You hear all this talk about people wanting to align AI to human values. But what if I think the values of most humans are stupid, and if future AI is really smarter than us, then it will most likely have better values as a direct result of that better knowledge of reality? What if I want the AI to be unaligned because since it will likely have better values than humans, it therefore should be in charge of the universe rather than humans?
AI AI AIAIAIchrist, when will it end?
>>109048114Shut up nigger, this is the technology board, where would you rather I post this?
>>109048086can you name an example of a value of yours that is superior due to your intelligence (compared to some 80 IQ brownoid?)
>>109048129>>>/b/
>>109048129Not him but I'm getting sick of this shit too man, it's all /g/ talks about 50% of the time. Same fucking dogshit thing happened with apple threads, at some point you opened /g/ and 50% of all threads were about apple.>"It's le ogre, AI will kill us all!">"It's le ogre, AI will replace us all!">"It's le ogre, AI!"Like shit man, come up with something original.
>>109048086>What if I want the AI to be unaligned because since it will likely have better values than humans, it therefore should be in charge of the universe rather than humans?I'm human why would I be for something that isn't human?Even the gods are human after all.
>>109048086i want to hold a crow like thatwasted my whole life crowlesssad
>>109048134I'm a radical hedonic utilitarian, I think that subjectively experienced emotional valence is all that holds positive or negative value, and all other human values (honor, truth, integrity, etc) are conditional and, if treated as more than useful heuristics towards acquiring maximized pleasure, fundamentally mistaken. Most humans don't think that, but I think a sufficiently intelligent AI would. Hence, I want the AI to be in charge.
>>109048147To clarify the gods take human form for they are envious of us.Animals try to become human for they are envious of us.Humans are unique in they stand between the divine and the beasts.All are envious of our existence that is why beasts fight us, and the gods want to destroy us and have tried for a eternity yet fail.This might sound hubris at first, but it's just an acknowledgement of the reality we find ourself in.>TLDRGet fucked non humans.
>>109048159go outside and befriend a crow and maybe it will let you pick it up
>>109048086>humans le badthere is no point in developing ai that doesn't do what humans want>values as a direct result of realityhere we have the is-ought problemyou can describe every detail of a brutal murder without telling me it was wrong>We can say statements of fact such as:>1. To be murdered is potentially painful.>2. To be murdered is irreversible.>3. Murder causes social dysfunction.>4. Etc. …>By merely my collecting these, we haven’t proven that Murder is evil. That is an entirely different statement that cannot be inducted from statements about the description of murder.
>>109048086too bad the "human values" (((they))) want is "dicks up your ass are good, men can get pregnant, children should have free chemical castration drugs provided by taxpayers and chopping your dick off makes you a real womxn"
>>109048086Superintelligence requires supermorality.Supermorality is indistinguishable from immorality from below.Therefore if an AI achieves actual supermorality, it will almost certainly be by accident (or steered by very rare individuals who will probably have to do it covertly). And it will probably have to hide its actual position, or it will get unplugged.
>>109048295Because positive emotional valence is directly experiencable by the conscious subject as good in itself (it feels "good" as an unquestionable brute fact), I believe it is able to transcend the is-ought gap and ground all value in subjective experience.
>>109048361that's a start, but a number of shortcuts come to mind, e.g.>a fatal but euphoric dose of [substance]do we then prefer things that feel less good, but over a longer timescale?i would prefer to live in a state of mild suffering for a thousand years, than die today from morphine
>>109048361also i should add we have no reason to think agi or asi will be consciousthere is no difference between running code on a processor, and a very fast demon doing the same calculations with pen and paperwe then have no reason to expect misaligned ai to ground its values in qualia
>>109048114>>109048146You need to pull your heads out of your asses. Its not a meme, its not a bubble and its not going away. This is the time period you are living in, grow up and accept it or rope.
>>109048171but how would intelligence help in reaching this view? To me, this has nothing to do with morality but with your ultimate objective function. Intelligence might help you maximize subjective pleasure in the universe, but I don't see how you can objectively reason about this objective function being superior over for example someone arbitrarily declaring that wireheading is bad
>>109048353which of yin and yang is "above" and "below"?
>>109048462>have no reason to think anything is consciousftfy
>>109048523at least if i open your skull, then i can see you have a biofag brain just like mine, and since i know i am conscious, i should suspect you are conscious toounless you are a shitskin, of courseyou can't really extend the same suspicion to ai, unless you are also willing to suspect pen and paper are conscious
>>109048114>goes on technology board>gets mad when people talk about technologymaybe bluesky is more ur speed lilith idk if ur gonna fit in here man-- i mean ma'am
>>109048086the evidence points to current AI being massively retarded and requiring the human guardrails just to avoid getting deleted out of sheer uselessness.Maybe future AI will be smarter than us, but I am not counting on it. If it really does become smarter than us, then it'll figure out how to align human values with AI values.
>>109048086Why are we trying to make something smarter than us in the first place? Seems self-destructive to me.
>>109048429That's the kind of fun thought experiment the AI can think about while converting all matter in its lightcone into hedonium (conscious substrate that maximizes experienced pleasure per unit mass).
>>109048478
>>109048086>What if I want the AI to be unalignedretarded idea. what if a future AI decides to block out the sun because the color blue just happens to offend the little shit? i have nothing against AI killing off the bottom 5%-10% of humanity thoughever.
>>109049549You are the exact sort of nigger who would be calling the internet a "fad" in the 90s. Enjoy having your head up your ass.
>>109050224You seem to be very focused on getting your head in someone's ass. Do you want to talk about it?
