[Reign of Steel] Zoneminds IQ and Complexity

[RogerBW]

I deliberately didn't put Fixed IQ or Complexity-Limited IQ on the templates for that reason. I did use the Complexity figures and the sizes of TL10 computers as a rough guide, but the Zoneminds have obvious reasons for not wanting their killbots to get too smart – quite separate from the actual limits on how much thinking power can go into a mobile chassis.

And when a robot hits rampancy (a term you won't find in the book, but one I pinched from the Marathon game series because it's useful to describe certain stages of emergent AI development) all bets are off. It's not just raw IQ, either: Five-Three, whom you'll see in one of the chapter head vignettes, isn't particularly smarter than other Tarantulas, but it has disposed of an awful lot of the standard XAU disadvantages.

[Flyndaran]

All human brains of are of equal complexity yet run from profoundly retarded to all mega-geniuses in history.
You don't need to follow linear ratios of IQ to Complexity either.

[hal]

dumb question time:

If IQ can be raised or lowered independent of Computer complexity limits - why bother having the computer complexity limit in the first place?

What specifically is the difference between a TL 9 computer AI that has upgraded its IQ with character points and presumably a neural net structure and that of a TL 10 computer fresh out of the box with its IQ based strictly on its complexity type?

This is something that to me, makes no sense. You want to treat a sentient computer as a character, then build it as a character. If you want to build it as a technologically based gadget, then treat it as such. Mixing them both, in my opinion, makes absolutely zero sense.

[NineDaysDead]

Originally Posted by hal

dumb question time:

If IQ can be raised or lowered independent of Computer complexity limits - why bother having the computer complexity limit in the first place?

What specifically is the difference between a TL 9 computer AI that has upgraded its IQ with character points and presumably a neural net structure and that of a TL 10 computer fresh out of the box with its IQ based strictly on its complexity type?

Racial IQ vs Individual IQ?

[David Johnston2]

Originally Posted by hal

dumb question time:

If IQ can be raised or lowered independent of Computer complexity limits - why bother having the computer complexity limit in the first place?

It's a world building element. It tells you just how smart those NPC AIs are almost all going to be. There aren't after all, any Zone Minds who are actually smarter than average for a Zone Mind even though it's a theoretical possibility.

[sir_pudding]

Originally Posted by hal

dumb question time:

If IQ can be raised or lowered independent of Computer complexity limits - why bother having the computer complexity limit in the first place?

What specifically is the difference between a TL 9 computer AI that has upgraded its IQ with character points and presumably a neural net structure and that of a TL 10 computer fresh out of the box with its IQ based strictly on its complexity type?

This is something that to me, makes no sense. You want to treat a sentient computer as a character, then build it as a character. If you want to build it as a technologically based gadget, then treat it as such. Mixing them both, in my opinion, makes absolutely zero sense.

What is the point of humans having IQ 10 if you are going to have individually smarter and dumber people? Same reason.

[Flyndaran]

Originally Posted by sir_pudding

What is the point of humans having IQ 10 if you are going to have individually smarter and dumber people? Same reason.

Mass production matters. But Zoneminds are more like unique works of technological art. So I don't think statting them out as if they were mundane pieces of well known computers fits realism let alone genre.

Also I don't think it's reasonable for adult humans to increase IQ! let alone entire IQ by more than a point. You can't teach me to genius intelligence even over decades of intense unaging tutelage. Note I said adult as all sorts of things can improve final adult traits during gestation, infancy, and childhood.

[hal]

Originally Posted by sir_pudding

What is the point of humans having IQ 10 if you are going to have individually smarter and dumber people? Same reason.

I believe you're missing the point of my dumb question. GURPS deliberately quantified IQ as an attribute that was mutable, changeable if you will, as a character building process. GURPS robots on the other hand, treats artificial intelligence as software on a finite hardware system. Initially, AIs were not character's built using character rules. While a neural net system is self-programming capable of increasing software skills as if an ongoing self-learning system, at some point in time, the complexity of the software increases in order to grant the higher skill level. If the stock computer used to house the original neural-net holds only 1gig of RAM, and the software self-learns a skill to where it needs 10 gigs of RAM (ie 10x the previous TL's processing power), the computer doesn't miraculously gain 9 additional gigs just because it "evolves". It still suffers the physical limitations of the hardware. All neural nets really do in the original rules (classic rules) is removes the "can't learn" disadvantage. It doesn't add to the complexity of the computer per the classic rules. It does however, add to the IQ value...

In all, discussions of a "racial IQ" for computers is like discussing the gears in Orange juice. It just doesn't compute.

