> Even if an autopilot *has* been programmed to respond to all previous > emergencies, vehicles will still find new ways to go wrong that require > genuine *creativity* to handle. There is no possible way to program *every* > possible emergency.
First, I hope you read my post about Monti Carlo tree searches. You can find solutions to problems not seen before. Second, you are looking at it incorrectly. The goal is not to program it to handle emergencies but to program it to maintain a certain flight state withing given parameters or as closely and safely as possible. The program is given a set of sensors to monitor the flight state (momentum, speed, position, orientation etc), the plane or space ships state (, material stresses, engines, flap positions etc) and the state of the surroundings (air, ground, other objects). It is then given a bunch of things it can change, like flaps, engines etc. It also has an internal model of all this that it can test new inputs on it, hoping for the wanted output using a Monti Carlo system or something newer and better. The next goal is for it to maintain its goals as best as possible using the tools it has, a feedback system. Systems like this can also learn from trying things out and learning to predict outcomes and by watching others do this. This is best done in the design and prototype phases though but could be done throughout the products like given sufficient safeties. This is really a lot like how human brains work and the only thing stopping the AIs from doing it is the speed of the hardware and software which is coming along quickly. Have you seen big dog? http://www.youtube.com/watch?v=W1czBcnX1Ww&feature=channel_page How about this helo that learns from humans? http://www.youtube.com/watch?v=M-QUkgk3HyE&feature=channel_page Then there are the videos about the DARPA robot car races. Remember that current computers are 2 times better and half as expensive every 14 months. These vids are mostly at least a year old. We are talking about future SF AIs! . . . > obGURPS: Does the Routine Vehicle Operation program include things like "if > both wings are blown off and the fuel is on fire, and the megaton nuke > missile armed but failed to release"? :-) > > -- > Troy Guffey What would a human do? Pray? With the computer updating all info that a human could have at 20 times a second and responding at that rate as of last year, I would still bet on the computer AI. I would bet in this case the human would bail out and hope for the best as he roasted in the fireball. The computer AI would stay with the problem to the end, not that I see any way out in this case. -- Douglas E Knapp Why do we live? _______________________________________________ GurpsNet-L mailing list <[email protected]> http://mail.sjgames.com/mailman/listinfo/gurpsnet-l
