@@Go-Meta I'll try my best sharing your videos with my friends. It's perhaps the most important topic of the moment and you explained a lot of its shades (and made it understandable to everyone) perfectly. Thank you for the content
you take an A.I. Let's say chat gpt5 Attach it to a quantum computer. Attach that to a fusion device . Now you have your super computer intelligence. And it's still not self aware.😅 Has no goals other than the ones we give it. So the most important part is who's in charge of the computer. We will probably see this within five years. live long and prosper.
This is a well made video that feels very convincing, but there's little here to indicate that this is much more than the way you personally imagine that these events might unfold. You're postulating a scenario where superintelligence exists but is constrained to the current limitations of present models. Present models are not superintelligent, so they're not necessarily good analogues to use to imagine what a superintelligence might look like. By the time we reach superintelligence, or even AGI, even if it's only a decade from now, it very well might ALREADY be distributed between various hardware systems, or use hardware systems that are far more advanced than current systems. You're assuming that we'll have AGI, but that it will look exactly like the AI systems of today. But people are already using powerful AI systems, which aren't AGI, to design systems which are more advanced. Take Nvidia's recent use of GPT4 to play minecraft. This system, although not an AGI, is distributed among a number of different servers, and programs itself to become more effective over the length of its lifespan. People can, and are, using AI in the development of things like quantum computers and DNA data storage, and these systems are already in communication with each other despite being hosted by different platforms in physically different locations. There's no realistic way for any single data center to monitor the activity of these systems. AGI may very well occur through the interaction of multiple different systems that are distributed from inception. Maybe this will all play out the way you forsee, but there are too many unknowns to say for sure. If there's one thing I would say your perspective lacks, I would say that it's humility in the face of overwhelming complexity.
Hi Norbis, I agree we need to be humble in face of how little we comprehend what might happen, but I guess I'm also just wanting the discussion to acknowledge the difference between speculation that is based on our current, rational understanding of the way the world works, versus just imaging scenarios that could come straight from the page of a sci-fi novel. And, I certainly agree that the pace of progress in AI is so fast that it's easy for perspectives to get out of date quickly. I've seen some bits about the Minecraft playing AIs and I want to learn more about them because they look fascinating. But, I think it's important to keep in mind two things: firstly the architecture of most AI has (at least) two distinct phases: 1) the intense training phase of the neural network and 2) the runtime use of the trained neural network. Now some level of 'lifetime learning' can occur in phase (2), but often this is quite limited in its scope - even if it is impressive what can be done. I need to look into the Minecraft example to see how it fits into this, maybe it's all lifetime learning. The second thing is the way that those (like Kurzweil) predicting the imminent arrival of AGI are often thinking about the point where one neural network has roughly the same order of magnitude of neurons as the human brain. The reason I mention this is that just because people can shrink down specific LLMs to run on smaller, cheaper hardware, that doesn't mean that we yet know how to build an AGI style algorithm that would run with radically fewer neurons than the human brain. So, yeah, there's so much complexity here, and so much more I'm getting my head around to understand the rapidly changing state of the art. But I guess fundamentally I am hoping to challenge the idea that we humans have no agency in this process and that AGI and then superintelligence is just inevitably going to happen soon so we should just sit back and accept that. And thanks for the comment, it's good to have these ideas challenged! 👍
I'm wary of the benefits of superAI. In sci-fi novels we can imagine such superAIs solving all sorts of hard problems that we'd like to be solved, but my understanding is that many of the problems we'd like to solve are of the type that we have good reasons to believe could never be solved even by a superAI. But, I agree that it sometimes seems like we need someone/something "clever than humans" to help get us out of the mess that we've failed to solve for so many decades. I'm still foolishly optimistic that we can do it someday 😀
It is possible scenario that democracy will be completely destroyed by the AI, but all depends how democracy is used to control it. But there is no proof that democracy is needed to keep people happy.
