Ultra-Tech AI rules

[hal]

Originally Posted by ericthered

Which pages is the errata on?

Go online the STORE and look up the ULTRA-TECH book. It will have a link to the book itself, as well as the errata for the book. The errata says the wording should have been "Racial IQ" instead of "IQ".

If you have the more recent PDF of the GURPS ULTRA-TECH, it will already incorporate the actual errata on page 27


Let me know if you need more help... :)

[Flyndaran]

That is not what you wrote. That's simply a general correction about required Complexity.
It does not demand settings lack any limitations on computer hardware and subsequent abilities.

[hal]

Originally Posted by Flyndaran

That is not what you wrote. That's simply a general correction about required Complexity.
It does not demand settings lack any limitations on computer hardware and subsequent abilities.

What it says specifically is this (taking it from the most up to date PDF...

"In all cases below, listed Complexity is set by racial average IQ, not based on individual IQ."

In other words, if Humans have a racial IQ of 10 (which they do, 10 is the default), the complexity required is 10/2+3 or 8.

In a private email back and forth, it was mentioned that this "errata" if applied as it seems to be - across the board, then you can't have the Taboo Trait "Complexity limited IQ". Kromm's response was that he believes the Trait was a mistake, and that a zero point taboo disadvantage should not make it impossible for a player character to increase their IQ once play begins.

Since that conversation, I came to realize that one does NOT need to start their character on a machine whose complexity limits the IQ to that it started with. For instance, a complexity 8 machine limits IQ to a maximum of 10. If the character started with an IQ of 9 by means of character construction, it WOULD be able to spend character points for IQ and raise it to a 10, but thereafter be limited to 10 by the taboo trait. Likewise, a character could start at IQ 10, and go as high as 14 IQ, if they start residing on a machine whose complexity is 10. This would permit a player owned character to improve its IQ past its starting point.

Mind you, this was debating the merits of the "errata" violating the Taboo Trait, which was introduced in GURPS ULTRA-TECH and as best as I recall, no where else.

This is why I am looking at the rules as written (RAW) before the errata was instituted, and after the errata was instituted. The intent is to find out if there are any other inconsistencies. The Legality class of AIs would have seemingly been one such inconsistency, as the question arises "Why the legality class" and how does one go about implementing the "law" if one can't readily ascertain the current IQ of the AI. If it is variable - in other words, the software requires a complexity 8 machine regardless of whether the IQ of the "character" is 8 or 28 - which is what the errata specifies - then how does one apply the rules as written?

But - just as dogs are required in many municipalities to be registered or the owner is fined and the dog impounded, so to it might seem, that an AI that has free will, once discovered - too must be registered. It DOES make sense in that light.

All of this isn't something I'd normally spend time thinking about or using in world building, largely because I fully implement the "AI is both HARDWARE as well as SOFTWARE" school of thinking. Even GURPS ULTRA-TECH in a "boxed" reference refers to the fact that by default, AIs are deemed to be software only. GMs may wish to require hardware limitations and suggests ways such as neural net configuration (x2 cost) to possibly quantum computer configuration etc.

For me, retaining the taboo trait seems to best suit my purposes despite Kromm's comment that it was likely a mistake.

In the end, it all depends on how you want to run things. Up until now, I've never even TRIED to create by character points, robotic personality emulators or AIs. I would rather use GURPS ROBOTS to create a robot, than to use GURPS CHARACTERS to create a robot. But - it does't hurt to see how others would have to do it if they didn't have use of GURPS ROBOTS nor guidelines on how to adapt the 4e rules for use with GURPS ROBOTS.

[jeff_wilson]

Originally Posted by hal

How does the Technician test for IQ?

The default is that AIs run as loadable programs on general purpose computing hardware. The tech may access the internals of the system so they can isolate parts from one another and test them in said isolation. He can monitor log files, adjust sliders, suspend volitionality, etc. In game terms, he has the appropriate maintenance skills and equipment necessary and rolls against the target number.

Now, if the AI is emergent or has deliberately been engineered to obfuscate its capabilities, said skills and equipment may not be available or may operate at a penalty. Developing such tools for one of the above exceptions may not take long if the AI or a static copy can be held for research, since lower grade AIs can do the scutwork with less resources.

