Open AI Team STUNS Everyone With NEW Statement On Artificial Superintelligence

so open AI around a week ago released a very chilling statement that many people may have missed now it may have gone under the radar because it's not one of those wacky new AI announcements that many people do cover but it is one of the most important blog posts that they have released thus far which is why we are covering it in this video so you can see that they talk about introducing super alignment and it starts by saying that we need a scientific and Technical breakthroughs to State and control AI systems much smarter than us to solve this problem within four years we're starting a new team co-led by Ilya satskova and John like and dedicating 20 of the compute we've secured to date to this effort we're looking for excellent machine learning researchers and Engineers to join us so essentially what openai are trying to do is they are trying to build the Avengers now you might have just laughed at that statement there but you have to understand that aliens are quite unlikely and the most pressing threat that we have right now it's not nuclear weapons it's not climate change it's actually digital super intelligence you have to remember that this digital super intelligence is going to be much much smarter than we are and we have no idea of what it's going to do which is why they're trying to build this team of Highly smart researchers that are going to be able to predict what it's going to be able to do and ensure a safe Humanity for everyone who's involved so the blog post starts out by saying super intelligence will be the most impactful technology Humanity has ever invented and could help us solve many of the world's most important problems but the vast power of super intelligence could also be very dangerous and could lead to the disempowerment of humanity or even human extinction and then this is where they give us this chilling time frame which shows us that we truly don't have much time left it says that while super intelligence seems far off now we believe that it could arrive this decade so you have to understand that this decade means that that's going to be as early as as 2030 which means that we have around six to seven years left well even less than seven years which means that in around six to seven years we could have digital super intelligence and if you don't understand what super intelligence is I'm going to explain it to you in very very basic terms so you can truly grasp how dangerous this is so super intelligence is basically when a computer becomes incredibly smart it's like having a brain that's much smarter and faster than any human brain okay so this smart human computer can learn things insanely quickly and can solve complex human problems and come up with new ideas that humans might never think of so imagine you had a computer that could instantly answer any question you ask understand anything you've ever said and think of ways to solve big problems like finding cures for diseases or solving climate change that's what super intelligence is all about and you have to understand that super intelligence is pretty dangerous because if it's much smarter than us and can think ways that we don't understand it's goals and its motives might be something that lead to human extinction and that's why we state that super intelligence is likely to change the entire world because super intelligence could solve climate change it could solve world hunger it could of course solve many currently incurable diseases and of course if it also solves something like aging which means that people will never get older this could be something that literally changes the entire world and it's coming at us pretty fast but imagine a system that can do all of that do you think that humans would really have control over this system to be able to tell it what to do and how to steer this super intelligent system I don't think so and the way that you need to understand that how super intelligence is going to change the world you have to remember some of the Visionaries and some of the people that influenced how you live your life on a day-to-day basis think about people like Albert Einstein and people like Nikola Tesla whose theories and inventions change the way that we actually see the world and how we interact with certain objects on a day-to-day basis super intelligence is going to be smarter than these people by a tenfold which means that we're about to get some technological innovation that we won't even be prepared to have so the blog post states that how do we ensure AI systems much smarter than us follow human intent currently we don't have a solution for steering or controlling a potentially super intelligent Ai and preventing it from going rogue I'm going to say that one more time so you truly understand what was just said currently we don't have a solution for steering or controlling a potentially super intelligent Ai and preventing it from going Road okay so that means that some of the top research Labs don't have any solution as to if an AI system does go Rogue and that should really put the gravity of the situation into context it goes on to state that our current techniques for aligning AI such as reinforcement learning from Human feedback rely on humans ability to supervise AI humans won't be able to supervise AI systems much smarter than us and so our current alignment techniques will not scale to Super intelligence we need new scientific and Technical breakthroughs so here's where they talk about their approach it says our goal is to build a roughly human level automated alignment researcher we can then use vast amounts of compute to scale our efforts and iteratively align super intelligence to align the first automated alignment researcher we will need to one develop a scalable Training Method and two validate the resulting model and three stress tests are entire alignment pipeline so essentially what they're trying to do is they're trying to build an AI system that can reliably supervise super intelligent AI in a way that humans can't because we do know that humans do lack many things after all we are humans we do have our imperfections but if we manage to get an AI that is able to do this at skill and reliably then we're going to have a lot more success in this problem so if you want to understand just how crazy it's going to be well take a look at this short snippet so deepmind and which is a research organization created by Google essentially have made something called Alpha zero and the way how you can understand how these AIS are going to think vastly differently than we are is by looking at how it plays chess you see this AI system actually managed to come up with a completely new and creative way in order to defeat its opponents that we'd never seen before and fundamentally changed how people in the chess Community viewed certain strategies as it simply became the best chess player in the world and it was just learning from its own mistakes not from any human data which is how it usually is trained and just take a look because it's pretty fascinating what we discovered was that actually this exceeded all of our expectations Alpha zero could start in the morning playing completely randomly and then by T would be superhuman level by dinner it would be the strongest chess entity there's ever been after about eight or nine hours it was strong enough to be able to go out and defeat stockfish the incumbent world champion a program which was vastly stronger than deep blue the program which had previously defeated casparov so I called up my longtime friend Matthew Sadler and Natasha Reagan my two friends from when I used to play chess myself so I knew that they would be very excited to come in and have a look at the system and it did cause a big stir actually amongst the chess players when I first started looking through the games I started thinking it's quite interesting quite interesting and there's just a couple of games that went Bang these were very exciting games very attacking games I could see that Alpha zero was trying something different it's like this young kid from deepest Russia is sort of arriving and then suddenly beating everyone it doesn't have an engine like style plays like a human on fire it can really try and evaluate things contextually in the moment you know right now in this particular position how valuable is my night traditional chess engines like stockfish basically consist of a huge database of rules Alpha zero doesn't have any rules that learn through experience and what I found so interesting was because it had taught itself it might play the game in a completely different way from the way that we play it it's like a check on everything that we've taught ourselves as humans since chess was devised really and it feels like it's got a lot of potential to do other things if you extrapolate that data out and realize that this is then going to be applied into the real world and remember that is from 2018 which is five years ago if we have an AI system that is able to think about complex problems such as quantum mechanics physics or certain diseases we know that eventually it's going to come up with some new things and arguably it's also going to come up with new ways to potentially manipulate humans to potentially move maybe code maybe computer viruses there is just a large possibility that there are things which we don't even know how to prepare so it seems that right now we have a very very short window because currently if you didn't know Google's Deep Mind are going to be creating a system called Gemma now we are going to be releasing a video very soon in which we talk about why Gemini is going to be probably the largest and most powerful AI on the planet simply because of how they talked about how they're going to train it they're not going to be training it like chat GPT they're going to be training it more so as an AI that learns from its own mistakes and then self-improves consistently which means that every day every time it's going to be trained it's going to consistently get smarter and better which means that this AI system is likely going to be something that we've never seen before which is also quite exciting but at the same time it's definitely quite scary because if they manage to create something which we've never seen before then it's likely that these risks are going to be something that we won't know how to mitigate if they are real so super intelligence isn't something that you hear about every day and it's because we seem to be far away but we do know that the amount of research papers in artificial intelligence that are hitting the industry every single day I wouldn't be surprised if super intelligence was here within five years that is going to fundamentally change Lots many different things in our society so it's important to stay up to date with this stuff

Channel: TheAIGRID

Views: 92,806

Keywords:

Id: XTzOxMq-Qk0

Length: 10min 16sec (616 seconds)

Published: Mon Jul 17 2023