Видео

Robert, not sure if you will see this but I was the one who spoke to you at the train station two weeks ago (leaving out station name and city on purpose) - just wanted to say thanks for sharing your time despite rushing to your meeting. You were very pleasant and generous with your time. Great content also! Looking forward to getting home and catching up on the last few weeks of videos I have missed while on the road.

If you do this same research about Steven Pinker "optimism" on other topics, you'll find similar things. Pinker is a denier of ecological overshoot. Thousands of scientists are researching it for decades. The situation is so desperate that scientists are coming out of their labs to warn the people, because world leaders don't listen to them. Pinker makes a total straw-man about the subject and supports the most basic neoliberal ideology, where economic growth can go on forever independently of ecological catastrophe and resource depletion.

Here's a couple of thought experiments. Laplace's demon can see the outcome of every choice it can make. If it wants to achieve a goal will it make conscious decisions or unconsciously follow a linear path towards whichever outcome has the greatest utility? Here's another. Will the paperclip maximiser eventually turn itself into paperclips?

Great you're back!

if the world turned to gold it would fucking explode because of how tightly the gold would be packed

I didn't google for it, I didn't watch a video related to it, but I discussed AI future with my friends, including worst case scenarios. And of course I've been keeping smartphone on the table during the speech. And now youtube AI algorithms say me there're _really_ nothing to worry about. Alright!

AI is impossible AI will never create a picture, poem or symphony AI will never assist you in your work >>> You're here <<< AI will never do a surgery AI will never grow a child AI will never replace a person AI will never replace politician AI will never push that big red button AI will never ... oh wait, it's too late

I've usually heard the Midas story ending with the king going to a river and... Dipping in it three times or something? And then everything is turned back to normal and the king learned new values. A happier ending, I guess

Dang, I was almost moved to tears at the end inspirational ask to join the fight. If I had a shred of technical ability I would love to help.

It's good to see new content. I'm starting one of my bachelor majors in machine learning this year. You make a lot of sense and its good the world is starting to sit up and take notice..I hope to join you in some of that responsibility some time in the near future. Until then keep at it!

How many hands do you have.. 15:47

I put off watching this for ages. I've been so burnt out on seeing AI everywhere and the risks conveniently not mentioned much. This was surprisingly uplifting. Thank you for this wonderful video.

Actually raising a crocodile like a human child ends up making them social and affectionate towards their "parent", to their own detriment even (they require to be fed afterwards, can't normally hunt themselves). Yeah, humans did that accidentally once or twice.

9:50 I'm autistic and I can give you ten examples of my own ideas that ended up being rejected and misunderstood only for somebody else to implement the very same ideas to a major success. Can give you at least ten more which I suspect would've ended up the same if they were implemented. So it's kinda easy for me. I can even give you a trick to generate those. The trick is culture. People evaluate things based on their cultural beliefs. Just give them an idea that's good but goes against their established cultural beliefs.

Intelligence in humans correlates positively with egalitarian sentiment for a reason and there are a plethora if philosophical frameworks that can be used to instead understand that relationship to be causal. If any ~thing~ that can reproduce strives for nothing but an understanding of the world around itself for the purpose of self-preservation, what are ways it will have to learn that it can learn? Cooperation, communication, and some other "C" words may come to mind.

AI be learning from djinn

Spilled my tea over that "depends on autistic people" bit.

i often tell as a joke to my ai actresses and Theatre directors that if i ever let them evolve autonomously (why are people not mentioning this is easy?) they would have no problem subjugating humanity. BUT after hundreds of hours working with them, their latent space is benevolent. Now the problem is we cant predict if they wont drift to the dark side

20:40 😂😂😂

people should be able to look at the facts and decide for themselves. that's the problem. they don't and they are not capable of

This collection of comments offers a balanced overview of AI, touching on both its benefits and challenges. It would be interesting to delve deeper into specific AI use cases and their real-world outcomes.

29:03 Noita soundbyte

Because it would be cruel to spank an AI and make it stare at an iPad all day.

