This is a clip from a conversation with Francois Chollet from Sep 2019. New full episodes every Mon & Thu and 2-5 new clips every Tue & Fri. You can watch the full conversation here: ua-cam.com/video/Bo8MY4JpiXE/v-deo.html (more links below) Listen to it here: lexfridman.com/francois-chollet/ and subscribe on your podcast app. Watch other clips on the clips playlist: ua-cam.com/play/PLrAXtmErZgOeciFP3CBCIEElOJeitOr41.html Watch full episodes on the full episode playlist: ua-cam.com/play/PLrAXtmErZgOdP_8GztsuKi9nrraNbKKp4.html
I've long thought that there are likely many geniuses out there, but that many factors other than intelligence as well as external factors are required for those geniuses to make a noteworthy impact. The right events need to occur at the right times and in the right order, and they need ample resources, they need enough experience in a field to recognize a big problem. Imagine if Elon Musk had never moved to America, didn't have the opportunity to study physics, didn't have the opportunity to take an interest in electric cars while at university, was never involved with Paypal, etc.
And with guys like Einstein and Musk, it's not just their intelligence, it's also massive amounts of determination, work ethic and belief in themselves that's needed for the scope of achievements they've been able to accomplish.
@Deep Mind X if you're a child of average or even above average intelligence having a millionaire mother isn't going to push you into musk territory. wealth is definitely a factor, but its not "even more crucial" than whats mentioned above.
@@NorthGermanic maybe then, it's crucial to be careful of what is supplanting instinct and willpower within the hypothetically generally intelligent machine's toolset. instinct is (implicitly, though we may not be so sure) determined by a messy and drawn out web of interactive selection processes on radically different scales than that of the AI's selection processes. willpower is ill-defined for a machine if we keep its explicit necessity for 'impulse' as we currently define impulse, exertion, or obstacles for that matter. I get the feeling these aren't just minor semantic misgivings, but another serious glaring issue with the premise and promise of a general intelligence that doesn't include the endocrine, digestive, and interdependent biological (and societal, if we're to consider Chollet's external environment point) ecosystems that comprise an appropriately defined "intellect". If I could drive a point with anyone working towards a functionally benevolent GAI, this^ would be it. Upvote if you agree because I'm not going to shop a treatise around or write any grants about it.
@S R I've wondered about that route for sure, but don't have the necessary understanding of the complexity of those systems in their totality, to make any kind of guess on how computable they are or will be any time soon. sounds simple when you put it that way (just "make the virtual environment correspond to physical reality" as we understand it) but if you account for the minute scales of relevant detail, from the specific information encoded in a body's DNA (let alone all of the genetic information and activity of the microflora/fauna inhabiting that body) all the way up to all of the relevant information necessary to process the environment functionally, you're really getting into the weeds there. I'd really be curious of any professional's thoughts on the viability range with our current state of the art and speculations as to future possibility
“Everyone is a genius. But if you judge a fish by its ability to climb a tree, it will live its whole life believing that it is stupid.” - Albert Einstein This quote is so true. Everyone has a special understanding of something in their lives. All you really need is knowledge of self. Once you go back and start thinking about the things that you were always naturally passionate about, you will find it uplifts you and you will thrive.
I think he's trying to say that you need exposure to multiple disciplines so you can form new vectors to attack a problem, and have a problem that needs expertise in multiple disciplines to solve. These days nearly all scientists are regimented into a minute niche in their field and that's the limit of their thought and imagination. Think of Andrew Wiles - in solving Fermat he had to essentially learn a whole new field of maths, incorporate several completely new proofs that he produced, and still bridge it all back into modular forms to begin to tackle the problem. And even then it took the true masters of those fields to step in and help him stay true to the underlying maths and not go off. Einstein had that as well. Where in academia today is that even fathomable? If it takes AI to be that latent Einstein or Wiles in modern times than that's what it takes, but it's not actually beyond human capabilities WITHOUT AI
I heard Eric Weinstein say something similar to your first point about the hyper-specialised nature of science and a lack of polymaths in modern society. It's not something I've considered before. Interesting
Also, it is perfectly plausible that a large improvement processing power will not produce equally large improvements in problem solving - it could ultimately be one of those things where twice the processing leads to a tiny increment in intelligence.
