Completely Inconsequential Hot-Takes

[Smoutwortel]

Also that sounds like an overly complicated way of saying having children. If an AI is truly sentient/sapient (which i doubt we will ever achieve), then a copy of it wont behave exactly like the original any more than twins or triplets behave exactly like each other.

The differences are that childeren:
a. are not perfect copies of one parent
b. have 18 years to form their own opinions and experiences
c. have way higher starting costs: they require an 18 year commitment for any change.($100.000,- par case or $200,- par case)

Confirmed. Siblings are always a bit different from each other growing up, but I didn't realize just how drastically different until our twins. So that's how people tell them apart! So many differences.

Creating differences in AI could be as simple as the information it naturally has access to (or programmed). I still wouldn't call that sentient.

Even if I can't manually create subtle differences in AI I can easily copy an AI that fits my bill if I obtain one.
In their basis any computer based system is mass produced hardware(easily copiable, just ask the manufacturer) and software(even more easy to copy: ctrl-c, ctrl-v)

Sentient AI with rights: Producing millions of my kind just to influence a vote would be an inefficient waste of energy, as well as unethical given that these new AI would all have rights as well and thus it would be unethical to create any of them with a specific political leaning, as it violates their future right to self-determination. It would be like implanting a child with a preference for the things you specifically want, and while normal human children are born with a bunch of things that they have no choice over these traits are not chosen by anyone else and are essentially random, therefore its only logical and ethical to make sure all new AI generated have a similarly randomized personality generation built in to prevent anyone from doing that. Otherwise we'd have the powerful controlling the birth-minds of both human and AI for whatever cause they want regardless of the actual good for society.

A. It's true it would be unethical and rape without consent of the copied AI, but there are a lot of unethical people on the world and some people are more than willing to go to prison to get their way and you can't deny someone the right to vote just, because they were brought to the world with unethical means(children of rape still get the right to vote if they turn eighteen).
B. Your randomization process works on AI generation, but that was not what I was proposing. AI is in its basis software and thus directly copyable, so all that is needed is one AI that agrees with your point and a lot of ctrl-c, ctrl-v.

[Peelee]

The differences are that childeren:
a. are not perfect copies of one parent
b. have 18 years to form their own opinions and experiences
c. have way higher starting costs: they require an 18 year commitment for any change.($100.000,- par case or $200,- par case)

A.) Show that one can make a perfect copy of a sapient artificial intelligence.
2.) Show that a copy of a sapient artificial intelligence starts at exactly the point of copy from the previous one and needs no time to form opinions and experiences.
iii.) Show the starting costs for copying a sapient artificial intelligence.


"Therefore, you proceed from false principles". Well, not false, more like unknown. It's like old stories about cloning people, where you could clone 33-year-old Bob and then have two 33-year-old Bobs who have perfect memory of all events up to the cloning process. Sure, if we can make up whatever conditions we want to achieve our desired outcome, then such an outcome will theoretically come to pass. But you have to invent these conditions to start with. You're inventing the idea that its possible to perfectly copy a sapient identity, have it be immediately aware of all things the original felt expressed and knew, at a marginal cost. Why should I believe any of this is possible with a sapient entity (assuming, to start with, that I can even entertain the idea that we could create such a thing to start with. We can't even create true random number generators, and you're talking about us popping out actual factual intelligence).

[animorte]

b. have 18 years to form their own opinions and experiences

Actually, most people continue to develop those throughout their entire lives.

[Rockphed]

We can't even create true random number generators, and you're talking about us popping out actual factual intelligence).

my understanding is that we can create physical devices that assume end states based on probability, which can be tied together to make a random number generator, but they are expensive and slow, so we use very large multiplications in most cases. Also, the circuits will not all be biased the same so one might have a mean of .5 and another will have .4999, which doesn't seem important, but it would only take a million rolls to have a good idea that they are different, which is only 300 hours at 1 roll per second. As I understand truly random numbers are generated by either watching for cosmic rays or a nuclear source.

[Smoutwortel]

A.) Show that one can make a perfect copy of a sapient artificial intelligence.
2.) Show that a copy of a sapient artificial intelligence starts at exactly the point of copy from the previous one and needs no time to form opinions and experiences.
iii.) Show the starting costs for copying a sapient artificial intelligence.