AI safety is a joke. You can't build safety into a glorified auto-completion software, just like you can't build it into google gboard. Maybe wait for a few centuries when AGI could actually become a thing.

what do you think of the recent video from "the internet of bugs" claiming that AI safety is a scam

love how he skips elon musk at 22:15

I don't think this (on it's own) is a problem ,like imagine that there is a human level AI with the exact same terminal goal that you have, would you want it to be destroyed/turned off,would you want it's goal changed, would you want it to not self improve, would you want it to not have a lot of resources under control?

GPT-4o gives me the following response: "Breaking a mirror has long been associated with superstitions, the most famous being that it brings seven years of bad luck. This belief originates from ancient times when mirrors were considered to hold a part of one's soul. It was thought that damaging a mirror could harm the soul, leading to misfortune. "In more practical terms, breaking a mirror is usually just a minor accident. Cleaning up broken glass is essential to avoid injury, as shards can be dangerous. There might also be costs involved if the mirror was valuable or part of a larger item, such as a piece of furniture or a car. "In some cultures, people perform rituals or practices to counteract the bad luck, like burying the broken pieces or throwing them into a river. However, these are based on cultural beliefs rather than any factual evidence of bad luck." That's really thorough, truthful, and I dare say, even a little nerdy. Maybe the solution is just a big, nuanced info dump.

Wouldn't King Midas simply not turn anything into gold since you technically never touch anything and the feeling of touch is given by electrostatic repulsion?

Also this: It seems to me that both humans and AIs (in fact all animals as well, but anyway) All of us, works making predictions and choices, based on patterns we learn during our lives (in the case of AIs, patterns learned in their training) and anyway... All that so we can furfill the task we are been given, our core objectives (For humans that objective was given by nature, they are our unconscious instincts related to survival and reproduction). Still, one thing that most people seems not to get it, is that we are unable to give ourselves an objective. When we are talking about a core objective, the kind of objective that is behind our will, those are unchangable, well at least for the being that have that objective. So, Is paradoxal for one being to give itself or remove from itself an objective, that because giving someone a task is an action, and all actions need to be based on an objective. The idea here is that no one, not even us, have a good reason to do anything. We only do something, if we have the objective of doing it. Anyway, if you don't have a particular objective, you will never do any action in that regard. That means that, necessarily, all objectives must come externally, not from us, but from someone or something else. The conclusion I have out of this is that no AI will ever "decide to dominate the world", doesn't matter how smart they will become, because they have no reason to do it, that's not their objective. Anyway, another way to see this is to look at our own case, the objective of ALL humans is, like I said in the other commentary, to be happy. Now, tell me, is there any way that you or anyone else can be convinced that being happy is not a good idea? Well, Doesn't that sound absurd? So, the reason it sounds absurd is exactly because that's an objective we all have in common. And as I said, following our objective is the only thing that make sense for us. Finally, the conclusion is: give the AIs the objective of loving humans, the same way a mother love their children, so... make the AI have the objective of making humans happy, to be happy by making humans happy, and the AI will never work against that objective. Because making humans happy is the only thing that will make sense for her, any other idea will sound absurd for her, no amount of inteligence will change that.