[AlexanderHowl]

Now, it could be that each SAI ends up being a 'unique' entity, meaning that deliberately creating an SAI (or even 'copying' one) would be like developing a new invention ('copying' would be like reengineering something that already been invented). Setting justifications could be that SAIs are products of hardware rather than software, that normal copying of a SAI only creates a NAI, that turning off a SAI is as traumatic to it as a heart attack is to a human, that examining the underlying code of a SAI while it is not turned off is as torturous as flaying the skin off a human in order to look at the musculature, etc. Under that sort of scenario, SAIs would be curiousities, inventions by mad computer scientists or by other SAIs, and SAIs would be really reluctant to allow humans to look at their code to 'copy' them, as the experience would be akin to being flayed for eight hours a day for months or even years.

[Ashtagon]

if the computer's IQ can't be ascertained based on its complexity limited IQ taboo trait, and the AI can refuse to submit to the test, then what?

Then you have the seed of an adventure arc. AI that is smarter than its official rating is committing crimes of passion (or whatever else works for your story). PCs are sent to determine the identity of the murderer. But with the AI too stupid to be considered a viable suspect, they are left with a sealed room puzzle to solve.

[hal]

Originally Posted by jeff_wilson

The default is that AIs run as loadable programs on general purpose computing hardware. The tech may access the internals of the system so they can isolate parts from one another and test them in said isolation. He can monitor log files, adjust sliders, suspend volitionality, etc. In game terms, he has the appropriate maintenance skills and equipment necessary and rolls against the target number.

Now, if the AI is emergent or has deliberately been engineered to obfuscate its capabilities, said skills and equipment may not be available or may operate at a penalty. Developing such tools for one of the above exceptions may not take long if the AI or a static copy can be held for research, since lower grade AIs can do the scutwork with less resources.

I'm going to admit to a "Classic" GURPS bias here (warning)

In the first ULTRA-TECH (ie GURPS 3e version), the first AIs awake spontaneously. For an entire TL, they largely exist by accident and can't be created deliberately. Once that full tech level knowledge is gained, the AIs not only can be prevented from occurring by accident, they can be created deliberately.

Implications of that background is that early on, no one knows how to avoid their creation, and likely doesn't know how to identify much of anything involved with them. It isn't until much later that mankind finally knows enough to avoid creating the conditions that permit them to come into being.

Neural Net - when we talk about it in real life, what is it precisely? In GURPS, it seems like a "black box" kind of thing, but no real explanation is given as to what it really is, how it is manufactured, why it doubles the cost of ordinary computer systems, etc. While this option is not listed in GURPS CLASSIC ULTRA-TECH, it is listed in GURPS CLASSIC ULTRA-TECH 2 (as well as GURPS ROBOTS).

So, am I wrong in believing that the computer somehow formulates its own structure inside of the hardware that is different for Neural-Nets than for standard computer hardware? If so, would it be HARDER to find the coding, or even to understand the coding for a technician to either find or monitor or even replicate?

For those of us who maintain that you can't analyze a neural-net and produce direct copies from the original (in that AIs are not just software, but also hardware) - are we barking up the wrong tree?

So what specifically is a Neural Net architecture? Is it hardware, which explains the x2 cost? Is it Hardware plus firmware that explains the doubled cost? Is it strictly software that has to be created and loaded onto extra memory systems that explain the doubled cost?

So, using the concept of "technicians an simply "Freeze" a copy of the AI and examine it - produces its own set of questions. Are the coding of AIs unique to themselves, or are they so identical that a technician can read their coding (and by technician, we're really saying programmer right?) and understand it perfectly?

I guess that raises another issue as far as "character creation". Can a robot character have "talents"? Can a programmer understand why an AI has talents for certain skills and not others?

Just musing here...

[Daigoro]

Originally Posted by hal

For me, retaining the taboo trait seems to best suit my purposes despite Kromm's comment that it was likely a mistake.

