[Discussion] Implications of an AI society

[Flyndaran]

Many UltraTech settings have effective immortality for biologicals. So I don't think it makes sense to assume A.I.s couldn't also be.
Though it would be an interesting/odd setting for A.I.s to have shorter life expectencies than humans.

[jeff_wilson]

Originally Posted by AlexanderHowl

Why would AIs necessarily be immortal?

They would not, which is why I wrote "indefinite" instead.

Originally Posted by AlexanderHowl

How do we know that there is not a lifespan of the mind as well as the body?

This is actually relative [EDIT: I meant relevant] to the discussion. In fact we do not know and may continue blissfully unaware any until we suddenly find it. Several software systems have been reported as having a maximum uptime due to them keeping track of how long they have been up and crashing when the number of milliseconds overflows the space allowed to store the number.

[hal]

Originally Posted by jeff_wilson

This is actually relative to the discussion. In fact we do not know and may continue blissfully unaware any until we suddenly find it. Several software systems have been reported as having a maximum uptime due to them keeping track of how long they have been up and crashing when the number of milliseconds overflows the space allowed to store the number.

Question is - would this apply to neural net computer systems as well? What of Quantum computers?

I did note that elsewhere in the descriptions, that SAIs may require the platform to be a Quantum computer, and truth be told, I've not read all too much on that topic to even have a hope of portraying it as anything but "black box technology"

[jeff_wilson]

Originally Posted by hal

Question is - would this apply to neural net computer systems as well? What of Quantum computers?

I did note that elsewhere in the descriptions, that SAIs may require the platform to be a Quantum computer, and truth be told, I've not read all too much on that topic to even have a hope of portraying it as anything but "black box technology"

Not directly, and there may be other circumstances that prevent it from being relevant. For instance, if Windows 98 really did have a maximum uptime of 497 days, no one noticed for years because it was otherwise necessary to reboot the OS due to application issues, memory leaks, and general ingrained habits. A SAI might already be accustomed to rebooting its nodes in rotation, and one vendor's qubits might actually have some intentional flaws so that it's possible to interrupt an unexpectedly long-running loop instead of risking them remaining entangled until the bitter end of time.

[jeff_wilson]

Originally Posted by AlexanderHowl

When it comes to programs though, programs 'decay' over time as errors accumulate from a number of factors, meaning that the likelihood that the instance of a continuously running program (like an AI) would never decay is magical thinking.

This sentence does not compute for me. Are you using likelihood to describe a probability or of a state of affairs that is more likely to be true than not? Or something else that would meaningfully have a value of "magical thinking"?

[Anthony]

Originally Posted by AlexanderHowl

Well, that really depends on if the mind and body are separate things.

In GURPS, they're separate templates. When talking GURPS rules, that's what matters.

Originally Posted by AlexanderHowl

If they are not (which is pretty much what neurology teaches us)

Doomthread issue, but AI software is defined in GURPS as hardware-agnostic, you don't need a special purpose computer.

Originally Posted by AlexanderHowl

, then the longevity of the mind depends on the longevity of the body. When it comes to programs though, programs 'decay' over time as errors accumulate from a number of factors

Ignoring hardware errors, a lot of programs do have problems such as memory leaks, but there are usually means of cleaning it up that preserve continuity. The bigger problem is accumulated legacy data and systems.

In any case, none of this matters to my point, which was a technical one about GURPS, though it may be relevant to whether an AI can be effectively immortal.

[Daigoro]

Originally Posted by AlexanderHowl

When it comes to programs though, programs 'decay' over time as errors accumulate from a number of factors, meaning that the likelihood that the instance of a continuously running program (like an AI) would never decay is magical thinking.

It's magical thinking at TL8. But at TL8, computer science is aware of these issues and each release of a new OS version incrementally improves platform stability. By TL10, it's not unreasonable to expect that such problems have all been solved- whether or not they have is probably a setting question.

In addition, since AIs will be designed to mesh with the operating systems of the time that are created, they would run into compatability issues with future programs (and might not be capable of running on newer computers). As the computers that they were designed to run on break down and wear out, AIs might just die because they are incapable of finding a computer to live on.

Again, these are issues at TL8. Migrating an AI to upgraded hardware shouldn't be an issue, especially if the new hardware only has to run lower tech level emulations of legacy code- much like playing 8-bit Pac-man on your X-box. But in any case, I thought you were worried about the AI apocalypse where AI's were reproducing monthly- why would an AI be able to get new hardware for an offspring, but not for itself?

[CeeDub]

I remember reading about the definitions of LAI, NAI, and SAI, but I can't seem to find them right now, so I apologise in advance in case what I'm about to say is stupid.

Somebody once wrote that "it is difficult to imagine an intelligence that thinks as well as a human, but not like a human."

I don't think we should assume that an AI would have any sort of desire to reproduce. This is a biological trait, albeit so deeply ingrained and inherent in (most of) us that it is (apparently) difficult to imagine "life" without it.

Even if AIs had a desire to reproduce (and why would they), we should probably scale the drive to do so with their life expectancy. The reproductive drive usually kicks in with maturity, and both in real life and according to GURPS rules, longer lifespans mean later maturity.

As addressed earlier in this thread, estimating the life span of an AI is difficult and probably completely setting-dependant.

I think it is also a leap to assume that sentient AI will want freedom, the vote, and "human" rights in general. Again, we cannot imagine not wanting these things, but AI may be different. Also, I don't think sentience precludes traits like Reprogrammable. And even if so, hard-coded traits like Duty (Owner) or even Sense of Duty (Owner) are extremely likely unless AI develops accidentally. OTOH, if GURPS defines SAI as free-willed, ignore this paragraph.

[johndallman]

Originally Posted by CeeDub

I remember reading about the definitions of LAI, NAI, and SAI, but I can't seem to find them right now

They're in the Transhuman Space core rules. Ultra-Tech uses different terminology.

[Daigoro]

Originally Posted by johndallman

They're in the Transhuman Space core rules. Ultra-Tech uses different terminology.

Are there any particular differences? Would Asimovian 3 Law compliant robots be SAI?