Good point on AGI. Carmack of Doom fame should be an excellent contender for the next improvement in AI. Still current state of deep learning for narrow AI is amazing! I did not expect so much progress since 2010 and MNIST success. Also the last 3 years with generative AI in art and language models seems super amazing!
Glad John Carmack left Meta to work on his own company. They take too much credit for his work! Also funny he named his company "Keen Technologies" named after his classic video game "Commander Keen". and AGI? somehow that reminds me of Apogee. Very confident he will be successful in his work, he's already developed Raycasting/Raytracing decades before NVIDIA knew what that was, he invented FPS gaming, and he developed a early version of AI within Quake 3 Arena where the in game bots would actually talk to you, listen to you and respond back.
He's an underrated genius. I feel like the community has largely forgotten him, despite being an OG. I never played commander Keen, but I still love doom and quake 1 to this day.
When I was a teenager fond of electronic valves and about to enter college the first transistor implementations started to enter the news in magazine like Popular Mechanics. At the time my fascination fueled my imagination with ideas that my teachers stamped as absurd and provided the foundations for my peers to have me as the target of mockeries. Now adays weird people like I was at the time are respectfully identified at least as geeks. The reason for the resulting, (just described,) circumstances was my idea (hypothesis) that artificial life was only at a stone throw distance of being a possibility. My idea was based on the knowledge that (at the time ,) we were scientifically able to reproduce any waveforms and due to the fact that we were miniaturizing electronics to the sizes of transistors and diodes I was able to imagine an hypothetical black box in which (by a special technique I will describe next,) the life of a conscious human being could be contained. Let’s imagine that the technological capabilities described are available to us contemporarily. And we had a subject volunteer for a medical intervention in which each neuron in his brain would be replaced by a single (transistor as I used to called over 65 years ago,) device producing the exact same waveforms. One of the criticisms I used to get was; “the subject would die”, hence; my idea was dead on arrival. To what I would reply that if the procedure could be carried out following a special protocol it would be possible to achieve the goal without killing the subject. The protocol was that for each neuron signal being cloned (and connected to the subject’s brain,) the neurologists will ask the subject if he still was the same individual as before. And only after a positive confirmation the neuron will be detached leaving the transistor carry out the tasks that the neuron had. And so, keep on doing for each and every other neuron remaining on the brain. Any biological brain (from the simplest to that of humans) is the result of millions of evolutionary mutations. And we are trying to create not just an electronic object that can mimic the functionality of circuitry but also consciousness. As you said, there are theoretical works that describe some possible evolution for our machines, but perhaps there is also a way to kick-start the consciousness on machines by transferring it from us to of them. ---
What about the plasticity of human brain? We can adapt, grow and modify our circuitry by means of neurotransmitters, biochemical reactions and past experience. If you recreated a human brain using this hypothetical method it would be at most the same exact brain at the same exact second. One second later the human you cloned would be a different being potentially capable of a different decision based on the same inputs.
What is the route carmack is testing? He has mentioned that old papers might have ideas which didn't pan out due to computational constraints in a couple of interviews. Video request: alberta plan.
Maybe John Carmack has the right qualities for someone to develop AGI, but does he have a team of skilled ML research to support him? Because I believe that that is still a requirement.
@@dumb8671 Richard Sutton may be smart, but he is only one person (not a team, which you need), and I'm not sure how much he knows about deep learning either, which you need to know a lot about since that is what the current AI boom is based upon. Only combining present work will not be sufficient if you want to be the first one to develop AGI, because the major players are already doing that, plus they are carrying out their own research. But Maybe Carmack's goal isn't to be the first to develop AGI. Maybe he is okay with being an adapter; there is nothing wrong with that as it can be a very profitable business model too (just look at Apple).
@@dumb8671 It doesn't matter so much whether he knows about AI. He and Sutton are still just two people. OpenAI, Google, Microsoft etc. all have large teams of very skilled machine learning researchers. That's what you're competing against, and that's why I originally asked whether he has a team of skilled ML research to support him. Without that, he won't be able to develop AGI even if he himself and Sutton both are very skilled.
Hi Phil, I think you are 93.07% "right" here. 😁 However, I think 2/3rds of the triumvirate are not particularly captured by their corporations. Specifically Hinton is : A) Interested in biologically plausible neurons (my words) and B) willing to cannabilize his own work i.e. gradient descent Also I think Bengio has a stronger grasp of mathematics than you appear to give him credit.