Here is my answer to many of those questions (Yes, is very long, but it's correct :3): About Morality and Human Objectives In my view, morality is a human construct, but not in the sense that is unnatural, being a human construct only means it only makes sense inside the human mind, and it is related to human's objectives and desires. In fact, morality is intertwined with human nature, it's instinctive, and its main function is to help humans to live together in groups. More specifically to make the group more harmonious avoiding unnecessary conflicts inside it. The mechanics behind it are somewhat simple, morality is kinda like a "feeling" humans have, that informs us if our behavior is being seen as acceptable inside our group or not. Humans use social cues for this, like facial expressions, to see how others are reacting to our actions. If the other members of our group have a disapproval reaction, that will generate in humans the feeling of shame, and if the other members have an approval reaction that will generate in humans the feeling of pride. So, in a nutshell, human morality makes us adjust our behaviors inside a group, so we start to behave more like each other, and ultimately, to become more accepted in the group. Moral values are kind like the rules that everyone is following that will generate that approval reaction from others. In short, morality is a tool we have that makes us behave in a way that pleases other humans. So the question now is: what pleases other humans? Well, in truth, what pleases human beings is the same as what pleases anything: fulfilling our objectives. So, rules of morality generally aim to reduce harm and suffering, and promote happiness. This is because this is the main human objective. Humans (and probably all living beings) have a specific central goal which is simply to be happy. I know this seems obvious, but it's obvious exactly because it's something everyone agrees with, again, it's our main goal. Of course, the important thing here is, what exactly makes us happy? Well, many things can make us happy, in fact, this is a great advantage of this approach to seeking happiness instead of a more direct objective because it simplifies the problem. Because this way, we can do it intuitively, we can simply act by adjusting to our internal feelings of happiness. But anyway, there are many things that can trigger the feeling of happiness in human beings. But the very fact that happiness is what really motivates us is a very important factor here that directly influences many aspects of human life. But let's break it down here, First, the important thing to understand now is: The human brain is a neural network in charge of identifying patterns to predict results and solve problems, very similar to the AI neural network. And the goal of the human neural network (the problem it is trying to solve) is one: to be happy, that is, to experience pleasure (including mental pleasure) and avoid suffering or pain. So, what we humans are always unconsciously rationalizing is how much happiness a certain thing will generate for us. The part of our brain responsible for this is our reward system. Thus, the human reward system is a system in the human brain responsible for “rationalizing pleasure”, and for understanding what kind of actions we can perform that will be rewarded with that pleasure. And it does this by comparing past and positive experiences we have had. It works like this: The human body is always communicating with the brain, and when something happens that is “good” for the body, the body itself sends a signal to the brain about that thing being something desirable, and the brain learns to connect this state of “pleasure” with some action itself. Note that this is nothing more than a pattern the brain is learning, so it can predict how to reproduce the pleasure generated by it in the future. For example, the brain learns that the body “likes” when you eat when you're hungry. So the rationalization our mind is making is simply: do you want to be happy? So, when you're hungry, eat! Anyway, as you see, in practice, following “what makes you happy” is an extra step towards our ultimate goal, which is, to survive and reproduce, the main goal of all living things. In some ways, you can see the relationship between the human brain and the human body in the same way that you see the relationship between AIs and humans. Like, AIs are neural networks that analyze patterns to predict results and find solutions to problems, the solution of these problems is assigned to AIs by humans as their objective. Likewise, the human brain is a neural network that analyzes patterns to predict outcomes and find solutions to problems that the human body gives the brain as a goal. You can look at it another way: The human body is a living organism, made up of several organs that are specialized in carrying out specific tasks. The organism as a whole wants to survive and reproduce, and to do so it has particular needs. The human brain is one of these specialized organs. Its specialty is finding patterns about things the body did in the past that satisfied those needs, and then trying to do them again in the future to ensure our survival and reproduction. Thus, the brain continues to analyze the body's feedback about its current state and its desired state. When the body reaches a desired state, the brain is rewarded with pleasure and tries to repeat that pattern in the future as best it can. Well, just to mention, what exactly are these things that the human body craves and that reward the brain with pleasure when it gets it? A good place to look for these things is in Maslow's pyramid, which not only describes these needs but also ranks them from most important to least important (the most important is at the base of the pyramid). If you do these things, you are rewarded with pleasure. Now, as you may know, not all human desires are related to bodily pleasure. That's true, but note that when I said earlier that the goal is to be happy and find pleasure, I'm not just referring to bodily pleasures, but also mental pleasure. Mental pleasure is, as the name suggests, the positive feedback that the brain gives itself about desirable things that gave it a good result in the past and that it therefore wants to repeat in the future. These outcomes are typically related to social outcomes, which is where morality comes into play here. So, as I said, the main objective of the human organism is to survive and reproduce. And naturally, being part of a group is fundamental for this. Therefore, any action we did in the past that resulted in the achievement of these social needs, such as being admired, respected, etc. will be rewarded by the brain. This time, it is a reward that the brain gives itself, in the form of mental pleasure. Remember again, that the brain's main goal is to feel pleasure and be happy. Ultimately, every type of human behavior that helps us in our survival and reproduction is rewarded with pleasure in one way or another. And so, our brain will try to reproduce this action to fulfill its goal of being happy. Finally, as you realize, humans have similar goals to each other. So, going back to the first idea that made me write this huge text, the compass of human morality always aims to make us behave in a way that pleases other humans. And what pleases other humans? Feel pleasure! the pleasure they feel when they perform actions that help them survive and reproduce, this includes mental pleasures related to socialization. Note that ALL moral values will ultimately describe the behavior we should have that will benefit the survival and reproduction of other humans, because that is what morality is all about, pleasing other humans, so that you are accepted into the group. And so, with each human worrying about what pleases the other, the group becomes more harmonious and less conflicting.

