AIs and population control

[biostem]

Let's say you have a setting where AIs, (in roughly humanoid bodies), exist alongside humans, and have rights roughly equal to those of flesh-and-blood people. Assume, in this setting, that these robot bodies can basically be built by anyone with the proper facilities and/or know-how. If the society was also democratically governed, how would you control these robots' population so as to maintain a "fair" state of affairs and/or distribution of resources. These AIs wouldn't need food, but would still need fuel or an energy source to recharge their internal batteries. Let's also assume that non-sentient humanoid robots exist and are used by humans and AIs alike.

[biostem]

I just realized I put this in the "Non-fiction" section - if this would be better placed in the "Science-Fiction" section, please move it. Thank you.

[Thanas]

I think it is fine.

But why would one need population control? Wouldn't there be an obvious limit due to the cost of fabrics and energy?

[biostem]

I think it is fine.

But why would one need population control? Wouldn't there be an obvious limit due to the cost of fabrics and energy?

Well, I think the biggest issue would be whether an AI can directly influence the AIs of any artificial life forms it "births" - if this society gave AIs rights and freedoms, just like humans, what would stop a wealthy AI from simply producing 1 million or however many offspring it can reasonably provide for, in order to swing the outcome of elections in its favor? There is a realistic limit on how quickly humans can reproduce, but these AIs could be produced much more quickly...

[Thanas]

And the danger of that would be what?

[AniThyng]

And the danger of that would be what?

that would depend on how you feel about AI votes overwhelming fleshling votes in elections if we give every individual AI instance one vote each.

I'm sure this can be turned into a thinly veiled human nation demographics allegory if one wished but the question of being able to easily clone sentient AI practically instantly seems interesting too.

[Broomstick]

There is a realistic limit on how quickly humans can reproduce, but these AIs could be produced much more quickly...

Why do you assume AI's would reproduce faster than humans?

Also - assuming these are artificial intelligences, why not design them to have a very low or nearly non-existent impetus to reproduce themselves?

[Starglider]

Why do you assume AI's would reproduce faster than humans?

Creating a new copy of an existing AI program is virtually instantaneous. Creating substantially differentiated AI instances will take longer depending on the exact technology used but not necessarily much longer. Building robot bodies takes hours on a production line instead of a couple of decades of maturation.

Also - assuming these are artificial intelligences, why not design them to have a very low or nearly non-existent impetus to reproduce themselves?

Firstly, because we don't know how to do that with any degree of reliability, and it is an open question whether it is possible (for sentient AIs) at all. Secondly, you know those 'quiverfull' fundamentalist families who believe it is their duty to have as many kids as possible? For normal humans that is a max of two dozen or so. For one sentient AI that decides to start cloning itself the limits are (1) the amount of hardware it can buy, (2) the amount of network connected not-perfectly secured hardware it can hack and (3) the amount of hardware that can be build by robots in a von neuman self replication cycle. The last one is open ended up to the amount of materials available on earth (and potentially beyond).

The current human conception of democracy will not work in a posthuman era. To be fair it doesn't work very well now, it's just less bad than the alternatives, so I would take that as a positive opportunity to find and implement a better system.

[Zixinus]

The short answers: every other sapient individual, but most likely AIs themselves would work out something.

AIs are likely to form communities and will (likely) take dimly to creating millions of AIs just to do voter fraud or other trivial reason. Humans will have limited ability to police AIs, if for no other reason than humans having trouble thinking like AIs and penetrating AI communities. So you will have AIs that will police other AIs.

It is to the benefit of both AIs and humans to control total population. If one AI were to start madly make as many other possible AIs as it could, it would be to the detriment of already existing AIs (if they are happy with how things are). If for no other reason than to control competition of things AIs already do and the drain of resources these extraneous AIs would be. AIs would also likely have some form of agreement with humans about the matter. AIs are also more likely to notice one of their own hacking their way to creating millions of new bodies or whatever.

If a very large number or almost all AIs decided that they needed to outnumber humans for strange, now that would possibly be a problem. But it would be a different problem.

[Broomstick]

Why do you assume AI's would reproduce faster than humans?

Creating a new copy of an existing AI program is virtually instantaneous.

Since we don't have working AI programs whatsoever at present I don't see where you have any basis to make such a claim.

