From: hal@finney.org
Date: Fri Oct 06 2000 - 12:02:31 MDT
JR points to:
> Rodney A. Brooks online papers:
> http://www.ai.mit.edu/people/brooks/papers.html
Brooks has had good success with his little robots, and now he's trying
to build a humanoid robot. As was pointed out earlier, a number of AI
labs are moving in this direction.
I can see that there is practical value in making robots that can walk,
carrying things, etc. But I can't help thinking that a significant
part of the motivation here is purely public relations. People are
more interested in a robot which looks human, is able to nod its head
and smile, shake hands, etc. Labs doing this work may get more funding
as a result.
But is this really the right direction for AI research? Recall what
the goal is: approximating human level intelligence. AI is a cognitive
effort which seeks to create a mental architecture. Giving a robot a
face and hands does not directly advance towards that goal.
Researchers try to justify this work by saying that forcing the robot
(really, the researchers) to solve real world problems will require the
development of intelligent algorithms. Some go so far as to say that
the reason AI has failed to make progress is because the evolution of
intelligence requires interaction with the real world in a real body.
I don't buy this because it seems that AI is so primitive that there is
plenty of work that can be done in simulated environments. It's not
like AI creatures are so smart that there is no more challenge there
and they're ready to move up to the real world. The slow progress in
AI is because it's hard, not because of a lack of challenge.
The problem with the current direction is that a great deal of time is
spent on mechanical tasks of layout, construction, lubrication and repair.
This is wasted effort in terms of trying out new cognitive structures.
It seems to me that this can only slow down AI progress in the long term.
We are sacrificing true AI progress in favor of emotionally impressive
demos.
Hal
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:31:26 MST