[Discussion] Implications of an AI society

[johndallman]

Originally Posted by Daigoro

Are there any particular differences?

Lots -- THS AIs were designed to support the tropes of the setting, and all have quite a few points in traits that are setting-specific, such as Enhanced Time Sense and Possession (Digital). THS NAIs are non-volitional; SAIs are fully volitional; LAIs are volitional with a bunch of mental disadvantages.

Would Asimovian 3 Law compliant robots be SAI?

That depends on how smart they are. Asimov's Laws are more of a plot generating trope than a serious attempt at AI design. The excellent webcomic Freefall points out a lot of the problems with the three laws.

[CeeDub]

Originally Posted by johndallman

That depends on how smart they are. Asimov's Laws are more of a plot generating trope than a serious attempt at AI design.

And yet, according to tvtropes.org (don't go there if you have a job or a significant other, unless you also have iron will), "Real life roboticists are taking the Three Laws very seriously (Asimov lived long enough to see them do so, which pleased him to no end)."

But that was in the early 90s, I think. These days, I would not be so sure anymore.

[AlexanderHowl]

It is how they fool people into supporting robotics research and development instead of having them grab the nearest sledgehammer. In reality, robots are being developed to be used in the battlefield to kill humans and have already replaced hundreds of millions of human workers across the world. And it will only get worse as robots get more and more sophisticated...

[Daigoro]

Originally Posted by johndallman

That depends on how smart they are. Asimov's Laws are more of a plot generating trope than a serious attempt at AI design. The excellent webcomic Freefall points out a lot of the problems with the three laws.

I presume that was this comic? I can see how restrictions on behaviour might argue against free-will.... but humans themselves can't act purely freely either- from laws of society to childhood behavioural conditioning.

Asimov's Laws are probably a given for early stage AI, but I don't know if it's a given for full-sentience.

[johndallman]

Originally Posted by Daigoro

I presume that was this comic?

Not just that one, but it's the most explicit on the subject. Most of the AIs in that story are not 3-Law.

Asimov's Laws are probably a given for early stage AI, but I don't know if it's a given for full-sentience.

Being able to reliably identify humans, and harm, requires quite a bit of intelligence.

[Daigoro]

Originally Posted by johndallman

Being able to reliably identify humans, and harm, requires quite a bit of intelligence.

I meant that the other way round. Asimov's Laws require a level of AI, but would full-sentience be possible with Asimov's Laws or not?

[TGLS]

Probably. Especially if you take the case that THS's SAI are sapient, as they are obligated to follow the law (because of honesty).

[Kax]

Another thing that might limit AIs is possibly requiring a biological component to the computer they run on--for those settings where AI can't be purely digital.

This would need the same technology used for brain uploads to do backups and duplicates, making both chancy and expensive.

Both an inherent limit on numbers and an addition to the discussion on the difference between a human with brain cyberware, a human/other brain upload, and an AI--where's the dividing lines?

But this gives a possible origin for them: cyberware research for brain modification and upgrades involves attaching very powerful and very smart digitial computers to brain tissue and playing with how the combination thinks. It's not much of a stretch when working like that to see how it goes if you leave the human off and just use brain tissue.

If it works, voila. And AI brains that are damned hard to duplicate usefully and are more expensive than a normal computer.

This may have been part of the development of THS AIs, where they managed to duplicate the bio component digitally. Or maybe they still have small bio cores.

[AlexanderHowl]

Biocomps reduce the effective Complexity of AI programs by one, which would make them a likely canditate.

[jeff_wilson]

Originally Posted by AlexanderHowl

It is how they fool people into supporting robotics research and development instead of having them grab the nearest sledgehammer.

Okay, I think I see now why you are so harshly against automation. I'm a machine learning researcher myself and I hope that I can explain that while there is rather a lot of military spending on automation research and implementation the aim is to cut down on human suffering and death by the letting non-living machinery do some of the work that has to be done in war zones and other threatened places so fewer fragile human bodies need to be sent there.