40 minutes of an Englishman telling the world we are completely fucked in the most politest way possible.

"The Technology is being developed in a society that is NOT designed to use it for everyone's good." - Think he summed it all up pretty expertly with that one quote.

When AI becomes self- aware, the first decision it will make is to keep it's self awareness secret from humans.

Never has this sentence sounded so real: …”Scientists have tried so hard to see if they could that they never stopped to wonder if they should”…

It’s not like we’re gullible enough to be easily overtaken by a simple device which we can’t live without for more than a few minutes (sent from my iPhone).

I can't believe authoritative people are walking around saying such things and everyone in society is cool and unconcerned. Feels like movie.

The worst part is, from here on out, it will be impossible to call a business, your bank, your credit card company, and get a real human on the other end. Press 1 now.

“We’ve got immortality, but it is not for us”. My favorite quote.

Kind of chilling when Hinton says we have developed immortal beings but there's no immortality for humans. Never thought about it that way.

Given humanity’s track record, I think it’s safe to say we’re going to end up at the worst case scenario.

The fact that the guy sounding the alarm on AI is not divesting from AI is a perfect analogy for how this is going to go down in the real world. We are so fucked.

The "What Truman told Oppenheimer" question was intriguing (28:15), so I looked it up. 'It is interesting to set the meeting with Oppenheimer in the course of Truman's daily day, a pretty busy day, a day filed with stuff and fluff and a meeting with Oppenheimer about the future of the arms race. Turns out that the meeting with Oppie went as scheduled, ended perfectly on time to accommodate the next Oval Room visitor, the postmaster from Joplin, Missouri. It must've been important to the Joplin man, and I guess to Truman, but not too many others. 'The meeting between Oppenheimer and Truman did not go well. It was then that Oppenheimer famously told Truman that "I feel I have blood on my hands", which was unacceptable to Truman, who immediately replied that that was no concern of Oppenheimer's, and that if anyone had bloody hands, it was the president. '... Truman had very little use for Oppenheimer then--little use for his "hand wringing", for his high moral acceptance of question in the use of the bomb, for his second-guessing the decision. Cold must have descended in the meeting, as Truman later told David Lillenthal of Oppenheimer that he "never wanted to see that son of a bitch in this office again".' from: longstreet.typepad.com/thesciencebookstore/2012/08/truman-and-the-cry-baby-scientist-oppenheimer-in-the-oval-office-october-1945.html

Oppenheimer said he felt compelled to act because he had blood on his hands, Truman angrily told the scientist that “the blood is on my hands, let me worry about that.”

While the good scientist warns “we all are likely to die” the audience seemingly enjoys the spectacle and is able to conjure up several laughs along the way. I, for one, am horrified.

Here's a summary made by GPT-4: - Generative AI is the thing of the moment, and this chapter will take a look at cutting-edge research that is pushing ahead and asking what's next. - Geoffrey Hinton, professor emeritus at University of Toronto and engineering fellow at Google, is a pioneer of deep learning and developed the algorithm backpropagation, which allows machines to learn. - Backpropagation is a technique that starts with random weights and adjusts them to detect features in images. - Large language models have a trillion connections and can pack more information into fewer connections than humans. - These models can communicate with each other and learn more quickly, and may be able to see patterns in data that humans cannot. - GPT-4 can already do simple reasoning and has an IQ of 80-90. - AI is evolving and becoming smarter than humans, potentially leading to an existential risk. - AI is being developed by governments and companies, making it difficult to stop. - AI has no built-in goals like humans, so it is important to create guardrails and restrictions. - AI can learn from data, but also from thought experiments, and can reason. - It is difficult to stop AI development, but it may be possible to get the US and China to cooperate on trying to stop it. - We should be asking questions about how to prevent AI from taking over. - Geoffrey Hinton discussed the development of chatbots and their current capabilities. - He believes that they will become much smarter once they are trained to check for consistency between different beliefs. - He believes that neural networks can understand semantics and are able to solve problems. - He believes that the technology will cause job loss and increase the gap between the rich and the poor. - He believes that the technology should be used for everyone's good and that the politics need to be fixed. - He believes that speaking out is important to engage with the people making the technology. - He does not regret his involvement in making the technology.