I find it odd that John carmack is working on “agi” considering his background. I suppose when he worked on rocket ships he wasn’t exactly qualified to do that either lol
It seems you lack sufficient information on the extent of Carmacks competences. From what I remember about his rocket company, they were the first to solve vertical landing for rocket reuse. Big deal. Especially given he had no prior experience as an aerospace engineer
that is absurd reasoning!! that is like hospitals deciding nobody was qualified to perform life saving surgery prior to it being done for the first time... So, let's see, then... The best way of reaching the point of the realization of performing actual successful life saving surgeries is not with an acclaimed team of physicians with proven, respected, and relevant training, research, and practice in the nascent field of surgical medicine... But instead our real hope lies in someone with little to no actual related practical knowledge base, even less displayed theoretical/analytical/practical or even hypothetical experience in the specific field, and though may be very smart, and accomplished amazing (if only tangentially related) things, clearly has a rather inflated sense of self-efficacy that has long remained unchecked... But, ya know how all advancement goes... If it hasn't been done yet, "who can really even say which qualifications are even relevant to doing serious surgery on someone!" I assume you'd be the first volunteer under the dubiously qualified, self taught surgeons knife? now, sounds to me like that line or reasoning starts to make less and less sense the more i ponder it... with my brain recoiling in absurdity.... @@MachineLearningwithPhil
Good to see here that AI is realized to be nonexistent; total marketing BS today. No amount of deep learning and processing power will parallel human intuition
A.I needs a strong theoretical background. Carmack has none of that. He is a low level engineer with practical knowledge. So his scope is very limited.
@@dumb8671 He has no academic-level knowledge, and has never been the guy to acquire it (his math skills never exceeded school level, for example). He is a low level engineer with a strong intuition from the very bone, but at some point it is not enough to progress further.
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
Good point on AGI. Carmack of Doom fame should be an excellent contender for the next improvement in AI. Still current state of deep learning for narrow AI is amazing! I did not expect so much progress since 2010 and MNIST success. Also the last 3 years with generative AI in art and language models seems super amazing!
Glad John Carmack left Meta to work on his own company. They take too much credit for his work! Also funny he named his company "Keen Technologies" named after his classic video game "Commander Keen". and AGI? somehow that reminds me of Apogee. Very confident he will be successful in his work, he's already developed Raycasting/Raytracing decades before NVIDIA knew what that was, he invented FPS gaming, and he developed a early version of AI within Quake 3 Arena where the in game bots would actually talk to you, listen to you and respond back.
He's an underrated genius. I feel like the community has largely forgotten him, despite being an OG.
I never played commander Keen, but I still love doom and quake 1 to this day.
When I was a teenager fond of electronic valves and about to enter college the first transistor implementations started to enter the news in magazine like Popular Mechanics.
At the time my fascination fueled my imagination with ideas that my teachers stamped as absurd and provided the foundations for my peers to have me as the target of mockeries. Now adays weird people like I was at the time are respectfully identified at least as geeks.
The reason for the resulting, (just described,) circumstances was my idea (hypothesis) that artificial life was only at a stone throw distance of being a possibility.
My idea was based on the knowledge that (at the time ,) we were scientifically able to reproduce any waveforms and due to the fact that we were miniaturizing electronics to the sizes of transistors and diodes I was able to imagine an hypothetical black box in which (by a special technique I will describe next,) the life of a conscious human being could be contained.
Let’s imagine that the technological capabilities described are available to us contemporarily. And we had a subject volunteer for a medical intervention in which each neuron in his brain would be replaced by a single (transistor as I used to called over 65 years ago,) device producing the exact same waveforms.
One of the criticisms I used to get was; “the subject would die”, hence; my idea was dead on arrival. To what I would reply that if the procedure could be carried out following a special protocol it would be possible to achieve the goal without killing the subject.
The protocol was that for each neuron signal being cloned (and connected to the subject’s brain,) the neurologists will ask the subject if he still was the same individual as before. And only after a positive confirmation the neuron will be detached leaving the transistor carry out the tasks that the neuron had.
And so, keep on doing for each and every other neuron remaining on the brain.
Any biological brain (from the simplest to that of humans) is the result of millions of evolutionary mutations. And we are trying to create not just an electronic object that can mimic the functionality of circuitry but also consciousness.
As you said, there are theoretical works that describe some possible evolution for our machines, but perhaps there is also a way to kick-start the consciousness on machines by transferring it from us to of them.
---
A modern ship of Theseus
@@MachineLearningwithPhil You have been the first ever to give sucha a compliment. It is greatly appreciated.
What about the plasticity of human brain? We can adapt, grow and modify our circuitry by means of neurotransmitters, biochemical reactions and past experience. If you recreated a human brain using this hypothetical method it would be at most the same exact brain at the same exact second. One second later the human you cloned would be a different being potentially capable of a different decision based on the same inputs.
What is the route carmack is testing? He has mentioned that old papers might have ideas which didn't pan out due to computational constraints in a couple of interviews. Video request: alberta plan.
