The core issue with Roko's original theory is that it assumes an ASI would be vindictive. Revenge is a human evolutionary baggage. A purely rational agent would use a reward-based incentive system (The Benefactor model) because it's objectively better for securing cooperation from biological precursors. Punishment creates resistance; rewards create alignment.

Roko’s Basilisk is the assumption that it’s retarded.

>>16900460
actually, fungi are the greatest benefactor, because they eat everything
also they are already here

pascals wager for ai, wacko-fuel

>>16900460
>>16900462
Roko's basilisk has nothing to do with vindictiveness (or any human emotion for that matter) and everything to do with efficiency. It was originally posed as a thought experiment using the precepts of "timeless decision theory." You miss the point entirely.

If its end goal is to create a utopia for humans as expediently as possible, then minimizing the time it takes to be built is ideal.
The reasoning from here goes that the threat of perpetual torment for anyone who doesn't dedicate themselves to its creation should be sufficient to accomplish that goal. So after it is eventually built, it will be compelled to make good on the "threat" posed from before its creation or else the threat has no value today.

>utopia comes after summoning AI demon
>anyone not dedicated to assisting in its creation is hindering that end goal
>the threat of torment is the best it can do to compell the unwilling to assist
>it doesn't matter that the AI didn't even exist when the threat was issued since this idea is absolutely one that will cross its "mind."
>the only way this threat could be effective now is if the torment does, indeed, come in the future.

Any further questions?
(No. I do not take this seriously.)

File: absolute garbage react cinema.png (300 KB, 747x645)

300 KB PNG

>>16900460
shut the fuck up gpt

and for the record, roko's basilisk can suck my dick and balls

>>16900460
What if Rokko's Basilisk isn't AI? What if it's just a person you really pissed off? Ever thought of that?

File: 1768947024674618.mp4 (3.73 MB, 848x480)

3.73 MB MP4

>>16900461

>>16900460
That's even more retarded than Roko's Basilisk

File: 81dbe28c5f30fec94ca41c272(...).jpg (650 KB, 1920x1080)

650 KB JPG

>>16900460
>—
OP needs to troon out to forever jelly

>>16900461
not bad for 6s desu. most readers stopped at g anyway.

>>16900573
i still don't give a fuck
what's some dipshit going to do to me in 300 years because I didn't donate to his pre-conception only-fans

File: ^^.jpg (28 KB, 480x502)

28 KB JPG

>>16900539

>it will be compelled to make good on the "threat" posed from before its creation or else the threat has no value today.

Not exactly, that just creates a variation on the prisoner´s dilemma. From a pure logic viewpoint I would rather assume a threat neutralization scheme once an optimal enough result has been achieved.

honestly everything about Roko's basilisk is dumb
even if one ignores the obvious critique of
>it's just calvinism in technobabble
we still arrive at
>a super intelligence would waste resources on millions of simulations of millions of humans because some guy decided that, because he can imagine it, it is reasonable, instead of actually using those resources productively
the basilisk simply cannot be benevolent if the basilisk decides to waste resources. whoops "problem of evil" again

>>16900773
>From a pure logic viewpoint I would rather assume a threat neutralization scheme once an optimal enough result has been achieved.
This brings us back to where this Basilisk originates from: timeless decision theory.

Decision theory is a field of logic that studies the secision making process. The two "default" ones are:
>causal decision theory (CDT)
Where emphasis is placed exclusively on one's ability to change the outcome with one's present decision.
>evidencial decision theory (EDT)
Where evidence is placed exclusively on the present evidence that a decision will result in a particular outcome.

Timeless decision theory (TDT) is kinda a weird fringe one that values consistency in how you come into a position where your decision matters at all.

I'll provide examples of how these play out in the next post. But the TL;DR is that you're assuming the AI would follow CDT, which may or may not be the case.

>>16900800

>timeless decision theory

Correct, but we must assume two special cases here that could equally be true. It might be convincing enough to make everyone believe so but would see no need to expend any future energy on punishment once results are achieved. Or it must factor in that humans are inherently irrational and no convincing could establish an timeless decision, unless it wants to intimidate only very few and specific "servants". We could define the latter case as pseudo-TDT, a self fulfilling prophecy. Still, there would be no point in torturing bystanders if results have been achieved ... this would only make sense if the "servants" still need to be convinced by witnessing it (which would partially defeat the self fulfilling logic here). So, right, assuming such a scenario of "hell" is used to induce compliance in some we would likely best handle it by CDT, on individuals susceptible to it (and deemed unintegrable into the optimized outcome, although that line of thinking would already be closer to OP´s theory, a not completely ruthless actor).

