal grammar

Subject: Re: Universal grammar
From: Mike Wright
Date: Sat, 30 Dec 2006 13:57:26 -0600
Newsgroups: sci.lang
Franz Gnaedinger wrote:
Hans Aberg wrote:
Happy Solstice and Seasons Greetings.

Thank you, I had a very nice time. Yesterday I read
a mind-boggling article and wish you to ask a question.
Hud Lipson, Josh Bongard and Viktor Zykov at Cornell
University built a robot with a central modul and four
bendable extremities, informed it about its elements,
and ordered it to walk. The robot began to calculate,
it carried out sixteen calculations for every move,
then it began to walk, not elegantly, but it moved on,
and when one extremity was removed, it began to
calculate anew, and then moved on in a new way.
The robot was not built for walking, it was just informed
about its elements and given an order to walk. Now for
my question. How can a robot understand such an order?
Can you also give it an order such as: Maintain yourself
... ? Living beings are entitites that maintain themselves.
Are robot crossing the treshold to life? I never hold this
possible.

It's pretty sophisticated, but seems to be wholly algorithmic. See <http://www.videoplayerweb.com/video/a-self-aware-robot-2/1sZgjyQspeQ/>

Also see: < www.videoplayerweb.com/video/a-self-aware-robot/YiSPWxQK38E/">http://www.videoplayerweb.com/video/a-self-aware-robot/YiSPWxQK38E/> and < www.videoplayerweb.com/video/the-self-aware-robot-2/u5yUFL9ReL4/">http://www.videoplayerweb.com/video/the-self-aware-robot-2/u5yUFL9ReL4/>

A working model probably requires some kind of feedback mechanism(s) connected to a level sensor--the equivalent of the semicircular canals in humans. When the sensor indicates that the robot is not level, it calculates which legs need adjustment to restore the sensor state that indicates balance. That would be combined with sensors that provide info on the relative locations of the legs and the body. For a fairly simple explanation, try the "Two legs" paragraph here: < www.channel4.com/science/microsites/R/robots/m_legs.html">http://www.channel4.com/science/microsites/R/robots/m_legs.html>.

After all, there is no mysterious life force that lets humans maintain their balance. There's just a set of gravitational sensors < hyperphysics.phy-astr.gsu.edu/hbase/sound/eari.html#c3">http://hyperphysics.phy-astr.gsu.edu/hbase/sound/eari.html#c3> that provides input to the nervous system.

When I had an inner ear infection, about a decade back, I couldn't keep my balance (and every time I moved, I threw up). Being "alive" wasn't sufficient once the mechanism quit working.

The robot doesn't have to "understand such an order" at a self-conscious level. It just has to respond to input with appropriate output--as it is designed to do. The phrasing in "informed it" and "ordered it" is extremely misleading. What has been done is to design an algorithm that permits the robot to adapt its gait to environmental changes--either internal or external. Note that some trials are required for the robot to come up with a new gait.

What's special about the Cornell robot is that it is more adaptable to extremes than previous designs. However, the basic "motivation" is provided by the designers, and the parameters of that motivation are probably relatively simple (and reflected in the design of the algorithm).

Rather than taking this as an example of a robot becoming somehow "alive", take it as an example of how fairly complex behavior can be achieved in a completely mechanistic fashion. Are you familiar with "artificial life" programs that model apparently complex behavior algorithmically, such as flocking behavior in Craig Reynold's "boids"? See < www.red3d.com/cwr/boids/">http://www.red3d.com/cwr/boids/>.

There is no "threshold to life". "Life" is just a particular set of mechanisms based on DNA and proteins. Granted, it's an extremely complex set, and appears to be pretty well-tuned to the survival requirements of the set of "living" things, but it's still a set of mechanisms. Even the brain is now being understood at this level. If you doubt it, read Eric Kandel's _In Search of Memory_.

Life (as we know it) is based on millions of years of development using all the resources of the earth. I doubt that we'll ever see explicitly-designed robots that have anywhere near the complexity--and generality--of living things. For one thing, the evolution of living forms is an incredibly wasteful process if looked at in terms of goal-oriented, bottom-line-driven corporations.

What's being learned with projects like the Cornell robot seems to be how to apply "selection" to inorganic systems in ways that mimic artificial and natural selection in organic systems.

That'll be two cents, please.

--
Mike Wright
                                www.raccoonbend.com">http://www.raccoonbend.com

<Prev in Thread] Current Thread [Next in Thread>
Privacy Policy