It seems like he's playing his cards close to his chest. I'll see if I can dig something up.
Maybe John Carmack has the right qualities for someone to develop AGI, but does he have a team of skilled ML research to support him? Because I believe that that is still a requirement.
@@dumb8671 Richard Sutton may be smart, but he is only one person (not a team, which you need), and I'm not sure how much he knows about deep learning either, which you need to know a lot about since that is what the current AI boom is based upon.
Only combining present work will not be sufficient if you want to be the first one to develop AGI, because the major players are already doing that, plus they are carrying out their own research. But Maybe Carmack's goal isn't to be the first to develop AGI. Maybe he is okay with being an adapter; there is nothing wrong with that as it can be a very profitable business model too (just look at Apple).
@@dumb8671 It doesn't matter so much whether he knows about AI. He and Sutton are still just two people. OpenAI, Google, Microsoft etc. all have large teams of very skilled machine learning researchers. That's what you're competing against, and that's why I originally asked whether he has a team of skilled ML research to support him. Without that, he won't be able to develop AGI even if he himself and Sutton both are very skilled.
I think R&D at Robocop should use OpenAI gym before giving them live ammo.
:D
I'D BUY THAT FOR A DOLLAR!!
If we ever achieve AGI it will be through reinforcement learning and reward based algorithm
I'm partial to that path myself.
Hi Phil,
I think you are 93.07% "right" here. 😁
However, I think 2/3rds of the triumvirate are not particularly captured by their corporations.
Specifically Hinton is :
A) Interested in biologically plausible neurons (my words) and
B) willing to cannabilize his own work i.e. gradient descent
Also I think Bengio has a stronger grasp of mathematics than you appear to give him credit.
Great points. I didn't mean to down play the intellect of the giants in the field. I just meant a fresh set of eyes is a big advantage.
I think you are 5% right
c-mon, at least 6.5%... quit being a hater!
AGI(T) will be built by AGI (T-1)
too many colourful video clips. but the clickbait title worked, and thanks for the pointer!
Thanks for the feedback.
I think that AGI will be made from a kind of neural networks customized threw genetic algorithm and that learned using DRL theory.
I'm curious what you think of AIXI, especially considering your RL background!
I honestly haven't heard of it. I'll check it out!
Where is that video from at 0:26? Sry not to discuss the topic :D
It's Robocop. I highly recommend that you watch it.
I think Ben Geortzel's ideas are most concrete for achieving AGI in our lifetime
I will have to look into what he says. Thanks.
But believing decentralized block chain leads to AGI is madness.
@@dhiraj_shah Opencog isn't on the blockchain, SingularityNet is. Opencog is the AGI framework.
"HEre's why" then zero real explanation
I find it odd that John carmack is working on “agi” considering his background. I suppose when he worked on rocket ships he wasn’t exactly qualified to do that either lol
Nobody is really qualified to work on AGI since nobody has done it before. We don't know which qualifications are relevant.
It seems you lack sufficient information on the extent of Carmacks competences. From what I remember about his rocket company, they were the first to solve vertical landing for rocket reuse. Big deal. Especially given he had no prior experience as an aerospace engineer
Carmack's rocket company was a big success in terms of tech advancements, the only reason it was sold is because it wasn't profitable.
Qualifications are imaginary
that is absurd reasoning!!
that is like hospitals deciding nobody was qualified to perform life saving surgery prior to it being done for the first time...
So, let's see, then...
The best way of reaching the point of the realization of performing actual successful life saving surgeries is not with an acclaimed team of physicians with proven, respected, and relevant training, research, and practice in the nascent field of surgical medicine...
But instead our real hope lies in someone with little to no actual related practical knowledge base, even less displayed theoretical/analytical/practical or even hypothetical experience in the specific field, and though may be very smart, and accomplished amazing (if only tangentially related) things, clearly has a rather inflated sense of self-efficacy that has long remained unchecked...
But, ya know how all advancement goes... If it hasn't been done yet, "who can really even say which qualifications are even relevant to doing serious surgery on someone!"
I assume you'd be the first volunteer under the dubiously qualified, self taught surgeons knife?
now, sounds to me like that line or reasoning starts to make less and less sense the more i ponder it... with my brain recoiling in absurdity....
@@MachineLearningwithPhil
Good to see here that AI is realized to be nonexistent; total marketing BS today. No amount of deep learning and processing power will parallel human intuition
A.I needs a strong theoretical background. Carmack has none of that. He is a low level engineer with practical knowledge. So his scope is very limited.
@@dumb8671 He has no academic-level knowledge, and has never been the guy to acquire it (his math skills never exceeded school level, for example). He is a low level engineer with a strong intuition from the very bone, but at some point it is not enough to progress further.