Keeping humans relevant in the shadow of TL10 AI.

[Michael Thayne]

Originally Posted by ericthered

Here are a few key things to remember about TL10 AI and computing:

• IQ 10 Volitional AI's are complexity 8. This means their hardware at a minimum costs $10,000 and that version weighs 40 lbs. An IQ 12 Volitional AI runs on hardware costing $100,000 dollars, and the cheap version weighs 400 lbs. As for why the IQ 10 version is not necessarily sufficient:
• Software to give +2 to the use of any skill is at most complexity 7. I generally do not allow AI to take advantage of this bonus, so the average human skill has a generic +2.

Combined, these may be enough to keep meat-space humans in the economic running, at least for some tasks. Its a close thing though.

Of course, when I run a TL10 setting, I carefully decide how AI works in this setting and alter the allowed technology to fit this vision. Some things I've used or considered include:

• AI is slaved to humanity, and most humans can command most AI's. Humans remain relevant because they are in charge. AI's prop up the traditional way of life
• AI is fairly alien, and struggles with some areas, especially art and psychology. Humans are necessary for those tasks, and remain a vital part of the economy.
• AI is more complex than we thought. This one is sort of a cop-out, because what I'm really doing is using TL9 AI instead of TL10. Occasionally I'll instead say AI can't be miniaturized effectively, so you have a handful of big, smart machines without the pervasive human-level stuff everywhere. Embrace the madness. Humans can be uploaded and become machines themselves. Or just have the players all be AI's to start with.
• Legal concerns make an AI much more expensive than its hardware and software. After all, barracks style work-camps are fairly rare in the modern first world.

One wrinkle here: $100,000 sounds like a lot, but it's less than a couple years salary for an Average job at TL10.

The idea of not letting AIs get bonuses from software tools is interesting, but I'm not sure it's consistent with RAW.

[weby]

You have to put some thought into how you want to handle things, options include things from the social to technical.

Some possible things to use/mix and match:

Social limits like:
-many jurisdictions in Transhuman space where true AIs have licensing requirements and are considered people, so you have to pay them a salary and similar things.
-Perhaps a forbidding or true artificial intelligences due to something like a robot revolution in the history like the Mass Effect universe.

Make human better by things like:
-Having most be advanced races like those from Biotech
-Making things like the +IQ brain implants very common.
-Allow some very fast skill learning/use methods for humans. (the rapid learning things, skill chips and more)

Use Raymond Kurzweil solution where humans have "evolved" into hybrid biological/technical beings by adding an AI layer and thus making MUCH smarter.

Handicap the AIs by making them less competitive at things. Some examples:
-perhaps also have hidebound even in the full AI template.
-making it impossible to buy skills for AIs as such instead forcing the AIs to learn by being taught or self study.
-make skills learned require program slots with high raise in the complexity at advanced point levels thus limiting AIs to low points in skills.
-make it hard to make high level AIs, perhaps most AIs over a certain limit go insanely very quickly.


In my own scifi setting I have the "True AIs are forbidden" as base thing, as AIs are still done illegally and such, so I also use quite many of the other things in many ways as there are still AI. Thus even the "Big bad enemy AI central computer" will not be that much better at things than PCs.

[AlexanderHowl]

Of course, volitional AI is even more science fiction than commercial fusion or human genetic engineering right now. We have experimental fusion reactors, we do not have experimental volitional AI. We have animal genetic engineering, we do not have even animal intelligence volitional AI. As far as we know, volitional AI is as much superscience as artificial gravity, FL drives, and reactionless engines.

[Flyndaran]

Intelligence cannot be superscience, because we exist. How easy it is to create regardless of our research intensity is the question. Whether it's possible to invent within the next century or if it takes a millennia or longer is what we cannot know just yet.
Personally, I think it's closer to a millennia, but I wouldn't start looking for magic if some kind gets developed before I turn 94. Stranger things have happened.

[David Johnston2]

Originally Posted by Michael Thayne

One major omission is Ultra-Tech is any sense of the availability of AIs with various skill levels. This represents a huge unanswered question about any ultra-tech setting. If AIs with skill-18 are ubiquitous, there's nothing for humans to do. But Ultra-Tech seems to assume that TL10 societies will still be fairly recognizable, with human obsolescence not becoming a real threat until TL11 or even TL12. But how do you enforce this in-game?