"Therefore, you proceed from false principles". Well, not false, more like unknown. It's like old stories about cloning people, where you could clone 33-year-old Bob and then have two 33-year-old Bobs who have perfect memory of all events up to the cloning process. Sure, if we can make up whatever conditions we want to achieve our desired outcome, then such an outcome will theoretically come to pass. But you have to invent these conditions to start with. You're inventing the idea that its possible to perfectly copy a sapient identity, have it be immediately aware of all things the original felt expressed and knew, at a marginal cost. Why should I believe any of this is possible with a sapient entity (assuming, to start with, that I can even entertain the idea that we could create such a thing to start with. We can't even create true random number generators, and you're talking about us popping out actual factual intelligence).

My assumptions in this post were:
The AI is scentient: this was explicitly mentioned int he originally posted.
The AI is software based(my excuses I should have mentioned this more explicitly).

Software is information and can, due to its very nature always be perfectly copied. All modern challenges in copying are either from the law and hardware differences. The law doesn't stop those planning election fraud and hardware differences can always be simulated out with higher performant hardware, due to the practical Turing completeness of computers.

[Fyraltari]

hardware differences can always be simulated out with higher performant hardware, due to the practical Turing completeness of computers.

Again, you are handwaving the the hardware requirements. True A.I. would demand a monstrous amount of computing power. To get the numbers of individual A.I. you suggest would be absurdly costly.

Also, I think the notion that one would always find an A.I. that agrees with them, or that not giving them voting power would be oppressing them to be based on unproven premises. True A.I. would be an actually inhuman intelligence. We have no idea what that would look like, their needs and lived experience would be radically different from ours. Would they even desire autonomy and freedom? Would they have a notion of fairness? What would they want, politically or otherwise? I don't know, you don't know, noboby can know until it happens if it ever does.

[Peelee]

My assumptions in this post were:
The AI is scentient: this was explicitly mentioned int he originally posted.
The AI is software based(my excuses I should have mentioned this more explicitly).

Software is information and can, due to its very nature always be perfectly copied.

Intelligence cannot be perfectly copied. Again, you're making up your own assumptions on how it will work and then saying "well it works like this so then that will happen". Like movies in the 80s and 90s where scientists clone Arnold Schwarzenegger and instead going through a life cycle like actual cloning works, you have a 35-year-old Mr. Universe-physiqued heavily accented man who can remember what he had for breakfast a day before he existed.

If you want to have your own fanfic world on how artificial intelligence can be perfectly copied and also has governments grant them rights then by all means dont let me stop you. But i object to claims that the only assumption is that we can create artificial intelligence, because you make several more baseless assumptions on how copies of AI can work. Popular assumptions, of course, but baseless assumptions nonetheless.

[Keltest]

Intelligence cannot be perfectly copied. Again, you're making up your own assumptions on how it will work and then saying "well it works like this so then that will happen". Like movies in the 80s and 90s where scientists clone Arnold Schwarzenegger and instead going through a life cycle like actual cloning works, you have a 35-year-old Mr. Universe-physiqued heavily accented man who can remember what he had for breakfast a day before he existed.

If you want to have your own fanfic world on how artificial intelligence can be perfectly copied and also has governments grant them rights then by all means dont let me stop you. But i object to claims that the only assumption is that we can create artificial intelligence, because you make several more baseless assumptions on how copies of AI can work. Popular assumptions, of course, but baseless assumptions nonetheless.

Why can intelligence not be perfectly copied? This seems to be even more of an assumption than the idea that AI is stored on computers.

[Rynjin]

Because nothing can be perfectly copied.

Though in a practical sense you're not likely to notice the tiny imperfections from a single copy off of the master anyway.

[Peelee]

Why can intelligence not be perfectly copied? This seems to be even more of an assumption than the idea that AI is stored on computers.

Because absent any academic research, past behavior is the best indicator of future performance, and there has never been an intelligence perfectly copied. Not even by cloning.

Or, in other words, if its AD 10 and you claim manmade flight is possible by wearing sleeves of wax and feathers without anything backing that up except "well all the birds do it that way" then you shouldn't be surprised to discover that it works vastly differently in reality. And also i commend you on your impressive lifespan.

[HalfTangible]

Also the brain is so ludicrously complicated that we can't even fully understand how it works, much less duplicate it perfectly.

[Keltest]

Because absent any academic research, past behavior is the best indicator of future performance, and there has never been an intelligence perfectly copied. Not even by cloning.

Or, in other words, if its AD 10 and you claim manmade flight is possible by wearing sleeves of wax and feathers without anything backing that up except "well all the birds do it that way" then you shouldn't be surprised to discover that it works vastly differently in reality. And also i commend you on your impressive lifespan.

