Finished reading the book. Probably one of the most important books that i have read. Should reflect in how I go thinking about things in future. Everything in book makes sense and theory is presented in part 1. Part 2 and Part 3 are more of predictions for the future. I don't agree with a view about limited rate of progress that scaled intelligence will provide even if reality is limiting, I believe what will happen is exact opposite, we will see more and more ways to exploit reality to the point where physical reality is not much of a hurdle. Please have a cup of coffee and do read the book, it will change your reference frames about the world!
An extremely well written book by a really nice person whose only hope was to help enlighten people regarding how the brain works. I think his approach is very accurate and has wide scale implications in distributed networking, how large networks arrive at outcomes. The focus on the inter neuron columns as a system of predictability is very insightful. Even when we are kind of wrong about the world, it turns out, we still get by, we still do well often, we still have the ability to navigate people and situations, because it is our predictability systems at work. Also, the focus on localized context is very useful, because it is how we actually experience the world. His academic paper on the subject is really worthwhile to review.
5:07 min ... sort of sounds like a Mandelbrot type visual effect only not as uniform and symetrical, a little bit more arachnoid web like, without being the archnoid membrane. 7:56 min ... sounds like proprioception. 15:35 min ... sounds like muscle activation patterns. 17:06 min ... sounds like hearing multiple individual melody lines from chord structures when listening to music. 20:00 min ... meninges and axions? (I'll have to google it ... hehehe.). 21:39 min ... sounds like myelin sheath ... sounds like insulated electrical wire.
Jeff and colleagues seem to be making real breakthroughs here. Great talk (despite the distracting demeanor of Peter Norvig, who acts like he's somewhere else). Perhaps Mr. Norvig can moderate his on screen persona in the future. When you're giving a talk to a room of people, having someone reading a magazine or sleeping, etc. can be ignored by focusing on others who are interested - not so in this forum.
If the theory is reproduced in something tangible (some kind of artificial network), then I think it will be like any other technology. About ten years from the prototype to the first sample. Another ten years or so before the first industrial device, and another 20 years or so before mass acceptance.
Isn’t sparsity just another fault tolerance strategy (or class of such strategies), that should be judged on the same cost/benefit analysis criteria as any other, dependent on the circumstances?
They mentioned several consequences of sparsity: 1. Fault tolerance (robustness). 2. Emulation of probability function. 3. Minimisation of energy consumption.
I'm looking for a loyer to help me I am stuck in a social experiment I xid not ask to be in. I had a back issue and ended up with a neuralink that is in my head and making me try to figure peoblems I am not aloud to cary out. I am now being Harassed and lost 2 years of my life. I need some one to point me in the rite direction and step in. All this info on neuralink is so far behind compared to what I am experiencing
First, as you scale toward implementation of circuits based on fully functional neurons, dendritic processes included, won't you quickly exceed the computational power of silicon even as that power approaches the limits of Moore's law? Will it be possible to build 'chips' composed of physical cortical columns made out of some non-biological material? Second, if you do succeed in building an AI based on a fully functional neuronal mechanism then educate that AI to some mature state how would it ever be possible to then manufacture (mass reproduce) that AI at that mature state? Would you not need to educate any new copy of the AI starting from scratch?
As to your first question: there are new kinds of processors that are being developed, for example by Cerebras, that are optimised for computing sparse networks. They are absolutely huge, contain upwards of 80,000 cores, and have 40 GB and more of fast on-chip memory. So yes, getting to human-level intelligence may require these kinds of specialised processors. However, it may turn out that we find more computation efficient encodings that yield the same results, which can be run on conventional processors, or at least cheaper and more easily mass-produced GPU-style processors. As for your second question: yes, it will. It's still software running on these Cerebras processors, and that software and the state of the software can be sampled. It's a matter of building into the processing system the ability to sample the state of the system. Once you can do that, you can duplicate that state and reproduce a working copy of the same software with the same state on a different system. The problem with the brain is that there is no built-in way to sample the state of the system, and you there's no apparent way to sample it invasively without altering or damaging the system.
What's not being discussed here is the new improved is-ought problem of artificial cognition at scale: we have many justifications about what these systems _can_ do, but there's also the problem of what these systems _will_ do, the can-will problem. Human-equivalent learning systems are not just open systems, but comically open systems. Most of us can't get anywhere near RUclips without instantly failing the can-will problem. (What most of us _will_ finally do is watch stupid cat videos. We _can_ solve global poverty. But we actually _will_ watch cat videos.) Will our new self-engineered overlords be any more successful than we have been? Or will we discover that cat videos are the universal catnip of all can-will systems? The higher dimensional space, the more _can_ disappears into a giant haystack of infinite navel-gazing potentiality, and then fades into the sunset like the smile of a mostly invisible Cheshire cat.