Transhuman Space vaguely gestures at ways of doing this. The first Personnel Files book is explicit that AIs with skill levels above what it calls "ordinary expert" are rare. In practice, many off-the-shelf AIs have as few as 4 points in their main skill. While AI software can theoretically have complexity as high as 10, I'm only aware of one stray reference to such a powerful AI in any of the sourcebooks. Most of the examples of AIs mentioned are NAIs and LAIs that are at best as smart as the average baseline human. The smattering of SAI examples range from SAI-7 to SAI-9, with the latter being implied to be mostly in use in the military or cutting-edge research facilities.

Ultra-Tech uses slightly different rules for AIs, but you could do something similar. Make AIs more complicated than a fast non-volitional AI with IQ 10 (a Complexity 8 program, requiring a microframe or fast personal computer to run) at least unusual. Don't be afraid to limit AIs to 4 points in their main skill, and 1 or 2 points in other skills. You can make exceptions to the 4 point rule - maybe 8 points in some Electronics Operation specialty is common because most electronics in the setting "just work" even for untrained users.

Thoughts on this approach? What have other people done? At some point, I want to try working out the effects of this approach on the role of drones in spaceship combat, but that's a project for another day.

In my TL 10 setting, autonomous weapons are massively illegal for private ownership, robots for the civilian market have hard-wired Pacifism, and even military autonomous weapons are heavily restricted by interstellar agreements and policy. There have been hacking incidents. It got ugly.

[AlexanderHowl]

There is no proof that non-biological volitional intelligence is capable of existing except through superscience, it is just a hope of computer scientists, science fiction fans, and transhumans that it is not superscience.

[ericthered]

Originally Posted by Michael Thayne

One wrinkle here: $100,000 sounds like a lot, but it's less than a couple years salary for an Average job at TL10.

The idea of not letting AIs get bonuses from software tools is interesting, but I'm not sure it's consistent with RAW.

Skill rules for AI are quite sparse. I mostly justify the AI bonus denial on the grounds that they're both using the same technology and you can't stack very fine on a sword twice to get +4 damage. But yes, its somewhat idiosyncratic


And I acknowledge that the two above aren't sufficient by themselves to keep humans relevant. They just make it easier. As I implied earlier, and should have said more clearly: they help keep humans relevant, but you still need to add a little extra something to make it work.

[David Johnston2]

Originally Posted by AlexanderHowl

There is no proof that non-biological volitional intelligence is capable of existing except through superscience, it is just a hope of computer scientists, science fiction fans, and transhumans that it is not superscience.

There is no reason to think it can't though. And this is beside the point anyway. You don't need volitional AI to replace humanity for most functions. You don't even want it. Limited AI is perfectly adequate.

[vicky_molokh]

Originally Posted by AlexanderHowl

Of course, volitional AI is even more science fiction than commercial fusion or human genetic engineering right now. We have experimental fusion reactors, we do not have experimental volitional AI. We have animal genetic engineering, we do not have even animal intelligence volitional AI. As far as we know, volitional AI is as much superscience as artificial gravity, FL drives, and reactionless engines.

Amusingly, what people consider to be criteria of rudimentary volition ('experimental volitional AI') seems to be a pair of shifting goalposts, for as soon as a computer begins being able to do it, people start saying that nope, constructing sentences or making a painting or performing a tactical decision doesn't count as rudimentary volition. It's also amusingly similar how people shift the goalposts of what counts as a sign of intelligence - first it was any symbol use/recognition or improvised tool, then syntax and toolmaking, then syntax that is subject to regional dialect, then something else ad infinum.

[evileeyore]

Originally Posted by vicky_molokh

Amusingly, what people consider to be criteria of rudimentary volition ('experimental volitional AI') seems to be a pair of shifting goalposts, for as soon as a computer begins being able to do it, people start saying that nope, constructing sentences or making a painting or performing a tactical decision doesn't count as rudimentary volition. It's also amusingly similar how people shift the goalposts of what counts as a sign of intelligence - first it was any symbol use/recognition or improvised tool, then syntax and toolmaking, then syntax that is subject to regional dialect, then something else ad infinum.

I've got a pretty solid line of where I draw "Intelligence"*... and a lot of animals have crossed it. I haven't found evidence yet of AI crossing it.


* Read as the steps from sentience to sapience and finally sophonts. There are a lot of animals I class as sapient. They lack only the ability to communicate effectively† with us to be consider 'sophonts'.

† There are animals I consider as sophonts as well.