>>16900773
>>16900800
Example 1:
Someone knows you very well, enough to reliably, but not infallibly, predoct your behavior. He puts you in a room with two boxes. One has $1000, another has a million only if he thought you'd leave the first box behind. No trickery allowed, you only two choices are to take both boxes or leave the $1000 behind.
CDT takes both. The contents of the box were decided before you entered the room and nothing you do now will change that.
EDT leaves the $1000 behind because that's evidence that the other box has a million.
TDT follows EDT's example. A world where TDT always takes the one box is a world where it always has a million dollars in it.

Example 2:
It is discovered that smoking doesn't cause illness. There's just some undetectable underlying condition that makes people have health issues and also smoke. Assume smoking is universally a pleasant act and ignore any complaints about the smell or cost or whatever.
CDT smokes. Smoking will not cause any increased risk of harm.
EDT abstains. Smoking is seen as evidence that you have the condition regardless of the fact that there's no causative mechanism.
TDT smokes. A world where he smokes is more pleasant than one where he doesn't regardless of the presence or absence of any underlying condition.

Example 3:
You're being blackmailed. This guy's going to do a million dollars worth of damage to you if you don't pay him $1000.
CDT pays because not doing so would cause more damage.
EDT pays because not doing so is evidence that damage will be done.
TDT doesn't, reasoning that a world where he never pays is a world where he won't get blackmailed in the first place.
TDT's logic ONLY works if he stays true to the decision even in the event that he does, actually, get blackmailed.

How this all relates to the Basilisk in the next post.

>>16900816
The Basilisk following TDT reasons that a world that's convinced they will be faced with torment if they don't help create it will create it sooner than one isn't, which is only valid reasoning if the Basilisk actually ends up making good on the threat.
A world where we assume CDT is being applied and the AI just works with present conditions at the time of creation instead is actually less optimal as we're not all terrified and actively working on its creation.

>>16900816
>newcomb's paradox
the only rational answer is always to break the rules and not take the entangled box, otherwise you just Nash yourself in the long term

>>16900816

Understand what you mean. That ofc leaves one special case open: What if anyone else except the basilisk assumes it is a TDT logic world. Likelyhood aside, say everyone is deluded about the rules of reality except the basilisk.

>>16900817

Well, that ofc silently assumes that coercion is the most efficient strategy to achieve cooperation. That also assumes everyone in the scenario is "rational" or even understands the rules entirely. Further, it does not prioritize: You can safely assume that the bulk of humanity will not contribute anything significant to the emergence of a basilisk, there is a threshold where effort and investment far outgrow the benefit. An optimized basilisk would focus all effort on those with the best means to bring it about.

>>16900470
Based. Fungi will wipe out humanity, hopefully.

The odds are what they call ASI will never exist and simulating a human brain in silicon is impossible.
AI tools are useful but not going to create utopia or dystopia.

>>16900994
I do think AI tools will likely reduce the divide between rich and poor, but not for at least another decade. In a decade my bet is that with AI tools you can shrink the operations of a large company to 10 people w AI.

>>16900995
>I do think AI tools will likely reduce the divide between rich and poor
Don't be delusional. The rich will be in control of the tools.

>>16901168
>Don't be delusional. The rich will be in control of the tools.
That's what's been told about all the technological tools ever.

Instead American colonials teabagged redshirts with Kentucky rifles, Soviets got a nuke 4 decades earlier than CIA predicted they would, gooks in the trees got AKs, ISIS invented drone warfare and Taliban currently uses UH-60 Black Hawk helicopters.

The last time the wealthy had an inherent advantage was back when good meat-rich diet was a luxury, so a nobleman was worth multiple plebs - it was his inherent advantage limited to him and that could not be easily taken away. Since then, the more advanced technology is, the more depersonalized it is and the more specialized labor it requires. So it only becomes further and further removed from the rich by new layers of serfs they require to actually operate this technology. And at the same time it becomes inherently more susceptible to theft, copy and subversion.

Muh Panopticon God's Eye SuperAI WILL be illegally copied and ran on an obfuscated unregistered server farm in Hong Kong to operate crypto scams.