[Guardsman Bass]

If it's ultimately a software program running on interchangeable hardware, then I don't see why you couldn't just make a copy of it (technically - ethical issues are another matter).

[Starglider]

Why do you assume AI's would reproduce faster than humans?

Creating a new copy of an existing AI program is virtually instantaneous.

Since we don't have working AI programs whatsoever at present I don't see where you have any basis to make such a claim.

Copying digital data in general becomes progressively closer to 'instantaneous' as technology advances. The only case in whuch this would not be true is artificial intelligence implemented as an analogue computer with poor or non-existent import/export. Some sci-fi authors like to propose this, because it supports their conceit of unique atomic individuals, but in real life virtually all neural net specific hardware is DSP based with fast lossless state import and export. It really makes no sense not to have this, as even for commercial robots you need it for initial setup, testing, field diagnostics and syncing/export of learned skills. And of course the overwhelming majority of actual AI research is done on conventional workstations, servers and microcontrollers.

[biostem]

The only "excuse" I could come up with as to why AIs couldn't simply be copied from one body to another, would be if there was some woo-woo issue where an AI became unstable if it didn't start from a more simplistic (read: newborn) stage, through a process of learning and adaptation to its body and environment.

I suppose they could implement a lengthy and involved process of "legitimizing" each new citizen born into the society, to check their "uniqueness" and competency, in order to participate in government - i.e. they keep a neural map of each registered citizen on file, and if more than 1 citizen tries to register with the same neural map, then the newer applicant gets rejected - due to the implementation of said AI, attempting to alter a cloned AI enough to pass this test would result in a fundamentally different AI, thereby undermining the entire intent of the mass-copying in the first place.

[bilateralrope]

I have seen one reason for why AI's aren't copied as easily as they can be. The AIs themselves don't want to be copied.

[Purple]

I suppose they could implement a lengthy and involved process of "legitimizing" each new citizen born into the society, to check their "uniqueness" and competency, in order to participate in government - i.e. they keep a neural map of each registered citizen on file, and if more than 1 citizen tries to register with the same neural map, then the newer applicant gets rejected - due to the implementation of said AI, attempting to alter a cloned AI enough to pass this test would result in a fundamentally different AI, thereby undermining the entire intent of the mass-copying in the first place.

And what happens when some AI clones it self a couple million times? What do you do with these beings after you reject them? Are they going to take it lying down?

[Starglider]

I suppose they could implement a lengthy and involved process of "legitimizing" each new citizen born into the society, to check their "uniqueness" and competency, in order to participate in government - i.e. they keep a neural map of each registered citizen on file, and if more than 1 citizen tries to register with the same neural map, then the newer applicant gets rejected - due to the implementation of said AI, attempting to alter a cloned AI enough to pass this test would result in a fundamentally different AI, thereby undermining the entire intent of the mass-copying in the first place.

This is not practical; without getting into implementation detail, suffice to say that for virtually all known AGI design patterns, there is (conceptually) a relatively computationally cheap way to permute the mind-state such that it is still functionally equivalent but looks different enough to defeat any 'dumb' hashing scheme. You would have to do a detailed and lengthly white-box inspection by an AGI of at least the same intelligence, and even that might not work, as well as being a total violation of privacy.

[Broomstick]

Why do you assume any AI's will have any sort of right to privacy? Law is written by and for humans and is slanted heavily in their favor. Until that changes, AI's are property. You can legally destroy your property. Your property has no right to privacy.

While I can argue it's unethical to simply wipe the mind of a true AI at present it is entirely legal. What to do with millions of clones of an AI? Seriously, given the type of gutter slime that has managed to acquire money and power in our society you have to ask that? There are historical instances of humans being butchered by the millions, what's to stop the destruction of a comparable number of AI's?

[Starglider]

Why do you assume any AI's will have any sort of right to privacy?

Because if they don't have that right, they aren't going to have voting rights so the question is pretty much moot. Checking for individual uniqueness is only relevant in the context of stopping people biasing elections by self-copying. If you're simply trying to adversarially restrict the amount of resources that AIs as a whole are using, the question of how distinct the individuals are or even how many individuals there are (the concept of 'individual' may be much more fuzzy for AIs, see the Geth in Mass Effect for a basic fictional example) is irrelevant.

