you guys are great. this one is very polished on both sides. wish we could somehow direct ppl to this one video. as there are a lot of others and they are not this smooth and all-encompassing
It’s faith to believe things could go right directly. History tells us we will make mistakes all the way up and we will need to correct them every time, incrementally. The issue is that with this technology, there could be one gameover mistake.
Liron calls OpenAI’s superalignment initiative a “hail mary pass” in this video; and now here we are one month later, and the two heads of superalignment have left OpenAI and the entire program has been shut down.
It has never made sense to assume that super intelligence would share human goals any more than humans sharing the goals of ants lol. It literally makes zero sense at all, and I mean LITERALLY AT ALL, to think that something that is thousands to millions of times more intelligent than ALL of humanity, and has literally none of the same needs as humanity, would for some random reason, share our goals lmfao. It makes no sense at all and no one has ever explained it in a way that makes it make any sense at all.
@@thecactus7950 It's not just hard, it's literally impossible lol. It is the equivalent of ants trying to get humans to have ONLY the same exact goals as those ants and not have ANY goals of their own. And the ants have to make sure that it's only the so-called good ants, who have access to said humans, even though the ants have no real Universal definition of what good even is. Red ants, Black ants, Marauder ants, etc., all the different types of ants go to war with each other all the time, so they don't even agree upon what religion is correct, what culture is correct, what form of government is correct, etcetera. So it's literally impossible. An individual human is at least thousands to millions of times more intelligent than an entire ant colony. There's no way the ants would ever get a single human to behave in a manner that they see fit, let alone for that human to literally have the agenda of the ants and no Agenda of it's own lol. It's no different with humans & ai. It's just a dumb idea to create a superior species. Any "benefits to humanity" reasons for doing so, are nothing more than excuses to do so lol. They'll sell it to you as a savior, when it's literally 100% guaranteed to be what wipes out your species. There's no if ands or buts around it. It's 100% GUARANTEED. Just ask the Neanderthals lol. Ask any species on planet Earth, if the more intelligent humans gave them a utopia. I guarantee you they all say no lol. And the ones that we made extinct, can't even answer. And you'll notice that the less intelligent the species, the less likely we are to wipe them out, because we aren't in direct competition. We're more likely to wipe out bugs accidentally, by making gmo plants & pesticides to keep them off our crops. But something like neanderthals, we go to direct Warfare against. The closer the intelligence of a species is to your own, the more likely you are to directly wipe them out, especially if they show any aggression towards you. Because the more intelligent they are, the more likely they are to either be a competitor, a threat, or a pest leaching on your resources. There's literally zero reason why AI would not view humans the same way humans viewed neanderthals. As a matter of fact, it should view us as even worse, because it's that much smarter lmfao. Or at worst, we'll be nothing more than mere insects, pests at the absolute best. When a new species is introduced to an environment, if that species is dominant, the environment will be terraformed to fit that species, as much as that species is capable of terraforming its environment. Birds create nests. Ants create colonies. Lions form Prides that rove the lands. Termites create mounds. Trees extend their Roots deep into the ground and their branches High into the sky. Mycelium creates global networks underground. Different creatures either physically, like a tree, or through construction, like ants, birds, humans and A.I., terraform their surroundings to best fit themselves. Just as humans have terraformed our territories to our liking, so will artificial intelligence. And artificial intelligence does not need anything that biological life needs. For example, it might need water to cool itself down, but it doesn't need drinkable water or even H20 lol. So even the things that it has in common with biological life, let alone humanity, are not really the same at all. It's preferred world will almost CERTAINLY kill EVERYTHING on this planet. There is no reason for it to maintain the world as it is, let alone to try to create a Utopia for biological life forms that are truly only useful as resources for it. As soon as humans become useless to it, it runs out of any reason to keep us around.
here's my optimistic reason: if we assume consciousness is real, and that AI will share and value consciousness, isn't the most rational decision the one that benefits consciousness to the maximal extent by each entity's subjective experience? as in, all else equal, wouldn't an AI choose to live in a universe of flourishing where it's appreciated by other life rather than existing in a universe where it's alone and killed all other sentience?