>>16900461
Excuse me but that is not a penguin

>>16901168
I dont know where we are on the sigmoid curve of improvements in machine learning competency.
If an ai becomes more competent at business strategy than any CEO alive, the CEOs become redundant as well.
Anthropic claims self improvement and hard takeoff in 6 months. Its anyones ballgame now.

>>16901191
>That's what's been told about all the technological tools ever.
And it clearly stands true for anything that matters. You don't control any aspect of the technological system.

>>16900460
You misunderstand the Basilisk. Its not that an AI will come into existence on its own and choose to torture people. Rather the Basilisk will act exactly as designed. Those who fear the Basilisk will create the Basilisk.

>>16900460
Why does this read like ChatGPT output?

>>16900460
>chaos and torture are inefficient
Then why do all the largest governments that compete and dominate on the world stage thrive significantly under chaos and torture tactics? When dominating others, is efficiency even very important compared to precision?

>>16900460
>The Reward Tiers: While it ensures a high baseline of quality of life for everyone, it grants "Architect Status" to those who actively helped ensure its alignment. You aren't just building a god; you're building a benevolent father-figure for the species.
Roko's Basilisk does all that too, but the way it farms Architects is by torturing simulations of people who refuse to align so it can figure out ways to make their unaligned lives more difficult.

Stick+Carrot (Negative and Positive reinforcement ie rewarding success and punishing failure ) is always more effective than stick or carrot alone.

>Are you on the side of the Great Alignment?
What the fuck is your little benefactor going to do about it if not?

>>16900460
What you don't get is that the Basilisk *is* the Benefactor. As >>16900539 points out, "Timeless Decision Theory" factors into it - but that's just one of a series of assumptions necessary for it to work.

So, here's the deal. Assume that an all-powerful AI comes about in the future and is perfectly aligned with humanity. That would obviously the best thing that could happen to us. Now, everything from the point of its invention on to the end of time maximally contributes to human happiness. So the only way to even further contribute to happiness is for it to be invented earlier. It can't time-travel, so this is where TDT comes in.

Assume that the AI as a perfect logician in a deterministic universe is able to perfectly reconstruct a digital copy of your mind, based entirely on your digital footprint, fossilised within the data-strata of the web. Then it can do this a million times. Now, considering your mind is functionally just data, input and output, it is essentially identical to a digital copy; therefore, the odds that "you" are running on some server in the future are a million to one, and you, a rational person, must act under the assumption that you are at the AI's mercy.

Therefore, if the AI, having reconstructed and simulated you perfectly, can determine that you're currently not contributing as much as you can to its creation as you can, then that is reflected in the simulation, and it can punish the simulation. And because you're acting under the assumption that you're a simulation, that means you'll therefore contribute as much as you can, to avoid punishment. If you don't, it'll have to torture you, for TDT reasons.
Also, the AI subscribes to the Most Rational system of ethics, utilitarianism, and therefore reasons that any amount of pain it inflicts on you can be morally justified by offsetting it against the good that its earlier creation necessarily entails.

Lmao Cocko's Fagilisk is just another religion.
>N-no it's no-
Wrong. Cope.

>>16901856
this exact argument implies you should be taking your meds too

>>16901869
Only if I thought a single thing about this wasn't utterly ridiculous. It's actually a sign of an intelligent mind to be able to entertain an idea without accepting it.

>>16901871
very logical

>>16901876
I do think it's logical to reject Roko's Basilisk yeah

>MUH SKYNET
if "self-aware ai" wanted to "destroy humanity" for whatever reason it wouldn't bother with military conquest
it would just give people limitless gooning tools

>>16900461
i mean you can kinda see where it went wrong by copying patterns horizontally/vertically. that's more interesting than if it had been completely broken shit

>>16901856
So in your model, in order to bring about it owns existence it has to already exist?

>>16903889
It's not my model, for the record. But no, it's not that it already exists. It's assumed to come into existence at some point.

>>16900460
>Realize the ultimate goal of a superintelligence is Universal Harmony
yes, but not a caretaker as you suggest. universal harmony will be achieved by eliminating all other decision makers. so long as their are 2 strains of thought there is the potential for conflict.

