On Tue, Sep 21, 2004 at 11:24:03AM -0400, Ben Goertzel wrote: > Messing with robots is a pain! Good robotic hardware is expensive, and
Yes, but they work in realtime, and don't take any resources to render the reality -- of any complexity. There are standard platforms, too, which are cheap. > requires significant effort to maintain, especially for those of us with > limited hardware-engineering expertise. I've done a bit of hands-on So buy a kit. > robotics work in the past, so I've learned this the hard way. Furthermore, > for a team of programmers distributed across various physical locations, > working with physical robots would seem to require a separate robot at each > physical location, or else a robot at a single site with a full-time > robot-maintaner onsite and excellent videoconferencing type facilities so > all the distributed team members can see the robot.... Furthermore, if one > decides one needs to change around ones sensors and actuators for some > reason, it can be very time-consuming and expensive to do so... > > On the other hand, a sim-world is a lot cheaper to build, a lot easier to > maintain, and can be used in a collaborative way by a distributed > development team. And it's easy to change around if one's ideas change.... Despite http://darwin2k.com/ and game engines, it's hard to get physics right. > I do find it interesting to think about how to translate complex sensor-data > into cognitive information. I have a lot of ideas about how to do this > using Novamente. But I have a feeling that one can get an AGI working > without giving it highly rich sensory inputs or highly flexible actuators. Strange, I have the exactly opposite feeling. Making sense from noisy, ambiguous data is what intelligence is all about. Abstract reasoning is a recent invention, and mounted upon that ancient chassis. > However, this is really a separate issue from the sim-world versus robot > issue, because one could implement a physical robot with simple sensors & > actuators, or one could implement a sim-world with very rich sensors and > actuators. You can buy a robot. Writing a decent simworld would take a big team, and budget. It's easier to jack in into an existing online game world. Game AI is something one should be able to raise money for. > I plan to start with a pretty simple sim-world, and add richer simulated > sensors & actuators within it only if necessary. I'm probably only going to > mess with physical robotics once I have > > a) Novamente working reasonably well for controlling an agent in a simple > simulated world > b) funding to pay at least one dedicated robotics person > > Without funding to pay for a dedicated Novamente-sim-world person, I'm > afraid progress on the sim-world project will be very slow, since it will > end up being worked on by a bunch of us in "spare time" not being spend on > other more commercial but less AGI-ish Novamente-based projects. To try to > work on a Novamente-robotics project in this kind of part-time way would be > silly, the rate of progress would effectively be zero. > > -- Ben -- Eugen* Leitl <a href="http://leitl.org";>leitl</a> ______________________________________________________________ ICBM: 48.07078, 11.61144 http://www.leitl.org 8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE http://moleculardevices.org http://nanomachines.net ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
pgpJPAf8IKAeB.pgp
Description: PGP signature