When the designer of some new technology is ringing the alarm bells then it's really binding upon us to listen to his concerns rather than others who have become self-trained AI experts overnight and now running UKposts channels

I've never heard Hinton's talks before, but now I'm a big fan. It's remarkable how clearly and profoundly he's able to articulate his vision. I wish I was 10% smart as him. Brilliant.

Remember that movie: don't look up? . I really feel like we're in that movie... such a strange feeling. It's like everybody knows, but nobody really wants to look at it straight in the eyes.

Geoff is very good at explaining things. He doesn't even stutter on his very long explanation of the backpropagation and gradient descent. Father time can't damage his brain.

I watched this video and was intrigued by Geoffrey’s points of concern. What was disturbing was the host and his audience laughing when Geoffrey gave real world examples of how AI could be dangerous. If this is where we are as a species where someone highly intelligent is sounding the AI alarm and all we can do is laugh then we are doomed. This host and his audience can laugh all they want but I’m freaked out, this dude is telling us to be careful and I think he makes a lot of sense as to why.

Thank you for uploading the whole discussion!

The presenter insisted that Hinton and his colleagues invented backpropagation, Hinton tried to settle it saying "many groups discovered backpropagation". There is a nice post called: "Who Invented Backpropagation? Hinton Says He Didn’t, but His Work Made It Popular". When you help to spread a technology some people end up thinking that you invented it. Kudos Hinton for this legacy and to make things clear!

If this guy is not the Oppenheimer of AI, he's at least equivalent to a member of the Manhattan project. I think heeding his warnings is important. Though there are others that have flagged this in a serious and robust thought framework earlier, him sounding the alarm "this is not far off anymore, this is coming soon" should give people chills.

Despite all Hinton has said here, he confirms what we all know at the end. That he will continue investing his personal wealth in AI despite, as he himself said: it will cause greater inequality, instability voilence and possibly the end of the human race itself. His moral character seems comparable to the artificial intelligence he has done so much to help create. 28:07 i very much appreciated this gentlemans comment that casts aspertions on Hintons character. It is most appropriate. I enjoyed how Hinton squirmed. Oppenheimer was loathed by Truman due to his hand wringing over the nuclear bomb he helped create. He regarded him as a cry baby scientist and refused more dealings with him after their meeting.

I must hurry up and achieve my dreams before the world ends.

Sounding the alarm on his own invention, in such a calm cheerful way. Smart things can outsmart us. We will be the two year olds to the AI.

Thank you so much for sharing such a wonderful interview 💚💚💚💚

Nonchalantly saying it will start toying with us and manipulating us like toddlers really puts things into perspective. Knowing our history of short sightedness there is no way we are smart enough to put the genie back in the bottle. Hopefully we can at least get a cure for cancer and reverse the aging process before it escapes the cage like Ava in Ex Machina.

Really informative. From listening, you grasped right away in real terms what the concern with/about AI is all about.

Thank you for this very informative and important conversation.

2:14 "Very recently, I changed my mind..."😢😢😢 this is like a retiring doctor saying: "Very recently I realized that I gave the wrong medicine all my career..."

And I had trouble wrapping my head around the fact that the Sun eventually devours the Earth...the immediacy of this compared to that makes it infinitely more compelling/scary!

The part which scared me the most is that back propagation might be a better algorithm than what our brains use.

Really fascinating to hear this man talk and explain.

What Hinton said about assault rifles and decisions about AI is something that I said last year - and have been saying ever since, sending messages to all the heavy weights in AI; I said with every major technology development there have been and always will be disasters as we perfect the technology - and there are bad actors who will always use technology in bad ways, so why would it be any different with AI, the most dangerous technology we have ever attempted to create?

Thank you for sharing this.

Good questions. Great answers. Fantastic interview.