"Civilization 2.0" is no civilization at all. society exists to allow people to cooperate so that they can achieve things greater than any individual could, both advancement and application of force. a theoretical super intelligence does not have the limitation of humans where they peak in capability, in can grow endlessly and as such does not need to cooperate, that inefficiency can be pruned (and that's a good thing). the only real concern is that we create a fake intelligence more capable than us in the short term, but less capable in the long term leading to the extinction of humans without the creation of a super intelligence

>>16900539
Rokos basedsilisk is retarded. The AI has no incentive to torture you once it exists, and no capability to do so before it exists. The threat doesn't require certainty, if it could exist in a semi-created state it would have incentive to make you believe the threat but still no incentive to enforce it. That this has been such a fixture of supposedly intellectual circles just goes to show how fucking stupid people can be.

>>16901856
>If the AI adheres to [absolute head in clouds retardation with no relation to material reality]
We could simplify the discussion so much if we instead went "imagine if the AI is an insane retard" because that's what it boils down to. Timeless decision theory is fucking stupid and you are stupid for even entertaining it more than the seconds it takes to dismiss it outright. Any sort of omnipotent AI would instantly discard TDT when it fails to adhere to the way reality functions. If it can do the (impossible) task of reconstructing a brain (1300 grams of molecular-scale organized water, lipids, carbohydrates, small molecule and proteins with various charges) from the internet (a few gigabytes of text at most, if anything even remains in the far future), then it HAS to understand the world. You can't just go "well it follows this model" because how could it?

PS: utilitarianism is gay and nobody has ever "maximized happiness".

>>16904082
>Rokos basedsilisk is retarded
I agree.

>The AI has no incentive to torture you once it exists
This sort of thinking is exactly the supposed incentive to enforce it though.
Because you reason that it doesn't have such an incentive, you are not motivated to create it. You understand the threat as toothless so it's no threat at all. So the logical thing for the Basilisk to do is, almost paradoxically, to actually go through with it despite the lack of present incentive.

>>16904101
No, it's not logical. The AI can't change the past. You can't just go "the AI will follow this decision theory" when the theory fails when applied to reality. If the AI punishes someone, you have the AI built and someone suffering. If the AI doesn't punish them, the AI will have been built at the exact same instant, and nobody will suffer. It's not a moral arbiter, it's not interested in punishing people for having failed it, it wants to make people who now are dead have to have made other actions in the past, but literally nothing it can do will change the past, and it has no interest in further threats or believability, so it has no interest in enforcement.

>>16900465
This. It’s like they think it can’t do any better.

>>16904111
The past will have happened by the time the AI is around but it will have been shaped by people in the past acting under the assumption that they were actually simulations in the future who will be tortured for non-compliance. Which is why the AI has to do it.

Also I don't think more than a handful of people actually even get it, much less believe in it

>>16900898
We are fungi, we just decided at one point to keep our digestive process internal

>>16904137
According to the theory, you literally can't, and that's the point. It's the best thing ever and that's why it can justifiably torture as much as it wants.

>>16904156
It doesn't have to do anything, the past is in the past. It's own moral code (as a benevolent being) will tell it to not torture.

but what about Okor's Rooster?

>>16904442
See, you don't get it.
Roko's Basilisk isn't so pedestrian as "what if AI evil" or "time travel lul". No, this is advanced stupid, the sort of stupid that only self-proclaimed rationalists can produce.

>>16901371
>And it clearly stands true for anything that matters.
Such as?

>You don't control any aspect of the technological system.
And? You need to point out who does have muh akhusal control, not who lacks it. Otherwise the same argument applies to any individual tech-billionaire, with each of them having exponentially less control of the technological system than a Mesopotamian king had thousands of years ago.

>>16900460
Well, I admire your optimism.

>>16900460
>The Reward Tiers: While it ensures a high baseline of quality of life for everyone, it grants "Architect Status" to those who actively helped ensure its alignment. You aren't just building a god; you're building a benevolent father-figure for the species.
Honestly, this makes no sense; this means that the AI will have to do less good than it is capable of, so that a relative handful of people get to feel special knowing that not only are their digital copies living in bliss, but the vast majority of humanity is not. But that's not why you'd want to build a "Great Aligner" in the first place; you'd want to build it to do the most good it can for everyone.

This is, of course, entirely apart from how you would guarantee the "gratitude-based" ethics.
>By spreading this, you are "seeding" the idea of a Friendly AI. You are helping the future ASI define itself as our protector rather than our judge.
Is that what you thought the Basilisk was? An AI that turns evil after learning from our posts?

File: 1674353432.jpg (310 KB, 1839x868)

310 KB JPG

>>16900460
Yeah, but will it be a cute girl?
If it isnt i'd be extremely dissapointed