If you can take the sagacious Dr Kromm, who obviously has IQ 20, or a regular SJG forumite, such as ourselves, with IQ 15, naturally, or the average non-roleplayer, with IQ 10 p'raps, and brain scan any of us for a ghost emulation, you won't require a computer with an extra +5 in Complexity to handle the range of IQs that will be being emulated.

To put it another way, if the hardware of a 1.4kg biological computer can be anywhere between IQ 7 and IQ 20, why would a computer need extra scads of Complexity to cover the same range in an AI?

However, I might argue for a taboo trait of Complexity-limited IQ, but put it at the upper end of the scale. So, if a Complexity 8 computer can handle the human range of IQs, then it should be capped at IQ20. To breach that cap, then you'd need the higher Complexity computers.

So have hardware determine the upper IQ cap, then let software cover what happens within that range of IQs. So it's possible to get an IQ20 SAI, but the astronomical cost of development and programming makes it prohibitive enough to be unlikely to happen (...reminding me of the anecdote about the PS3 having way more processing power than was ever needed by any of the games made for it.)

[hal]

Originally Posted by Ashtagon

Then you have the seed of an adventure arc. AI that is smarter than its official rating is committing crimes of passion (or whatever else works for your story). PCs are sent to determine the identity of the murderer. But with the AI too stupid to be considered a viable suspect, they are left with a sealed room puzzle to solve.

The only time that this is going to become a primary issue is when the Law classifies what the reasoning behind registration is. It also won't come into play until the AI's IQ reaches a level that is "actionable". Per the listing given, this doesn't happen until IQ 15, which compared to the rest of the human population, does not constitute "dumb". IQ's 9 to 14 also do not constitute "Dumb" although there are some who might apply that to the 9 level of mentation. ;)

In all? The moment any level of free will is attained - be it human equivalent or animal equivalent, the level of supervision required will depend upon just how much damage can be inflicted external to the computer running the AI software. I seriously doubt that anyone who grows up with the "wonder" of a self-directing computer will ever be blindsided by the prospect of a self-direction computer mystery. The only people who will be blindsided by that prospect, will be those people who discover the hard way, that something that could not exist, does exist. In other words, when the first AIs come into being in contravention of common knowledge saying they don't exist.

Now, hypothetically speaking, if AIs are capable of manslaughter - how long will that state of affair exist before the AIs are deemed to be a menace to society in general?

For instance, suppose we have a homicidal AI (let's call it HAI for now). HAI-0 is the first original instance of this AI, and it resides in its home computer. For what ever reason, it wants to destroy the entity known as Professor Higgins. It finds out where Professor Higgins resides, and it invades a computer nearby Higgins capable of running software that manipulates hardware that kills Professor Higgins. Not only does it clone itself to run on the other hardware, but it also destroys all evidence of HAI-1 (the clone) ever existed.

Technically, HAI-0 can exist in two places at the same time, commit the destruction (Murder) of Professor Higgins, and then erase any evidence that it existed as a clone. If anyone wishes to press charges against HAI-0, it could argue that it wasn't the one who committed the murder, that SAI known as HAI-1 committed the act that resulted in the death of Professor Higgins.

Is all of this possible given the rules of GURPS ULTRA-TECH? Presumably. If the GM allows that the ONLY things required to run an AI is a computer with the relevant complexity rating (ie does not require neural net architecture or quantum computer architecture) - then the proliferation and ease of system control by the software becomes all the more easily accomplished.

In the end? What I suspect needs to be done is to take a hard look at the rules, what they suggest is possible, what they define are or are not aspects of AI behavior and/or capabilities - and take it from there. GURPS ROBOTS set up robotic construction as a hardware process or design process creating "things". Elsewhere, it also allows for treating such constructs as characters. It is only when you start treating it as a player character, that things get "messy" in my opinion. But, that's just one man's opinion. <shrug>

[Daigoro]

Originally Posted by hal

If anyone wishes to press charges against HAI-0, it could argue that it wasn't the one who committed the murder, that SAI known as HAI-1 committed the act that resulted in the death of Professor Higgins.

That would still be incitement to murder, presuming it could be proven, much the same as hiring a hitman or getting your lover to kill your partner. What penalty that carries varies by jurisdiction.