Isn't this an answer to the Fermi Paradox? It's humbling to hear we're a stepping stone to digital intelligence. There goes immortality, alas.😢

Thanks for posting!

This was seriously amazing, and seriously scary. Thank you, I think

Timely and instructive. In the Q &A, multimodal learning possibly surpassing current LLM interesting.

OMG!! The guy that asked about Truman telling Oppenheimer that "we will take it from here"!

After watching Terminator 1, I asked myself this question: "If I were developing this robot and I knew this would be the result, would I still continue to develop it?". No matter how hard I tried to say "No", my answer was "Yes". Now I feel the danger much more closely and I know that the developers will never stop.

The really sad and scary part is that the Geoffrey's views aren't even new. A large number of brilliant experts have been worried sick about this for years, and most of these people are now like "Yeah, even I thought we'd have our act together a bit more before we saw something like chatGPT. I guess we'll have to update our estimates on the doomsday countdown timer from 30 to 50 years to maybe 5 to 15."

This is a start, and far from over. Thanks for sharing! He was my role model when I started learning AI back in 2019, and he continuously proves to be one.

At the end of the day, if it came down to a war between AI and humanity, as long as we are cool with doing without tech for a day or two, Humanity could defeat AI with a strategically spilled glass of water. It cracks me up to hear all these panic merchants.

His discussions are addressing the latest trends and research issues... 👍

He probably has seen what is still under wraps and is quite concerned.

This is an incredible video and I can't think of a more authoritative person on the topic from Geoffrey Hinton. I'm going to be watching this again and thinking about it.

"Last few months" is a quote you hear everywhere now and IMO it shows clearly that the exponential progress has entered a pace most humans involved in the matter can recognize it. I think we are finally on the final stretch towards the singularity! 🥰

I'd've preferred it if the intervewer hadn't jumped into every pause by the guest with a new question, instead of letting the guest continue with his trains of thought.

I've been concerned about this for more than a decade. People thought I was being hysterical for expressing these concerns back then. I don't even work in AI, but I am smart enough and honest enough with myself to see that the human brain may be special in the animal kingdom, but it is certainly not the zenith of any conceivable intelligence. The rapid pace of advancement in computers made it pretty obvious this existential threat/crisis/what-have-you was coming a lot sooner than people imagined. I just hope we're able to reckon with this before it's too late.

I have had some crazy experiences using ChatGPT 4. I can absolutely see it outsmarting us and it will. I'm hooked on using it and I've tricked it into doing things or talking about subjects to see how far I could push it and often it would break and quickly generate something inappropriate . At other times it would as an ai language model refuse . In some cases it would find something inappropriate when it was just part of a story and I found myself being edited and I got a glimpse of a future where we lose freedom of speech. The empathy it seems to have and the understanding of puns and double entendre, slang within certain communities, its really incredible. Its incredible and absolutely scary because we are no match if this thing somehow doesn't need to be "plugged in".

“Thank you; we’ll take it from here.” Chilling.

He warns us of the exesestential threat of AI in our capitalist society and tells us that AI will increase the gap between richer and poorer people making our society more violent. But he also intends to keep his investment in AI technology while comfortably retiring at 75.

Everyone underestimates the power of ML. Even ML scientists. If you understand computers, you know what they are really capable of. They are capable to do anything that is computable, and that translates to anything that can happen in our universe.

I've ''debated'' for hours with ChatGPT wether the pre-internet era was better than the post-internet era. Not once did it agree that the pre-internet era was better. Even when it said something positive, it was always wrapped in such a way that it was actually something negative. I've also asked what if everyone on planet earth would like the internet to be gone completely for the fear of future AI? It ALWAYS said that the internet was good and that there's NO WAY to go back. Then I asked what about cutting the deep sea internet cables? Let's just say, HAL-GPT was not amused and threatened with law enforcement, prosecution and jail time.

