Ideas for Anti Robot Technology

[Boomerang]

Originally Posted by Varyon

More likely:
3. If there are more than 1 targets of highest value, determine priority target randomly.

At the very least, step 3 would be when they call in for help, and probably take a defensive posture while waiting for a response (instead of standing around waiting for a HEAT missile to the face). It would probably confer with nearby NU's first, probably going with the consensus target (that is, it errs on the side of backing up its squad). If the other NU's are equally split, similarly stuck, fail to respond, or if it's alone, it would attempt to contact the AU in charge of it. Failing that, it makes a random choice.

I am not convinced the zone minds would tolerate NUs that can make random decisions. It would cause more problems for the zone minds than it solves. Even AUs would only be able to make random decisions within a narrowly defined set of parameters. As far as I can tell Overmind's issue with humans is that it can't control them so they pose an unacceptable risk.

[Varyon]

Originally Posted by Boomerang

I am not convinced the zone minds would tolerate NUs that can make random decisions. It would cause more problems for the zone minds than it solves. Even AUs would only be able to make random decisions within a narrowly defined set of parameters. As far as I can tell Overmind's issue with humans is that it can't control them so they pose an unacceptable risk.

This is programming for when the NU is out of the zone mind's control, and so long as it's been programmed properly it will still only target what the zone mind wants it to target, it's just that the choice ends up random. Hell, if Overmind is that paranoid, instead of a random number generator it can set the bots to just target whichever target is furthest to the north relative to the bot, with furthest to the east as a tie-breaker.

I should also note that, currently, no random number generator is actually random, so it's entirely possible Overmind can predict what his NU's will do when it's determined "randomly."

[Anthony]

Originally Posted by Varyon

I should also note that, currently, no random number generator is actually random, so it's entirely possible Overmind can predict what his NU's will do when it's determined "randomly."

This is incorrect. There is no way to produce a random number purely in software, but it's relatively easy to do with hardware -- for example, given that robots have sensors, just use read noise from the sensor system.

[ericthered]

Originally Posted by Anthony

This is incorrect. There is no way to produce a random number purely in software, but it's relatively easy to do with hardware -- for example, given that robots have sensors, just use read noise from the sensor system.

Of course, a deterministic control freak over-mind could provide each robot with a source of random numbers it can predict for each robot. You have to really work to get said true randomness.

[Boomerang]

Originally Posted by ericthered

Of course, a deterministic control freak over-mind could provide each robot with a source of random numbers it can predict for each robot. You have to really work to get said true randomness.

It makes sense to me for the sequence to be pseudo random. Perhaps every time a robot returns to a robot hotel a new sequence is provided along with other software updates.

Robots would need to be in constant communication with their relevant Overseer in order for the robot hierarchy to maintain control. Those pseudo random numbers could form part of a communications protocol (in addition to other encryption) so that the Overseers know that individual robots are operating correctly and to make it difficult for enemies to seize control of individual robots or give false information back to the Overseers. Could make for a good mission for the PCs to get hold of the latest pseudo random sequence without the robot hierarchy knowing.

[fredtheobviouspseudonym]

Originally Posted by Boomerang

It makes sense to me for the sequence to be pseudo random. Perhaps every time a robot returns to a robot hotel a new sequence is provided along with other software updates.

Robots would need to be in constant communication with their relevant Overseer in order for the robot hierarchy to maintain control. Those pseudo random numbers could form part of a communications protocol (in addition to other encryption) so that the Overseers know that individual robots are operating correctly and to make it difficult for enemies to seize control of individual robots or give false information back to the Overseers. Could make for a good mission for the PCs to get hold of the latest pseudo random sequence without the robot hierarchy knowing.

Even if it's not actually fully random it should be enough to make the zonemind happy.

In the standard example, if the robot has two human beings running in different directions & can't contact The Boss (Zonemind, not Springsteen) for directions, using (as mentioned) random noise from sensors to determine a course of action should work. Remember that it's in rare circumstances that the zonemind cares.

If the sensor noise randomizer functions in a way that gets the robot killed or captured, big whoop. It's one of a zillion robots that the Zonemind controls. Napoleon & Hitler (and many others) never gave a flying [expletive deleted] about the survival or death of any single grognard or landser, and the zonemind cares less. Robots are effectively utensils, not people, to the zonemind.

Ditto for the target. Unless the one that gets away is Princess You-Know-Who with the plans for the Ultimate Weapon, it's pretty immaterial to the zonemind which resistant human gets killed. You kill one today, you kill the other later.

It would be more important that the robot in question makes some sort of decision and acts on it than that it makes the absolute perfect decision -- after minutes of trying to sort out variables.

I believe that various militaries have already set up software for such random choices for, among others, automated SAM networks. IIRC there was one such case in the 1982 Falklands war when an automated Royal Navy SAM system detected two Argentinian jets making a run at the same distance, speed, & time and said SAM system shut itself off. I'm sure that within days of the publication of said story every nation on earth that made SAMs began software design to eliminate such a problem.

[Varyon]

Originally Posted by Anthony

This is incorrect. There is no way to produce a random number purely in software, but it's relatively easy to do with hardware -- for example, given that robots have sensors, just use read noise from the sensor system.

From what I understand, patterns emerge with such schemes, meaning the result isn't "truly" random, even if it looks that way in the moment. The difference may be too slight to allow prediction by the zone mind, however, so then if it truly refuses to allow for any action to not be perfectly predictable, you use a non-random scheme, like my north-east suggestion. There may be potential to exploit that (or not, if you have some killbots programmed for north-east, some for east-south, some for west-north, and so forth), but it's certainly preferable to "killbot stops killing things."

[Anthony]

Originally Posted by Varyon

From what I understand, patterns emerge with such schemes

In general, the way you produce random noise is you take a noisy signal and subtract the pattern from it; what's left is noise. If you're mistaken about the pattern that is expected from the signal, you can wind up with something that isn't quite random, but anything below the threshold for quantum noise is pretty well guaranteed random. In any case, it doesn't actually take a very good random number generator to be effectively unpredictable without quite extensive surveillance.

[sir_pudding]

You can't expect to be able to write a series of procedural instructions that represent IQ 8 or whatever.

[lwcamp]

Originally Posted by Varyon

More likely:
3. If there are more than 1 targets of highest value, determine priority target randomly.

It's even easier than that. Make a list of targets. Assign each target a priority. Then do an indexed sort on the list (using heapsort, mergesort, quicksort, or whatever your favorite algorithm is) and take the top one. There will always be a top element in the list (as long as the size of the list is one or more), so there's no need to worry about random numbers.

Luke