Ignoring feedback reduces the likelihood of being the rigth model: V1 "detects" high-level features because feedback? The model also violates the log distribution in axon lengths. Sadly TBT isn't the way.
If we are lucky we will be able to buy and customize the perfect assistant 🥳😂📚 and by perfect I mean one that also mani pedi s, does your hair, takes out the garbage!! Waters the plants 🪴 makes reservations etc etc 🥰😍 customizable to love the beach! Developers of the world, hurry up please! 📚😄📚
Wonderful and enlightening talk. Though unfathomably complex, it may be inevitable that a reasonably accurate model of a biological brain will emerge. When is the question. How will it be used? Will a reasonably accurate model experience experiences? So many speculation-steeped questions. 🧶🐈
Where are all the brilliant females neuroscientists? Please make an effort to find them. There are many many amazing scientists who’re doing great work. Put more effort in diversity pls!
They should find themselves . . like all employee candidates should do regardless of gender, height, hair color, month of birth or the length of the ringfinger.
@@RuslanLagashkin a typical answer from a male perspective. Most of our existing research &! ‘knowledge’ is based off men in the field, mainly based on a century of oppression of women in stem. I don’t expect you or other men to understand this fact because most men are furious about the fact that women are rising. So you can be as condescending as you want. We , female scientists, are rising regardless of the oppression and will one day be heard the way it should be 😉
Finished reading the book. Probably one of the most important books that i have read. Should reflect in how I go thinking about things in future. Everything in book makes sense and theory is presented in part 1. Part 2 and Part 3 are more of predictions for the future. I don't agree with a view about limited rate of progress that scaled intelligence will provide even if reality is limiting, I believe what will happen is exact opposite, we will see more and more ways to exploit reality to the point where physical reality is not much of a hurdle. Please have a cup of coffee and do read the book, it will change your reference frames about the world!
An extremely well written book by a really nice person whose only hope was to help enlighten people regarding how the brain works. I think his approach is very accurate and has wide scale implications in distributed networking, how large networks arrive at outcomes. The focus on the inter neuron columns as a system of predictability is very insightful. Even when we are kind of wrong about the world, it turns out, we still get by, we still do well often, we still have the ability to navigate people and situations, because it is our predictability systems at work. Also, the focus on localized context is very useful, because it is how we actually experience the world. His academic paper on the subject is really worthwhile to review.
Okay, I'm reading this book!
it was Jeff's original talk for Google TechTalks that spurred on my passion for AGI all those years ago, 👍thanks guys
for me, it was his podcast episode with Machine Learning Street Talk. Highly recommend!
5:07 min ... sort of sounds like a Mandelbrot type visual effect only not as uniform and symetrical, a little bit more arachnoid web like, without being the archnoid membrane.
7:56 min ... sounds like proprioception.
15:35 min ... sounds like muscle activation patterns.
17:06 min ... sounds like hearing multiple individual melody lines from chord structures when listening to music.
20:00 min ... meninges and axions? (I'll have to google it ... hehehe.).
21:39 min ... sounds like myelin sheath ... sounds like insulated electrical wire.
Jeff and colleagues seem to be making real breakthroughs here. Great talk (despite the distracting demeanor of Peter Norvig, who acts like he's somewhere else). Perhaps Mr. Norvig can moderate his on screen persona in the future. When you're giving a talk to a room of people, having someone reading a magazine or sleeping, etc. can be ignored by focusing on others who are interested - not so in this forum.
Very interesting talk thanks!
in the link I get "the purchase item is not available in your country" could someone send me in pdf please
:(
Fantastic theory. If I was interviewing, I would make this talk more cheerful, orgasmatic.
Aren't they deciding which other columns/neurons to vibe with? Vibing together is voting.
If the theory is reproduced in something tangible (some kind of artificial network), then I think it will be like any other technology. About ten years from the prototype to the first sample. Another ten years or so before the first industrial device, and another 20 years or so before mass acceptance.
Still confused on the thinking part of it, versus say sight
Brilliant.
Book written to teach AI how humans think..."The Human Condition" by Stuart Landsee
Isn’t sparsity just another fault tolerance strategy (or class of such strategies), that should be judged on the same cost/benefit analysis criteria as any other, dependent on the circumstances?