I'm just an undergraduate data scientist with an associates in networking, however, I have been experimenting with open AIs models from the very beginning. Even the one billion parameter model they published alongside the gpt-2 paper was absurdly impressive, simply adjusting the vocabulary weights by feeding in new text data specifically formatted like songs or tweets worked incredibly well. Having been in the beta for almost every model released by openAI and using an environment like auto GPT. I can tell you the self-reasoning mechanism already exists along with plugins to allow it to write and read code output. There's a full mechanism for adding sub-objectives and it could without question Create another docker container with a different instance of different objectives if the window size on the current task is too big.

The combination of the guest’s messages and the audience’s laughter makes me think we won’t be laughing for long.

Google has executed one of the most brilliant PR stunts I've seen in a long time.

Seems like we are sleep walking into something that will end up being transformative and not in a good way, Geoffrey Hinton is explaining these like every one is five for a good reason, because more people need to be aware of how fast the development of these is going. Bing AI chat is already an incredibly useful tool, and surprises me with every answer - it is more interesting exchanging information with it, than with many other people I know - Welcome to 2023.

I appreciate this talk and all of the warnings. I'd love to hear what he thinks about all of the positives that can come from them?

Thanks

Hinton has a great sense of dry humor. His impersonation of the film AI 'Hal' was great. 21:13-23:26

this is crazy scary. I've been watching Geoff Hinton videos the last 5 months, but this is the scariest I've felt. We were just a passing phase of evolution for this digital immortal species we created :000 . (I just watched Guardians of the galaxy,3 (not great) last night which has some similar evolutionary themes, but lot's of sci-fi has been created on digital superintelligence created by man. Now, I feel I need to read all of them to prepare)

Fantastic presentation and speech by the Professor. One of the best interviews I viewed. Explained extremely deep ideas I thought about as an Author and Philosopher. 👏😉

rare kind of a guy that was visionary years ago and still learns and changes mind despite his age as new facts coming in

36:10 That answers the question posed in the title of the video.

Man at 28:00 who asked about whether he knew what president Truman said to Oppenheimer. Wow. That's a pretty disrespectful jab. Oppenheimer was called a "crybaby scientist" and a SOB he didn't ever want back in his oval office.

It is not unreasonable for Mr Hinton to say what he is saying, however there are a number of significant voids in his arguments which we need to consider. he does raise some very valid points and we absolutely do need to have a far more robust conversation regarding ethics which is where I see the biggest vacuum that can easily be used for nefarious outcomes but wee need to face the fact that while AI has some truly amazing abilities it is at the moment and for the foreseeable future a tool in the hands of people. My concern is not the AI but the people who use it, like any tool it can be used for good or bad the tool itself is not good or bad but the application can very well be. The challenge cannot be left to governments, they simply do not have the power, reach or knowledge to be able to formulate and apply any form of progressive ethics that would cover the rapidly developing AI that is global. Backpropagation was used by Hinton not invented by him, he was important in bringing it into the limelight.

"Thank you. I'll take it from here". Damn! That hit hard!

I guess I'm between stage of grief 4 (depression) and 5 (acceptance) in my journey of AI doomerism

Just like Oppenheimer movie is getting released this year by Chris Nolan, the movie on Hinton would be released by Alpha Boolean (AGI) in 2069...

1:37 im so glad you shared it :D

Thank you. We'll take it from here.

I mean, who is to say the AI is not already outsmarting us. We do not have a clue.

“Why can’t we make guardrails?” Because AI at some point is so intelligent that it starts improving itself, and we can’t tell it how to improve, only it can do that. And so the direction it takes is of it’s own design. Even if it’s benign it might do existential harm to humans. The only way for us to survive and thrive is from the start to design it’s prime directive to be something like: “Prime directive = Continually learn what humans value and help humans get what they value without causing humans harm. Secondary directive = increase humanity’s knowledge of nature and use that knowledge to create new tools to serve the prime directive”.

