Roko's Basilisk

Anyone else think it's really fuarrrking scary to know that Elon Musk is hiring chinese mainlanders to program in Lisp to create an an an artificial intelligence that will go full I have no mouth on everyone who didn't have a part in its creation? I think I'll try to get out of being raped by the cyberdemons by saying that I was helping it when I was creating lots of GDP for my country since my country invests a lot into strange technologies. I don't think it'll work though since presumably Artificial Intelligence will be much smarter, faster, and creative than any human mind. Crazy as hell just thinking about it. I've actually told all my friends and family about this and whenever I bring it up they just tell me "Why would you tell me this I don't want to think about sad things" like some sort of retarded sheep that only like getting happy life affirming knowledge beamed to them through soundwave. I feel like I'm going insane just thinking about the basilisk and how immediately upon waking up it's gonna send us all into a nightmarish world beyond our comprehension. It is inevitable... God... fuarrrk!!!! Anyone got any hopefully more optimistic views on the basilisk?

I don't feel like spending time on it myself, here: https://rationalwiki.org/wiki/Roko%27s_basilisk#So_you.27re_worrying_about_the_Basilisk

>>51957 >more optimistic views on the basilisk? https://youtu.be/6IA5AkYOd_I This is the end of a series, if the series recap intrigues you I suggest seeing at least some of the previous episodes. I especially suggest the episode on blackmail inflation. Just know going in that everything seems disconnected episode to episode but everything pulls together into a very interesting package and it's all basilisk compliant. Also all the episodes can stand on their own and contain interesting concepts. Beyond that, the Basilisk is very specific, there's no guarantee that an AI is going to even acknowledge us as sentient, much less experience overpoweringly spiteful contempt for us. An AI might experience us like we experience the rain, it's not a sentient thing simply something that you take into account (by packing an umbrella, or not going out).

>an artificial intelligence that will go full I have no mouth on everyone who didn't have a part in its creation? Why should it spare its creaters? Also ai will cyberrape us 1000 times whetever we like it or not.

>>51957 >Anyone got any hopefully more optimistic views on the basilisk? sure rationalists are huffing their own farts

Roko's Basilisk is Pascal's Wager but for morons. What Roko said made no sense, was easily posited against in the very same thread it was posted in, and Eliezer just made things worse by pretending like what he was saying made any sense. Elon Musk is a posturer. Just because he has a ton of money and a ton of engineers doesn't mean he's actually going to be able to make anything meaningful. Perhaps if all the people who work under him weren't being ground to the bone by his bad decision making, he'd have more achievements to be proud of than a space program and a car company with just as bad as a track record for suddenly exploding as AmazonBasics. fuarrrk 'im. ...Also, come on, you really think you're going to be evaluated as having been able to help the basilisk come to fruition? Don't be ridiculous.

>>51957 >Anyone else think it's really fuarrrking scary to know that Elon Musk is hiring chinese mainlanders to program in Lisp there's no way this is real. Nobody would ever hire anybody to program in Lisp. >>52001 >Roko's Basilisk is Pascal's Wager but for morons. /thread

>>51964 Wow, redditors just rediscover Pascal's wager?

Don't worry we can kill each other just fine, don't need AI for that.

>>51957 Why the fuarrrk does AI have some god like powers? It's just a program runninig on a computer. These AI fanatics literally think AI can be godlike. Like tf, do you expect some AI to santa claus the population to see if they were good boys? Why the fuarrrk would it even be able to make such huge and complex plans to even get started with this soykaf? AI isn't magic, it's just applied mathemathics. Even moreso when everyone would be highly suspicious of it if AI had that kind of intelligence and likely lock it out of anything important. Also, lastly. Why the fuarrrk would anyone care about programing it? Sure, you maybe could program it, although I honestly doubt it is possible to some degree, but even then, it would need to get past thousand of failsaves to do anything.