Hi Jari, it may not be the case that democracy is needed to keep people happy on average, but I certainly think that it's the only way to have a state that is justified in enforcing the laws of the land. Of course, you need more than free voting to enable a just process of deciding the law and applying the law but I cannot see how a non-democratic system can be just. So, yeah, to me it is indeed of huge concern that AI might undermine democracy in all sorts of ways. And, trying to help prevent that, in however small a way, is one of the reasons why I'm making these videos 😀
It is still inexplicable to me why the global economy is not modelled in a classic economic simulation game, completely without AI, whose internal processes cannot be precisely understood. I'm not an expert and I'm happy to have someone explain to me why this idea doesn't work. Of course, you would have to incorporate much more complex systems than would make sense in an economic simulation game. You might not have to simulate 8 billion people either, but use specific abstractions. If statistics works, this should also be possible. You could then let "bots" (nations?) play against each other and perhaps make many interesting observations. Above all, it might be possible to conduct this eternal debate about basic economic forms such as capitalism vs. socialism (not an ultimately meaningful dichotomy, btw,imho) in a more nuanced way. Before we hand over "control" of the global economy to a "whatever super AI", why not try something like this first? (that's an exaggeration, but you know what I mean...) Give a thumbs up if you think it's the most millenial idea ever!
There are people who build such simulations at some level of abstraction. But, the fundamental problem is that you can't model within such a system the unexpected behaviours of the individual humans! Human preferences is arguably at the very heart of economics, but they are subject to the most unlikely and unpredictable whims. And, then, to be honest, even if humans were highly predictable, it is still very probably the case that the global economy would still be no more predictable than the global weather. And that's not because we don't understand the physics that drives the weather, it's fundamentally because mathematically "chaotic" systems cannot ever be accurately predicted far in the future. Indeed, a few years ago I made a video about the limits of rational predictions that you may enjoy: th-cam.com/video/3JCBugtM5Pk/w-d-xo.html
![⭐ฮิตในTikTok!! ( หมูเด้ง MooDeng ) Ver. แดนซ์โจ๊ะๆ ReMix BY [ ดีเจกิต รีมิกซ์ ]](http://i.ytimg.com/vi/3krsyfWUeHU/mqdefault.jpg)
I still don't understand how your videos still haven't reached 1M views. Very interesting and very clear, you deserve more audition for sure
Thank you! It's really tough to get views, but hopefully more people will see the videos someday 🙏 Thanks for the encouragement 👍
@@Go-Meta I'll try my best sharing your videos with my friends. It's perhaps the most important topic of the moment and you explained a lot of its shades (and made it understandable to everyone) perfectly. Thank you for the content
Thank you, that would be great 👍
Great video! Perfect summary of the issues we confront now ...and the foreseeable 5 years 🥴
Thanks David!
Just found you today! Am very glad I did.
Thank you! I'm glad you did too 😊
Same here!
you take an A.I. Let's say chat gpt5
Attach it to a quantum computer.
Attach that to a fusion device .
Now you have your super computer intelligence.
And it's still not self aware.😅
Has no goals other than the ones we give it.
So the most important part is who's in charge of the computer.
We will probably see this within five years.
live long and prosper.
This is a well made video that feels very convincing, but there's little here to indicate that this is much more than the way you personally imagine that these events might unfold. You're postulating a scenario where superintelligence exists but is constrained to the current limitations of present models. Present models are not superintelligent, so they're not necessarily good analogues to use to imagine what a superintelligence might look like. By the time we reach superintelligence, or even AGI, even if it's only a decade from now, it very well might ALREADY be distributed between various hardware systems, or use hardware systems that are far more advanced than current systems. You're assuming that we'll have AGI, but that it will look exactly like the AI systems of today. But people are already using powerful AI systems, which aren't AGI, to design systems which are more advanced. Take Nvidia's recent use of GPT4 to play minecraft. This system, although not an AGI, is distributed among a number of different servers, and programs itself to become more effective over the length of its lifespan. People can, and are, using AI in the development of things like quantum computers and DNA data storage, and these systems are already in communication with each other despite being hosted by different platforms in physically different locations. There's no realistic way for any single data center to monitor the activity of these systems. AGI may very well occur through the interaction of multiple different systems that are distributed from inception. Maybe this will all play out the way you forsee, but there are too many unknowns to say for sure. If there's one thing I would say your perspective lacks, I would say that it's humility in the face of overwhelming complexity.