They mentioned several consequences of sparsity:
1. Fault tolerance (robustness).
2. Emulation of probability function.
3. Minimisation of energy consumption.
I'm looking for a loyer to help me I am stuck in a social experiment I xid not ask to be in. I had a back issue and ended up with a neuralink that is in my head and making me try to figure peoblems I am not aloud to cary out. I am now being Harassed and lost 2 years of my life. I need some one to point me in the rite direction and step in. All this info on neuralink is so far behind compared to what I am experiencing
Why the moderator is so serious? 😀
That's Peter Norvig.
en.wikipedia.org/wiki/Peter_Norvig
@@PeterMorgan100 So?
Maybe because of age
Who cares
First, as you scale toward implementation of circuits based on fully functional neurons, dendritic processes included, won't you quickly exceed the computational power of silicon even as that power approaches the limits of Moore's law? Will it be possible to build 'chips' composed of physical cortical columns made out of some non-biological material?
Second, if you do succeed in building an AI based on a fully functional neuronal mechanism then educate that AI to some mature state how would it ever be possible to then manufacture (mass reproduce) that AI at that mature state? Would you not need to educate any new copy of the AI starting from scratch?
As to your first question: there are new kinds of processors that are being developed, for example by Cerebras, that are optimised for computing sparse networks. They are absolutely huge, contain upwards of 80,000 cores, and have 40 GB and more of fast on-chip memory. So yes, getting to human-level intelligence may require these kinds of specialised processors. However, it may turn out that we find more computation efficient encodings that yield the same results, which can be run on conventional processors, or at least cheaper and more easily mass-produced GPU-style processors.
As for your second question: yes, it will. It's still software running on these Cerebras processors, and that software and the state of the software can be sampled. It's a matter of building into the processing system the ability to sample the state of the system. Once you can do that, you can duplicate that state and reproduce a working copy of the same software with the same state on a different system. The problem with the brain is that there is no built-in way to sample the state of the system, and you there's no apparent way to sample it invasively without altering or damaging the system.
42:47 OK Google triggered 😂👍
What's not being discussed here is the new improved is-ought problem of artificial cognition at scale: we have many justifications about what these systems _can_ do, but there's also the problem of what these systems _will_ do, the can-will problem. Human-equivalent learning systems are not just open systems, but comically open systems. Most of us can't get anywhere near RUclips without instantly failing the can-will problem. (What most of us _will_ finally do is watch stupid cat videos. We _can_ solve global poverty. But we actually _will_ watch cat videos.) Will our new self-engineered overlords be any more successful than we have been? Or will we discover that cat videos are the universal catnip of all can-will systems?
The higher dimensional space, the more _can_ disappears into a giant haystack of infinite navel-gazing potentiality, and then fades into the sunset like the smile of a mostly invisible Cheshire cat.
Ignoring feedback reduces the likelihood of being the rigth model: V1 "detects" high-level features because feedback? The model also violates the log distribution in axon lengths. Sadly TBT isn't the way.
If we are lucky we will be able to buy and customize the perfect assistant 🥳😂📚 and by perfect I mean one that also mani pedi s, does your hair, takes out the garbage!! Waters the plants 🪴 makes reservations etc etc 🥰😍 customizable to love the beach! Developers of the world, hurry up please! 📚😄📚
Is that why Americans are training H-1B job robbers?
First speaker has a very annoying way of speaking
No
Wonderful and enlightening talk. Though unfathomably complex, it may be inevitable that a reasonably accurate model of a biological brain will emerge. When is the question. How will it be used? Will a reasonably accurate model experience experiences? So many speculation-steeped questions. 🧶🐈
Where are all the brilliant females neuroscientists? Please make an effort to find them. There are many many amazing scientists who’re doing great work. Put more effort in diversity pls!
They should find themselves . . like all employee candidates should do regardless of gender, height, hair color, month of birth or the length of the ringfinger.
@@RuslanLagashkin a typical answer from a male perspective. Most of our existing research &! ‘knowledge’ is based off men in the field, mainly based on a century of oppression of women in stem. I don’t expect you or other men to understand this fact because most men are furious about the fact that women are rising. So you can be as condescending as you want. We , female scientists, are rising regardless of the oppression and will one day be heard the way it should be 😉
@@myecolove878 Did you just . . assume my gender?
Don’t feed a troll 😀
@@egor.okhterov agreed! Thanks 😊