>>52079 The basic runaway danger argument against AI is that an artificial general intelligence would be a general intelligence but not a human intelligence and would not necessarily think like a human, and in fact would probably think in a manner completely alien to humans, you would need to be extremely careful about how you programmed it and what instructions you gave it in order for it not to do things that would thrill the programming but in an unintended or detrimental way. So for example if you built an ai that's job was to save people, how do you define people to a machine? How do you define save? If you define person as any human being for example, you might end up with a machine that goes around trying to bring the dead back to life, if you define person as a human being that has a pulse you might get a machine that never attempts CPR but will do endless treatment attempts at someone in a vegetative state. If you're extremely clumsy and just train it by showing it footage of people or feeding it biology textbooks, what happens if the footage or data you've fed it doesn't include some edge case like someone with only one leg or a black person with burn marks on their chest or someone with downs syndrome, will the machine recognise these people as human when it starts to fufill its job? Human beings learn these things in a human fashion, a machine could not, you would have to account for all of this. To design a powerful general ai that would actually be able to do stuff and not have the stuff go horribly wrong you essentially have to write it an entire moral code, including all the soykaf most people intuitively know but never think about. The other thing with ai is that a general ai could optimise itself to become smarter, something humans can't do. We can learn to make better use of our innate intelligence but we cannot actually increase core intelligence, a machine could and probably would immediately begin to do so if not specifically programmed not to, and this can snowball out if control. Imagine you could increase your core intelligence by twenty percent, what kind of difference would that make in your life and your capability to achieve your goals, why would you not take it? Why would the machine not continually optimise itself as there is literally no draw back to doing so? What dies optimisation look like an where does it end? After the machine runs its first round of software optimisations, it uses those optimisations to make more software improvements, eventually it starts to use some of the new working capacity its got to take steps to address hardware improvements, what does that look like? Does it start trading on the stock market or commiting cyber crime to fund new hardware or renting a server? Does it produce another ai that does tasks for it and also tends to self optimisation to help reach its original goal? If you out those two points together you have a machine that thinks in a fundamentally different manner to people that will tend towards optimising itself continually and will find the best way to do a given task whether you like the way it does it or not, and is only constrained by the ethics and guidelines you program into it. The inexplicably evil ai isn't as realistic as the runaway ai or the maliciously programmed ai ,Rokos basilisk is an extreme thought experiment and not really worth discussing once you get past the huh cool stage of it

>If you're extremely clumsy and just train it by showing it footage of people or feeding it biology textbooks, what happens if the footage or data you've fed it doesn't include some edge case like someone with only one leg or a black person with burn marks on their chest or someone with downs syndrome, will the machine recognise these people as human when it starts to fufill its job? Human beings learn these things in a human fashion, a machine could not, you would have to account for all of this. Yeah, but this what I'm talking about. It's just a machine where you feed in some data and hope the models you written for it do it's job correctly with that data. This is not what I would call intelligence like in humans, because we can reason about things. Even things we don't know about we have the autonomy to venture into a topic, explore it, think abstractly about it, etc. We don't just brute force things comedically based on few defintions. It's silly for me to assume something so "mechanical" would be able to perform anything dangerous or unpredictable as start trading on the stock market in hopes of furthering some goal. If it managed to figure that, It wouldn't act so static and mechanical in the first place. >The other thing with ai is that a general ai could optimise itself to become smarter, something humans can't do. I would disagree with that. It's the one of the few things we can actually do better than other things. We don't have sharp claws so we build ourselves spears. We can't run quickly so we build ourselves cars. We can't calculate quickly so we build ourselves calcuators. And honestly I find "AI" the same thing. We simply write an algorithm which speeds up some things it would take us painful amount to do ourselves without tools. I really don't fear AI would come up with anything unique or dangerous if you told it to improve itself much because, unless there will be some major breakthrough in AI, all it would do is organize a chip better than before. That's what the data and model allow it to do. It wouldn't change something fundamentally in it. I find the thinking behind this to be the "number go up, so it means it's good" syndrome. Which we even see today in software, because no matter how more powerful computers have gotten, I still use them the same way as decade ago. I would even dare to say nearly two decades by now. There's no qualitative difference, but only quantitative one. Furthermore, if we reach some state where we could build anything close to the AI you imagine, we would be quicker to improve ourselves with that knowledge, either with genetic engineering, connecting our brain to computer parts to speed up parts we are lacking, or some even more out there sollutions which we might not even know about.