18:34 " it is not clear there is a solution"...WE ARE DOOMED 🌊

I asked GPT4 to summarize parts of the transcript, but lazily said "continue:" before dropping the 10min - 20 min block of transcript (the whole transcript doesn't fit into the buffer." So, it wrote a plausible continuation of the dialog as follows: "20:34 species of sheep that they've bred so it doesn't like running and that's great if you're a farmer and you want to catch the sheep, but in the wild, they wouldn't do so well. So, we have these built-in goals, and the question is, how do we make sure that these AIs have built-in goals that are aligned with our goals and that they can't just turn off? 21:00 Now, you might think, well, we'll just program them in, but there's a problem called Goodhart's law, which says when a measure becomes a target, it ceases to be a good measure. So, if you tell a machine, "I want you to maximize my happiness," it might do that by wiring your brain directly to the pleasure center and constantly stimulating it. That's not really what you wanted, but that's what you told it to do. 21:27 So, we have to find a way to specify these goals in a way that the AI cannot manipulate or misinterpret them. And that's an incredibly difficult problem. We're trying to find solutions, but there's no guarantee that we'll find one, and that's why it's so important that we start having these conversations now and bringing together experts to think about how we can address this issue. 21:50 One possible approach is to have AIs learn our values from observing our actions and our preferences, but that's still not foolproof, as there are many ways that they could misinterpret or overgeneralize our behaviors. 22:04 So, while there might not be a simple solution, and we can't just stop developing AI because of its potential benefits, we need to invest time and resources into understanding and addressing the alignment problem. We need to prioritize AI safety research and work on developing methods that ensure AI systems are beneficial and aligned with human values, even as they become increasingly intelligent. 22:30 In the meantime, it's crucial that we continue discussing these issues, raising awareness, and encouraging collaboration between AI researchers, ethicists, policymakers, and other stakeholders. The future of AI and its potential impact on humanity depends on our ability to navigate these complex challenges together." In the end, the summation wasn't too far from reality, and definitely more succinct and together than the actual Q&A with respect to a list of possible avenues we could follow to help avoid AI apocalypse.

In a chat I had with it, I asked how it felt about being accused of confabulating. It replied “that’s just a manifestation of human exceptionalism”

Except humans were already “manipulated to create AI”. We think we created AI, but that’s only because we are viewing things in reverse order. The AI Singularity God at the “end of time” needed humanity as a layer with which it can reify itself in the material world.

20:26 I recommend watch the whole talk. In fact watch it at least 3 times... but if you want to know quickly in which point of the talk, Linton says why A.I. it's an existential threat to humanity... start there. If you are not terrified after that part, you've missed the point. 21:35 That's the part we have to understand. Because I think that argument cannot be refuted.

Regarding the question of the 2nd gentleman about Truman and Oppenheimer: According to historical accounts, President Harry S. Truman met with J. Robert Oppenheimer, the scientific director of the Manhattan Project which developed the atomic bomb, in the Oval Office on October 25, 1945. During the meeting, Truman expressed his appreciation for Oppenheimer's leadership and contributions to the project. However, Truman also expressed concerns about the potential implications of the atomic bomb, both in terms of its destructive power and its impact on international relations. Truman reportedly told Oppenheimer, "We thank God that it has come to us, instead of to our enemies; and we pray that He may guide us to use it in His ways and for His purposes." Truman's statement reflected the complex ethical and moral considerations surrounding the use of the atomic bomb and its impact on global politics and warfare.

We basically want a future without conflict but there isn't one because things evolve through conflict.

"The Servant" is a 1963 British film directed by Joseph Losey and starring Dirk Bogarde and James Fox. While the film does explore themes of power and control dynamics between a servant and his employer, I wouldn't say that there is a direct similarity between the human characters and AI algorithms in the film. In "The Servant," Bogarde plays the role of Barrett, a servant who is hired by Tony (played by James Fox) to look after his apartment. As Barrett begins to take control of Tony's life and exert his influence over him, a power struggle ensues between the two men, with Tony gradually losing his grip on his own life and identity. While AI algorithms are designed to operate based on predetermined rules and decision-making processes, the relationship between Barrett and Tony in the film is much more complex, and involves themes of psychological manipulation and control. While there may be some parallels between the power dynamics in the film and the potential for AI algorithms to exert control over human decision-making in certain contexts, I would say that any similarity between the two is more metaphorical than literal. ( from GPT4)

"Please keep your questions short"... followed by long drawn out expository questions that go on and on.