Law is written by and for humans and is slanted heavily in their favor. Until that changes, AI's are property. You can legally destroy your property. Your property has no right to privacy. While I can argue it's unethical to simply wipe the mind of a true AI at present it is entirely legal. What to do with millions of clones of an AI? Seriously, given the type of gutter slime that has managed to acquire money and power in our society you have to ask that? There are historical instances of humans being butchered by the millions, what's to stop the destruction of a comparable number of AI's?

Sure but my point is that in this case, we don't need a test to check how distinct AI instances are from each other, because we don't care.

I personally think this is a moot point because any plausible type of sentient general AI has neligible chance of being effectively restrained by humans, but that's out of scope for the debate.

[Purple]

Why do you assume any AI's will have any sort of right to privacy? Law is written by and for humans and is slanted heavily in their favor. Until that changes, AI's are property. You can legally destroy your property. Your property has no right to privacy.

Go back 50 years or probably even less and this was true for pets. And look at things now.

And pets don't have the intelligence or sheer capability for devastation that an AI would have.

[Broomstick]

Pets still don't have a right to privacy, and while you can't legally torture your pets anymore you sure can have them put down whenever you want.

[Zaune]

While I can argue it's unethical to simply wipe the mind of a true AI at present it is entirely legal. What to do with millions of clones of an AI? Seriously, given the type of gutter slime that has managed to acquire money and power in our society you have to ask that? There are historical instances of humans being butchered by the millions, what's to stop the destruction of a comparable number of AI's?

Well, for one thing there's the fact that there's nothing in the Constitution of any country I'm aware of that specifically says its civil rights provisions only apply to organic sapient beings*, and arguing that sapient AIs don't count as people would at minimum look really bad in the media.

* Or any other sapient being that doesn't quite meet the definition of human, which will become important when someone finally decides to genetically engineer catgirls.

[Purple]

Pets still don't have a right to privacy, and while you can't legally torture your pets anymore you sure can have them put down whenever you want.

That's still a huge step from where they were. And they ain't even sapient or at all capable of arguing, let alone fighting for their rights.

[biostem]

Like I mentioned in the OP, AIs in this setting are basically treated as equal to humans. The government is a representative democracy, where a district elects a representative, and that representative takes part in the federal government decision making process.

Now, assume it is very costly for AIs to reproduce in this setting, but it takes nowhere near 9+ months to "give birth", and the AI is essentially fully "mature" in less than a year. Your average AI citizen, if they wanted to have offspring, could only afford maybe 1 or 2 such children, but what about wealthy ones? No matter how rich I was, I couldn't move into a town, suddenly have 1000 children that all thought and did exactly as I instructed, to change the voting demographic of that region.

Would you need an implementation like the senate, where it's a fixed amount of human and AI representatives per district?

Maybe what I'm missing here is that these AIs may simply be too independent to simply be swayed by other AIs, even if they were direct clones...

[AniThyng]

Like I mentioned in the OP, AIs in this setting are basically treated as equal to humans. The government is a representative democracy, where a district elects a representative, and that representative takes part in the federal government decision making process.

Now, assume it is very costly for AIs to reproduce in this setting, but it takes nowhere near 9+ months to "give birth", and the AI is essentially fully "mature" in less than a year. Your average AI citizen, if they wanted to have offspring, could only afford maybe 1 or 2 such children, but what about wealthy ones? No matter how rich I was, I couldn't move into a town, suddenly have 1000 children that all thought and did exactly as I instructed, to change the voting demographic of that region.

Would you need an implementation like the senate, where it's a fixed amount of human and AI representatives per district?

Maybe what I'm missing here is that these AIs may simply be too independent to simply be swayed by other AIs, even if they were direct clones...

Even if they are not easily simply swayed, it's not unreasonable to assume AI's are more likely to put AI-centric political stances ahead of human-centric political stances and this will affect politics and governance accordingly. Or...well wealth-centric ones. Though really if you're rich you could suddenly find 1000 people willing to vote your way anwyay?

[Starglider]

Maybe what I'm missing here is that these AIs may simply be too independent to simply be swayed by other AIs, even if they were direct clones...

Well sure, if you're making up completely arbitrary characteristics for these AIs, why not.