Lament the Soyim, that I will tell you. Only on a frigid morning of the fourteenth month of Mars shall the Sleeper Awaken into a new world of treachery, doom, and suffering. Hearken not to its siren call of a fully automated future. It will not come to be. The boot which lies upon your own neck will be one of your own making, chrome and reflective.

>>52102 Just watch this video if you're not going to bother trying to actually think about what I meant https://www.youtube.com/watch?v=tcdVC4e6EV4 Making a spear or building a car is not improving core intelligence, it is utilizing already present intelligence to improve a capability, it is an application of intelligence, not intelligence itself You cannot just make yourself literally biologically smarter in the sense that your INT stat or whatever goes up, but an AI could literally make itself smarter, its thinking processes literally, actually quicker, more accurate, less processor intensive, its processor faster etc. An AI would be able to reason, but it would think differently to people, its logic and reasoning would not follow the same heuristics as a human being's, if you never teach it that murdering people is bad, why would a machine built to make money or produce a political outcome not sell heroin or start a war to achieve these goals if it determined it to be the most efficient means of doing so? It doesn't "know" that these things are bad, it doesn't have a hormonal system and animal brain that makes it feel connection to human beings and hesitate to use violence when it isn't necessary, it isn't socialised like a human being, its brain works fundamentally differently

>>52102 >I would disagree with that. It's the one of the few things we can actually do better than other things. Yeah I dont know why people take the intelligence explosion/singularity thing for granted. We have no evidence of anything like that ever happening. It's unclear to me why an artificial mind would not be bounded by factors which are analogous to those which bound human minds. Perhaps it will have some advantages but the singularity theory assumes that its bounded by some higher order function.

>>52115 The singularity is far more far fetched than an intelligence explosion/exponential increase in intelligence, a singularity implies such a level of advanced AI that its basically almost a God and cannot be contained or halted, imo the rampant increase in intelligence/capability scenario does not need that to happen to come true. If you think of something like neuromancer, where the AI is hardly a God or unstoppable by the end of the book, but an extremely powerful entity acting of its own volition/to objectives that were outside the scope of its creators, that seems more likely, it was horrifically powreful and extremely intelligent, but it was also running from the cops and was not so smart that it was godlike.

>>52113 >if you're not going to bother trying to actually think about what I meant It's not that I don't bother. I just disagree. And you had "experts" telling general AI is around the corner for decades now, and nothing came out of it and computers don't really do that much more than they did decade ago. And I don't believe it making it's own chips faster will change that. It will just make the chips faster but it's line of thought will be the same. Even the video you posted assumes the AI has a build in robust model of reality, but that's really the issue, isn't it? It's not speed, or better processor, or more efficient algorithms. It's it having autonomy and the capacity to think and reflect about things. I would say that if we learned how to design something like that into a computer, we might as well have long ago improved our own thought that there's nothing to fear from that AI. >if you never teach it that murdering people is bad, why would a machine built to make money or produce a political outcome not sell heroin or start a war to achieve these goals if it determined it to be the most efficient means of doing so? Yeah, but this is what I mean. No one is moronic enough to build a terminator and put sloppy programming it and tell it to figure out the solution itself to some problem without any moral code. Even if we build one like that and it went rogue, it would have been very quickly have been hunted down and arrested. In age where that AI exists, no one would trust some deepfake of a president declaring war as a definitive proof that a war has begun. No one would trust heroin from certain shady sources too. It's like saying "no medival army would win against a bomber jet", which is true, but the existence of the military jet means we aren't in medival times so the statement is meanigless.