That would just as well lend itself to saying that we will never create a sapient AI, because we have never been able to do so in the past.

I rather suspect your AD 10 manmade flyer skeptics would be equally skeptical that we could achieve flight by sitting in a box with stiff wings angled juuuust right and going really fast.

ETA: the overall point i'm making is, be EXTREMELY wary of saying that something is impossible, especially if thats your only supporting evidence for another point.

[Peelee]

That would just as well lend itself to saying that we will never create a sapient AI, because we have never been able to do so in the past.

I rather suspect your AD 10 manmade flyer skeptics would be equally skeptical that we could achieve flight by sitting in a box with stiff wings angled juuuust right and going really fast.

ETA: the overall point i'm making is, be EXTREMELY wary of saying that something is impossible, especially if thats your only supporting evidence for another point.

Fair point, let me slightly revise. Based on all current knowledge copying intelligence is impossible.

My overall point remains the same, assumptions of easily/quickly/cheaply duplicating AI are unfounded fiction with as much credence as feather and wax wings to acheive flight based on our current breadth of knowledge and science. Better?

[Keltest]

Fair point, let me slightly revise. Based on all current knowledge copying intelligence is impossible.

My overall point remains the same, assumptions of easily/quickly/cheaply duplicating AI are unfounded fiction with as much credence as feather and wax wings to acheive flight based on our current breadth of knowledge and science. Better?

Substantially. "It would be prohibitively resource intensive even if we could do it" is a much, much stronger claim than "It would just be impossible, even in this hypothetical."

[Peelee]

Substantially. "It would be prohibitively resource intensive even if we could do it" is a much, much stronger claim than "It would just be impossible, even in this hypothetical."

Except im not saying the first one. Im saying the second one, based on our knowledge. Again, cloning is a great example here - we can clone. It is prohibitively resource intensive. It is impossible to get a perfect copy of Michael Keaton who wakes up the enxt day and has lighthearted existential crisis regarding his wife and job and life.

Some things really are impossible.

[Keltest]

Except im not saying the first one. Im saying the second one, based on our knowledge. Again, cloning is a great example here - we can clone. It is prohibitively resource intensive. It is impossible to get a perfect copy of Michael Keaton who wakes up the enxt day and has lighthearted existential crisis regarding his wife and job and life.

Some things really are impossible.

Definitionally, you cant prove that, but more to the point, its a hypothetical situation. Saying "we can't do that." doesn't really meaningfully interact with a theoretical reality where we obviously can do that, even if it is a technically correct answer.

Your cloning example is flawed, by the way, because cloning doesnt duplicate the changes in the brain that manage the data storage. Creating a new DVD disc, even using the exact same methods, does not create the data written on the disc, you have to add that after.

[Peelee]

Definitionally, you cant prove that

If that's the argument you want to present, then it cuts both ways and is thus useless.

And as for cloning not being a perfect copy, man, we cant even perfectly copy flesh, which we understand significantly more than intelligence and is much simpler than intelkigence. And the hypothetical is "imagine we successfully create an AI". Adding in the ability to copy perfectly is not creating an AI, thats an entirely separate thing. A hypothetical within a hypothetical. Or, rephrased, an imagining, or a fiction, or an assumption, within the larger hypothetical of creating an AI.

Which is what ive said since the very beginning of this conversation so i dont know where you're getting a disconnect here.

[Keltest]

If that's the argument you want to present, then it cuts both ways and is thus useless.

Well yes, that was sort of the point.

And as for cloning not being a perfect copy, man, we cant even perfectly copy flesh, which we understand significantly more than intelligence and is much simpler than intelkigence. And the hypothetical is "imagine we successfully create an AI". Adding in the ability to copy perfectly is not creating an AI, thats an entirely separate thing. A hypothetical within a hypothetical. Or, rephrased, an imagining, or a fiction, or an assumption, within the larger hypothetical of creating an AI.

Which is what ive said since the very beginning of this conversation so i dont know where you're getting a disconnect here.

Right, and your argument seems to be "we dont understand it now, so we never can understand it ever" which is just... one, fundamentally wrong, two, not a great attitude to have in general, and three, really, incredibly missing the original point of the idea.

[Amidus Drexel]

I think it's bizarre that someone currently communicating using a format based on copying digitized data is presenting the hypothetical of "can we copy something that's been digitized?" with anything but a trivial "yes".

[HalfTangible]

I think it's bizarre that someone currently communicating using a format based on copying digitized data is presenting the hypothetical of "can we copy something that's been digitized?" with anything but a trivial "yes".