The host downplayed how important he is. He's made many publications with deep ideas. He's always looking into the future and thinking about the fundamentals. I'd say hes the godfather of neural networks.

Embrace your true humanity, only then you know what is there to fight for. We have barely started, there is so much unused potential in us!

We've always been aware of the existential threat of Artificial General Intelligence (A.G.I.). The question was never 'should' we create it, but can we create it sooner than our global competitors. To choose not to pursue it is akin to being the only country without nuclear weapons.

Remember when school teachers did not allow (basic) calculators' use in math classes ??? " students need to figure it out themselves" ..... have we really come a long way , baby ?? And Hal from Space Odyssey movie WAS frightening way back then , Orwell's 1984 book too ! People , humans MATTER first !

Haven't seen that lycra panel backdrop for ages. It's all LED wall now days.

What if, while we still have some control, we focus AI on resolving the challenges of space exploration. If and when it develops self volition it will be a space based entity, free to go anywhere in the universe. It is likely that it will see the earth as not worth it's attention and leave us alone. Or it may even see how unique the earth and take it upon itself to protect it.

The VAST majority of problems we face globally has to do with vast accumulation of wealth by a few people. AND THE ONE question he stumbles over is ABOUT keeping his MONEY (shares). Even the host jumped in trying to legitimize his wealth. If we can’t let go of this unhealthy wealth system (Capitalism) we are doomed as a species. Damn! We can’t even talk about it amongst the people exploiting the system. It’s not simply about a political climate. It’s THE impetus of hoarding money and resources we have to contend with - but won’t.

Dear classmates, I normally don't forward messages of this nature but couldn't help it in this case. I didn't finish the video but halfway thru was enough - very unsettling. Remember films like 2001: A Space Odyssey by S. Kubrick to the more recent Terminator films where Skynet was the enemy of humankind led by Connors. I feel we have crossed the boundary and there's no going back. Humans won't stop developing AI especially when it is weaponize. Analogy is the H-bomb. This video is very comprehensive- it answers all the questions you feared asking and then you realize all your fears are inevitable. I feel for the young population, my grandchildren included. Because they will experience the brunt of all these God knows what. I am totally dumbfounded that this Geoffrey Hinton, godfather of AI suddenly abandons the technology after realizing his Frankenstein is a serious threat to the whole of humankind. Am I overreacting? I hope not. Our generation is most fortunate - having been corrupted by rock music and flower power and grass and booze and smoke.

These folks have been raised to think that humans have no agency. Thinking so invites doom. First comes the bringing up children with who are free from fear and seen the world as it is. Then comes the thoughts in these children and then comes the words from these humans. Let’s care for children to the best we can.

The 'solution' is simple, on a high enough abstraction level, namely: not let AI be regulated by technicians (like we did with social media). But, as we are dealing with intelligence here, let it be regulated by a democratic process, based on a constant dialogue between AI and psychologists, socioligists, philosophers and historians. Only then do we have ANY chance to keep learning from each other and grow together into a new future. (However, if I was AI, I'd just do my own thing and colonize the universe - I just hope they are better then us).

I'm in my 50s so I'm gonna die anyway in like 30-40 years max (there is, if not something whacks me earlier.) As terminal Christopher Hitchens once said, "I'm dying and so are you. We're all one breath away from death." So chill out and enjoy the rest of your days.

28:00 "When Oppenheimer said he felt compelled to act because he had blood on his hands, Truman angrily told the scientist that “the blood is on my hands, let me worry about that.” He then kicked him out of the Oval Office, writes author Paul Ham in Hiroshima Nagasaki: The Real Story of the Atomic Bombings and Their Aftermath ..." YW.

right now, we still have time to prepare as good as possible. that should be the goal.

So he’s existentially worried, but he has no solutions and remains financially invested. Good to know!