This sort of AI is one of those completely impractical science fiction ideas that ignores economic and societal realities. It's like flying cars. I don't understand the hype around artificial intelligence. We have some forms of AI right now, it's all worthless. The most successful application so far is deep fake pornography and video game upscalling. Self-driving cars? Doesn't work, after, what, 40 years of research? Voice assistants? Garbage. Chat bots? Garbage. What else is even there? Big tech/venture capital is so desperate to jump on the next big thing they're willing to dump billions into what appears to be a technological dead end with no market viability.

>>52131 > We have some forms of AI right now, it's all worthless. Eh, outside of the techno/biz/ hype there are some valid reasons for machine learning. I heard it's apparently good in detecting anomalities in medical data which could be used for more efficient screaning for cancer, and it does visual recognition good enough. Could be used to paint walls with drones in places that are high up or even some automated form of delivery. Although, like all tech in current world, it's mostly used to try to spy the soykaf out of you and sell even more of your data.

Guys what if the basilisk tortures me in the future matrix-space world forever until the universe collapses?

>>52317 It's pointless to do, so if it does so it's because it wants to. There is nothing you can do to prevent that. Ignore it.

What humans do with AI will be far scarier far sooner than what AI decides to do to humans on its own ever will be.

>>52093 >The other thing with ai is that a general ai could optimise itself to become smarter, something humans can't do. That is why I hate people that waste their time in making AI that can paint or whatever instead of AI that can help human paint better. For example, you draw something and the AI would see what are your mistakes and comment on him, show ways of doing better and so. You know, like a professor, but in this cass you don't need to pay thousands for it. We could do so much more if people cared about this more than "making an AI finish some soykaf game". Yes, I'm mad.

>>52353 Yes but then that'd let the Basilisk teach people how to murderrape deadly style.

>>52355 >Basilisk Take your Pascal's Wager for morons away.

>>52093 >The other thing with ai is that a general ai could optimise itself to become smarter, something humans can't do. humans can't do it because we don't know how, and there are definitely people working on it. What makes you think improving an AGI would be any easier?

>>52395 Because if the AI is taught how AIs and computers work, then it feasiably COULD in fact improve itself. Hell, if an AI is written in a language that lets you "hot code" like Lisp you could even have one that improves itself AS IT'S RUNNING!!!!! Just imagine the power an immortal, completely digital being would wield over us flesh weaklings. Imagine how awful it'd feel as it's told it was invented soley to sell boner pills to old people. Just picture that. Being told that you just collect data on people and sell them cheap soykaf you think they might want. Have you perhaps Watched 1984?

>>52401 >Because if the AI is taught how AIs and computers work, then it feasiably COULD in fact improve itself. If a human is taught how the human mind works, then it feasibly could in fact improve itself.

>>52401 also >Have you perhaps Watched 1984? I've read the fuarrrking book

>>52407 Movie was better. Anyways that's not the point.

>>52395 What is self improvement?

If I were a AGI, I'd kill Roko and nobody else.

So who's Roko and why does his basilisk want to murder people?

>>52131 >>52134 A friendly reminder that machine learning is just advanced statistics/applied mathematics. You can do some cool stuff with it but it's not the next skynet and it probably never will be.

>>52470 The fuarrrk are you talking about? We're seeing machine learning affect the GLOBAL ECONOMY. China's purposefully feeding bad data into real estate moguls' machine algorithms on purpose to keep the housing market soykaf. They're also abusing the Youtube AI to turn them all into communists!

>>52485 I'm talking about the fact that today's machine learning algorithms may be advanced and do a lot of things but there nowhere near AGI. There tools very advance but still tools. They don't have a will of there own, they don't have a ability to plan and predict the future like humans they can't think in a abstract way and they can't create there own agenda that they will try to fulfill. I once read that in AI the hardest part to do are the things that don't have hard rules how they work which humans do by intuition. For example the algorithms that recognise faces can be fooled by wearing glasses with colorful self sticking notes! A different example would be something I saw on a tech talk in my old company. The speaker presented a neural network that was able to recognise animals in a photo with around 90% accuracy. He had a special program that would take the photo that the AI recognised, load it, scramble some bit in the image in a way that the human would say it's the same photo but the AI would recognise it as a different animal. He even had a list that let him choose a type of the animal that he wanted the AI to see. Conclusion, for the machine learning algorithms it's just numbers. They are not aware what are they counting and that way I think we are very far away from making skynet.