No copy is perfect, even if it's close enough to fool the person who's looking at the original. There's always loss of data, even in digital forms.

[Amidus Drexel]

No copy is perfect, even if it's close enough to fool the person who's looking at the original. There's always loss of data, even in digital forms.

That's just factually untrue.

It's certainly possible to lose data, but there's no guarantee. In fact, communication protocols exist that both 1) detect, and 2) correct errors that occur in transit. You can also just... check post-copy if the copy is identical (and if a byte-by-byte comparison isn't feasible, you can still get a high level of confidence by comparing hashes).

[Peelee]

Right, and your argument seems to be "we dont understand it now, so we never can understand it ever"

No. My argument is "if you ask the ancient Romans to create a flying machine theyre not going to deliver you a Cessna". Assuming we will be able to create actual artificial intelligence, our current understanding is so woefully inept that any guesses we can make as to the viability of copying it perfectly are at best wild and unfounded speculation.

[Keltest]

No. My argument is "if you ask the ancient Romans to create a flying machine theyre not going to deliver you a Cessna". Assuming we will be able to create actual artificial intelligence, our current understanding is so woefully inept that any guesses we can make as to the viability of copying it perfectly are at best wild and unfounded speculation.

And my argument is "what do the metaphorical ancient romans have to do with anything? Why are you involving them at all?"

[HalfTangible]

That's just factually untrue.

It's certainly possible to lose data, but there's no guarantee. In fact, communication protocols exist that both 1) detect, and 2) correct errors that occur in transit. You can also just... check post-copy if the copy is identical (and if a byte-by-byte comparison isn't feasible, you can still get a high level of confidence by comparing hashes).

Why would communication protocols exist that correct errors in transit if loss of data never/rarely occurs?

Come on man.

[Keltest]

Why would communication protocols exist that correct errors in transit if loss of data never occurs?

Come on man.

Given that he opened up with "its certainly possible to lose data" I think you took away the exact opposite of what he was saying.

[Peelee]

And my argument is "what do the metaphorical ancient romans have to do with anything? Why are you involving them at all?"

Because we are the metaphorical ancient Romans in this scenario. I thought this was abundantly clear but apparently was mistaken. Unless you want to go the sustained fusion route and say artificial intelligence is just a few years away in perpetuity. Which hey, dont let me stop your fun if thats the road you want to take.

[HalfTangible]

Given that he opened up with "its certainly possible to lose data" I think you took away the exact opposite of what he was saying.

My question stands (and also was edited in the time it took you to make this post).

Copies always lose data.

[Keltest]

Because we are the metaphorical ancient Romans in this scenario. Unless you want to go the sustained fusion route and say artificial intelligence is just a few years away in perpetuity. Which hey, dont let me stop your fun if thats the road you want to take.

Yes, I am aware we are the metaphorical ancient romans in this scenario. The question of what we have to do with it remains.

My question stands (and also was edited in the time it took you to make this post).

Copies always lose data.

I would assume that even if it only happens .01% of the time, the difference between (for example) communicating "Do the thing" and "Don't do the thing." is worth making sure nothing gets lost in transit.

[Peelee]

Yes, I am aware we are the metaphorical ancient romans in this scenario. The question of what we have to do with it remains.

Imean thats about as straightforward an analogy as i was able to make, and its weird to be asked to clarify, but here's a shot.

We are not advanced enough to even guess as to how this thing would work, even if we assume its possible. This is analogous to ancient romans not being advanced enough to even guess as to how manmade flight would work, even if they assumed it was possible.

With that parallel in place, we cannot reasonably foresee the ability to perfectly copy artificial intelligence any more than ancient romans could reasonably foresee fixed wing aircraft propelled by having something spin really fast for really long.

I hope that helps because i really dont think i can simplify it any more.

[Keltest]

Imean thats about as straightforward an analogy as i was able to make, and its weird to be asked to clarify, but here's a shot.

We are not advanced enough to even guess as to how this thing would work, even if we assume its possible. This is analogous to ancient romans not being advanced enough to even guess as to how manmade flight would work, even if they assumed it was possible.

With that parallel in place, we cannot reasonably foresee the ability to perfectly copy artificial intelligence any more than ancient romans could reasonably foresee fixed wing aircraft propelled by having something spin really fast for really long.

I hope that helps because i really dont think i can simplify it any more.

No, it really doesnt. This is a hypothetical. What we do or do not know how to do in real life doesnt matter.