In theory, per the mindset quoted above by Sir Pudding, a small computer with a neural-net option, can grow up (if that is the term), to have the same IQ as a Megacomputer. There are no limits to the IQ of a neural-net computer? That is why I asked the question in the first place. Why not just give racial IQ levels based strictly on computer "type" instead of giving a formula based on hardware complexity considerations?

To that end, racial IQs that can exceed hardware limitations essentially declare that there are NO limits. Characters have no real limits. A biological brain in theory relative to mechanical/electrical computers, has more storage space than a normal computer. A vat-brain computer might exceed its original design parameters, but a physical computer some how adding another 9 gig of RAM just because it has a neural net? I don't buy it.

Remember: an increase in skill level in a program requires a higher complexity computer go run on, than does a lower level skill.

In the end, you all can do as you please, and treat AIs as character's and allow IQ to improve independently of hardware limits, but to me, it does not compute. A DX 9 computer in a robotic body designed with DX 9, isn't going to evolve to DX 10 without requiring some hardware upgrades.

Ah well, don't mind me, I clearly have a dissenting viewpoint than the bulk of you seem to have. Doesn't make my way better than yours if you have fun your way. Point is, my way makes the hardware mean something. A tiny computer neural-net will never be able to equal a Megacomputer of the same tech level no matter how much experience the tiny computer has in comparison with the Megacomputer with zero experience (ie, a stock unaltered Megacomputer) AI.

Ah well, ignore the question.

[sir_pudding]

Who says there's no limits? Racial maxima exist and are zero points features. If humans in your game can't have IQ above 15, then it reasonable to limit AIs to +5 too (or whatever).

[Minuteman37]

Originally Posted by hal

I believe you're missing the point of my dumb question. GURPS deliberately quantified IQ as an attribute that was mutable, changeable if you will, as a character building process. GURPS robots on the other hand, treats artificial intelligence as software on a finite hardware system. Initially, AIs were not character's built using character rules. While a neural net system is self-programming capable of increasing software skills as if an ongoing self-learning system, at some point in time, the complexity of the software increases in order to grant the higher skill level. If the stock computer used to house the original neural-net holds only 1gig of RAM, and the software self-learns a skill to where it needs 10 gigs of RAM (ie 10x the previous TL's processing power), the computer doesn't miraculously gain 9 additional gigs just because it "evolves". It still suffers the physical limitations of the hardware. All neural nets really do in the original rules (classic rules) is removes the "can't learn" disadvantage. It doesn't add to the complexity of the computer per the classic rules. It does however, add to the IQ value...

In all, discussions of a "racial IQ" for computers is like discussing the gears in Orange juice. It just doesn't compute.

In theory, per the mindset quoted above by Sir Pudding, a small computer with a neural-net option, can grow up (if that is the term), to have the same IQ as a Megacomputer. There are no limits to the IQ of a neural-net computer? That is why I asked the question in the first place. Why not just give racial IQ levels based strictly on computer "type" instead of giving a formula based on hardware complexity considerations?

To that end, racial IQs that can exceed hardware limitations essentially declare that there are NO limits. Characters have no real limits. A biological brain in theory relative to mechanical/electrical computers, has more storage space than a normal computer. A vat-brain computer might exceed its original design parameters, but a physical computer some how adding another 9 gig of RAM just because it has a neural net? I don't buy it.

Remember: an increase in skill level in a program requires a higher complexity computer go run on, than does a lower level skill.

In the end, you all can do as you please, and treat AIs as character's and allow IQ to improve independently of hardware limits, but to me, it does not compute. A DX 9 computer in a robotic body designed with DX 9, isn't going to evolve to DX 10 without requiring some hardware upgrades.

Ah well, don't mind me, I clearly have a dissenting viewpoint than the bulk of you seem to have. Doesn't make my way better than yours if you have fun your way. Point is, my way makes the hardware mean something. A tiny computer neural-net will never be able to equal a Megacomputer of the same tech level no matter how much experience the tiny computer has in comparison with the Megacomputer with zero experience (ie, a stock unaltered Megacomputer) AI.

Ah well, ignore the question.

I kind of agree with your opinion Hal with the caveat that like what's been demonstrated in real life, Computer Programs and human minds can better utilize there hardware to greater effect. You can learn mental shortcuts to solve a problem faster and in tern Computer programs can be engineered to do things more efficiently. So yes there is a hard limit on the IQ an AI can have in a given platform, but it's reasonable to say it doesn't start out of the box at that limit and can learn it's way up.