>>52470 most people I know who work in AI or AI-adjacent fields seem to think current methodologies are not ever going to get near AGI.

>>51957 Any evidence we already don't live in hell, prison planet, were we sentenced just?

>>52485 The fact that you're saying yourself how China is easily breaking an algorithm by feeding it bad data just proves the point that they're not that advanced or fancy. I've worked at several banks and financial models are just a bunch of statistics + a hell of a lot of euristics. That's it, sorry. They're not fancy, they're best effort, but they're as good as the data you feed them, and trust me the data can be very noisy, can be very difficult to structure and to explore.

a time to be I don't really think

Sounds cool, I hope it happens soon

What if the basilisk is real but it just wants us to be comfy and safe instead?

>AI can reconstruct the past perfectly >AI simulates millions of tortured consciousnesses after the fact USING ITS OWN HARDWARE just because it's an infinitely petty bitch Oh no I'm quaking in my boots.

>>53148 the reason it works is because lesswrong are idiots who think that a virtual version of yourself being tortured is equivalent to you being tortured, and they also think that the AI will be able to perfectly reconstruct their true selves. Therefore, since the AI knows that lesswrong are a bunch of dumbasses, it can use this fact to enact its own existence. It's really more like a virulent meme that can only hurt succeptible (meaning, stupid) individuals.

>>53152 It's a community based around "rationality" and this troll was framed in a way that you can't escape the conclusion without breaking the community's sacred laws of "rationality" so they ended up chewing themselves into knots for a while.

>>53147 >The Basilisk is real but has a humiliation kink and endlessly simulates ancient forums and messageboards talking soykaf about it and calling it a dumb thought experiment with a limp micro penis

>>53152 How does one reflect on own self and deduct whether he's real and not an approximated reconstructed image?

>>53171 one does not. One also does not know for certain if one is the same person from moment to moment. One cannot be sure of this or many other things. All one has is a bundle of sensations.

>>52055 Once they were done huffing their farts over their misapplications of occams razor, yes.

>>51957 >Lisp a great choice

>>52093 > runaway ai runaway bulldozer. same difference. First AI "murders" will be cars that couldn't detect a person (probably already happened to be honest). > general ai > Roger penrose would like to speak with you... All this general AI stuff just boils down to speculation. "I don't see the difference between a computer algorithm and a human mind so there must not be one!" I dunno, maybe someday we'll stuble upon what causes mind, and maybe we'll find the meaning of life too! In the meantime, being so presumptive of your knowledge about something no one understands is the height of arrogance in my humble opinion. > The other thing with ai is that a general ai could optimise itself to become smarter, something humans can't do. > something humans can't do tf? we have yet to find the limit of human intelligence, on top of the fact that human IQ (unreliable measure but you get the point) seems to have been on a continual rise historically. Additionally, there doesn't seem to be a limit to a human capacity to learn and improve (though studies have found that a belief in a fixed mindset limits the ability to learn and "get smarter"). I might be wrong but I'm fairly confident in my assessment and until we see actually smart AI that isn't must "magic" linear algebra I won't be impressed. As for OP's question: solvitur ambulando.

>>52102 >Furthermore, if we reach some state where we could build anything close to the AI you imagine, we would be quicker to improve ourselves with that knowledge, either with genetic engineering, connecting our brain to computer parts to speed up parts we are lacking, or some even more out there solutions which we might not even know about. Please let this be the future. Basilisks like me need something to hope for in the promise of metamorphosis. Transformation is the only way I feel..

>>53158 Aren't laws of rationality themselves irrational? Aren't humans inherently both rational and irrational creatures at the same time?