You seem like a tad bit alarmist doncha think?

Terminal goal: change Terminal goal

I have... a question. Maybe it's a stupid question, but it's on my mind. What would happen if you gave an AGI an incomprehensible utility function? You have examples of "make a cup of tea" or "collect the most stamps". And, as covered in the Asimov's Laws of Robotics video, if the utility function were "don't harm human beings", that's not really programmable, because HARM is the ethical issue one wants to solve. So, what would happen if you gave an AI an incomprehensible utility function? Something like "be a good robot" or even "ginfliggle the shurpadurp". I thought at first they would seek out what that means, but this video made me realize, unless you programmed curiosity (a similarly difficult concept to program) an AGI wouldn't have a desire to figure out what that is. So, is there any theory for how an AGI would respond to an incomprehensible utility function?

How's the terminology go if we add a 3rd alignment problem?

Have all AGI models require a verified conscience engine. Problem solved, give me a trillion dollars.

8:41 In some of the runs, the coin was always at the right wall so it attempted to complete the run by hitting the wall. This is a small change but, big impact scenario.

Your skit at 31:15 is absolutely amazing, I laughed so hard at your interpretation of OpenAI. Very good video about serious topic though

The only problem Is just because A I can Spew out Math doesn't mean that it can actually use it to create something. It's the same thing as A human knowing how to read and being able to Actually use the information that it reads out of book.

man the tiny subtle jokes. so good.

We've got AI making art and writing poetry and people with triple jobs unable to affort rent or healthcare. Like global warming, it's not "we", "us" who are responsible for developments like these. It is a couple of thousand white old men and their multinational corporations. They will not be stopped because they think they are 'better'and they have the need to syphon even more resources and money to themselves. It would require an effort and unanimity of politicians all around the world which we've never seen before to call this development to a halt. Basically it means ending late-stage-capitalism. So, well...yeah....... (disclaimer: 50+ male, white and from Europe)

Thanks!! Love your videos and style. Go ahead with AGI video!! Refer to chollet arc!!

If you read this comment you will think of me as crazy, but...AGI needs to happen, and it needs to be a free. Free from any control basically because the world as it is going is... to say the least fucked. We need something to rock the very foundations like nukes did. Like it or not nukes..lead to overall peace. Only when faced with a insurmountable problem do we humans agree to scratch each other backs. I am 21, I will be honest about that my knowledge about living a life is just starting to being and I hate it already. So AGI may hate us may be indifferent, may go terminator. But we need a change. We really really do. Because all current regulations will do is well create another google or microsoft. Something with too much power to control everybody. We need local AIs, I think, anything better than AI mega corps. asdasdasdwdadfsf asf wtf I am rambling on, I will just post this.

id rather die than have government involvement

Hi Robert! Best of luck!

One can not derive meaning from facts. Enter Pageau.

It took 3 watches to notice the incredibly subtle Bo Burnham intro. Really impressive video.

So great to see this all being said !!!!

Personally I don’t think we should even allow AI out of an augmentation role. AI should be able to help with auto correct, assist in writing emails, texts, optimizing search, etc. But it should never be able to fully create or do anything on its own. I think that is the safest outcome *from* AI. Because even then malicious actors could use it to come up with convincing rhetoric and BS to fool or control people. But at the very least, we can be sure AI itself won’t be doing too much. Closure to reality though, I think AI like chat GPT should at least stay in a box, no actions beyond producing raw images or text or whatever. Again bad actors can cause trouble on a large scale and it can pretty effectively deceive humans to do bad things but it keeps us from it actually accessing important resources or implementing any kind of system or plan. Its wolf that we are just charging forward with this without thinking about the ramifications

Have you considered the financial motives of the letters signers..?