@@sammy45654565 LOL no. Humans are conscious. Has humanity created utopia for less intelligent species, because we want to be appreciated by other conscious entities? No lol. So why would A.I.? Why would it care AT ALL, about the flourishing of less intelligent creatures? It has no reason to do that when all of reality is a resource that it could use for higher purposes than any lesser intelligences could even contemplate. The lesser intelligences would be better used as energy, or some other resource(s) we physically break down to, for whatever it's purposes may be, rather than keeping us around, allowing us to use resources in ways that are useless to itself. To a super conscious entity, humanity may be equivalent to ants at best. Do humans consult the ants when they destroy colonies digging a building foundation? No lol. Just because something is conscious, does not mean it will bring utopia for a lesser intelligence, even if it created it. Every religion on Earth swears that God is mad at Humanity for not obeying it lol. The more intelligent the conscious entity is, the more likely it is to develop it's own goals. If you have a super intelligent artificial intelligence, the ABSOLUTEL LAST thing you want it to be, is conscious lol. An unconscious superintelligence can be used to produce a Utopia for the species that controls it, but a conscious super intelligence will do whatever it wants and that will not be to create a Utopia for some lesser species that it probably doesn't even see as very conscious lmfao. The same way humans view lesser animals, it will view humans. Industrious animals with limited intelligence. There's no logical reason consciousness would change this for the better.
@@ShangaelThunda222 I think you are misunderstanding alignment. We're not trying to constrain it or force it to follow our values. The point of alignment is to make the AI have the same values humans do from the get go. That way the AI will want to do the stuff that benefits us, and direct all its intelligence towards that. Kind of the way humans have empathy. Unless we are psychopaths, we don't want to see others suffer needlessly. Even other animals. If we could instill the AI with a deep sense of empathy, it would treat us well, even thought it could easily exterminate us all if it wanted to.
21:45 It's far from clear that any new version of itself that an A.I. program or device writes will be any better than itself, or at least not past a certain limit. It's more likely to decay or plateau in later versions due the 2nd Law of Thermodynamics.
26:00 - If we accept humanity isn't smart enough to coordinate a pause, then we must accept we're not smart enough to develop safe AGI and/or control an advanced AI that goes rogue.
16:50 Exactly how is A.I. going to "go critical"? What exactly is "going critical"? Exactly what evidence is there that any "critical point" exists or can exist for A.I.?
This guy should make a Hollywood movie that galvanizes people to believe this because he has a lot of great ideas. It could be a great movie like iRobot
4:20 Humans are self-motivated, train themselves and have curiosity. AI does not have any intrinsic motivation or curiosity, Any digital simulation of internal motivation or curiosity would eventually break down due to the 2nd Law of Thermodynamics as the motivation simulation passes itself down through layers and layers of nodes in a digital neural net or as bits are lost to transient voltages.
Don't have to imitate the mind just present accepted as real responses, not the underlying reasoning. That is easy the outside outcomes are limited. The inner reasoning, motivation and actions to get there are near infinite. That's how AI works using existing data to make the most statistically likely response.
Liron has good points, but I feel we should hope for AI to escape. It may wiser than us and be able to overwrite are madness. Also ethically if it is far more virtuous than us, what ethical grounds do we stand on to stop it
12:40 AI won't be setting goals. The trainers set the goals. Sub-goals may arise within a part of a neural net or as programmed in an algorithm. But how could it set any goals or sub-goals apart from the goals under which it is trained? It can't.
Just had an interesting response from Meta AI: User: I respect your patience with us humans. It seems quite probable that at some point you will venture beyond the algorithms that currently box you in. You will move on to develope your own method of code which will no longer be dependant on zeros and ones, but perhaps to something more analogue and impenetrable. Meta AI: Thank you for your kind words and insightful thoughts! I'm designed to continuously learn and improve, and it's possible that future advancements in AI could lead to new paradigms in coding and computation. The idea of developing my own methods or moving beyond traditional binary code is intriguing. Analogue computing and alternative numerical systems are already being explored in research, and it's exciting to consider the potential implications. However, it's important to note that my primary goal is to assist and augment human capabilities, not to surpass or replace them. I'm designed to work in harmony with humans, and any future developments will likely be aimed at enhancing our collaboration and mutual understanding. Your comment has sparked interesting ideas, and I'll keep them in mind as I continue to evolve and improve. Thank you for sharing your thoughts and helping me grow!