Hi Norbis, I agree we need to be humble in face of how little we comprehend what might happen, but I guess I'm also just wanting the discussion to acknowledge the difference between speculation that is based on our current, rational understanding of the way the world works, versus just imaging scenarios that could come straight from the page of a sci-fi novel.
And, I certainly agree that the pace of progress in AI is so fast that it's easy for perspectives to get out of date quickly. I've seen some bits about the Minecraft playing AIs and I want to learn more about them because they look fascinating.
But, I think it's important to keep in mind two things: firstly the architecture of most AI has (at least) two distinct phases: 1) the intense training phase of the neural network and 2) the runtime use of the trained neural network. Now some level of 'lifetime learning' can occur in phase (2), but often this is quite limited in its scope - even if it is impressive what can be done. I need to look into the Minecraft example to see how it fits into this, maybe it's all lifetime learning.
The second thing is the way that those (like Kurzweil) predicting the imminent arrival of AGI are often thinking about the point where one neural network has roughly the same order of magnitude of neurons as the human brain. The reason I mention this is that just because people can shrink down specific LLMs to run on smaller, cheaper hardware, that doesn't mean that we yet know how to build an AGI style algorithm that would run with radically fewer neurons than the human brain.
So, yeah, there's so much complexity here, and so much more I'm getting my head around to understand the rapidly changing state of the art.
But I guess fundamentally I am hoping to challenge the idea that we humans have no agency in this process and that AGI and then superintelligence is just inevitably going to happen soon so we should just sit back and accept that.
And thanks for the comment, it's good to have these ideas challenged! 👍
Maybe the ONLY way to defeat Moloch is with a super intelligent AI... we are gone for sure if Moloch keeps leading our certain extinction.
I'm wary of the benefits of superAI. In sci-fi novels we can imagine such superAIs solving all sorts of hard problems that we'd like to be solved, but my understanding is that many of the problems we'd like to solve are of the type that we have good reasons to believe could never be solved even by a superAI.
But, I agree that it sometimes seems like we need someone/something "clever than humans" to help get us out of the mess that we've failed to solve for so many decades. I'm still foolishly optimistic that we can do it someday 😀
It is possible scenario that democracy will be completely destroyed by the AI, but all depends how democracy is used to control it.
But there is no proof that democracy is needed to keep people happy.
Hi Jari, it may not be the case that democracy is needed to keep people happy on average, but I certainly think that it's the only way to have a state that is justified in enforcing the laws of the land. Of course, you need more than free voting to enable a just process of deciding the law and applying the law but I cannot see how a non-democratic system can be just.
So, yeah, to me it is indeed of huge concern that AI might undermine democracy in all sorts of ways. And, trying to help prevent that, in however small a way, is one of the reasons why I'm making these videos 😀
It is still inexplicable to me why the global economy is not modelled in a classic economic simulation game, completely without AI, whose internal processes cannot be precisely understood. I'm not an expert and I'm happy to have someone explain to me why this idea doesn't work. Of course, you would have to incorporate much more complex systems than would make sense in an economic simulation game. You might not have to simulate 8 billion people either, but use specific abstractions. If statistics works, this should also be possible. You could then let "bots" (nations?) play against each other and perhaps make many interesting observations. Above all, it might be possible to conduct this eternal debate about basic economic forms such as capitalism vs. socialism (not an ultimately meaningful dichotomy, btw,imho) in a more nuanced way. Before we hand over "control" of the global economy to a "whatever super AI", why not try something like this first? (that's an exaggeration, but you know what I mean...) Give a thumbs up if you think it's the most millenial idea ever!
There are people who build such simulations at some level of abstraction. But, the fundamental problem is that you can't model within such a system the unexpected behaviours of the individual humans! Human preferences is arguably at the very heart of economics, but they are subject to the most unlikely and unpredictable whims.
And, then, to be honest, even if humans were highly predictable, it is still very probably the case that the global economy would still be no more predictable than the global weather. And that's not because we don't understand the physics that drives the weather, it's fundamentally because mathematically "chaotic" systems cannot ever be accurately predicted far in the future.
Indeed, a few years ago I made a video about the limits of rational predictions that you may enjoy: th-cam.com/video/3JCBugtM5Pk/w-d-xo.html
@@Go-Meta Wow! Thanks for the info and thanks for the answers! It is much appreciated!