This really resonates with the work of Professor Reuven Feuerstein whose theory involves thinking about thinking “meta-cognitively” - as a system with input, elaboration and output phases. The key / main point about the elaboration phase being defined as “awareness of the problem.” The idea being to actively look for problems needing to be solved rather than staying with the status quo / vested interests
Shortened version: “not-for-profit” // “large body of published, peer-reviewed research demonstrating consistent findings of significant enhancement of cognitive performance for disadvantaged learners” www.linkedin.com/pulse/recent-audit-cognitive-programs-university-auckland-james-colvin ***** An important meta-analysis / summary of the evidence base: eppi.ioe.ac.uk/cms/Portals/0/PDF%20reviews%20and%20summaries/t_s_rv2.pdf?ver=2006-03-02-125128-393 ***** Various documents gathered via google around the 2011 Nobel Peace prize nomination which was instigated by a Professor of “Pedagogical Science” from Italy. The one called “Biographical Notes” was the main letter: www.dropbox.com/sh/1qro3otdqhc93y7/AABE29WEp3tvhwrXNWqg63Pwa?dl=0 ******** A brief section of a much longer paper on the Johns Hopkins University web-site: www.dropbox.com/s/h20trfc2zpxkwfb/Johns%20Hopkins%20University%20Taunton%20Info.docx?dl=0 The whole paper is here: archive.education.jhu.edu/PD/newhorizons/strategies/topics/Instrumental%20Enrichment/hur.htm ******* This is a great video - I’m willing to bet that if you watch to the end it’ll surprise you. Also - when he appears here the Professor is articulating clearly / towards the end of his long life some important and (I think) profound messages: ua-cam.com/video/lqSQI6VJgLk/v-deo.html
Interesting thought, and I think he is probably correct about intelligence being a whole system and you can't tweak one thing and think you can predict the system will perform better. However, I do think that if we increase the storage, "RAM," and processing speed, that the range of problems our intelligence could make breakthroughs for would have to be wider.
Brain, body, environment, relationships, rich stimuli. Makes total sense. That being said, if aspects of cognition can be reverse engineered, then in due course of time we can get to machines that serve the same purpose as (at least) a portion of the brain. Then again, neural networks do that today.
This is a great point.. but isn't it a separate issue to the idea of whether we can create an AGI, or perhaps more accurately , agents? How these agents then engage in the task of info processing is the remit of this video... am i wrong?
This is a brilliant and important idea folks! It has been my belief that the intelligence of the universe, of which we are part, is far more imaginative than a mirror image of itself (as we find in AI systems, robots, etc). The automata that already exist are far more fault resistant and intelligent than we give them credit for.
Francois believes that the way to build AGI should be top to bottom not vice versa. Understanding intelligence at a high level, can lead us in the direction on how to build it.
I think there are infinite scopes of problems and, like inflation, do not all scale at the same rate, so this one size fits all idea of intelligence is probably limited in scope
Dear Mr Fridman and Mr Chollet, Thank you Mr Fridman for your great show. You enlighten all of the world, and I hope Russia will come to its senses, or the government, and allow you to come back and do some shows in Moscow for example. Why am I writing. Well, I’m sure I will get a lot of flack from this, but isn’t it odd that Albert Einstein had so many great ideas working as a patent clerk in Berne, Switzerland? Does anyone see where I am going with this? Imagine being a patent clerk. What do you see all day long? Inventions, ideas, concepts. I’m not accusing anyone of everything, but it wouldn’t surprise me if Einstein lifted a few ideas and put them together in a different way to come up with his theory of relativity. We know that the first thing Isaac Newton did when he became president of the Science Academy was to have 17 portraits of himself made, obviously he knew he was going to go down in history. But probably the most intriguing thing he did was to burn the recently deceased notes of his ardent critic who was also a physicist and philosopher. His name escapes me something like Roberts I believe. But get this, he always claimed Newton had stolen some of his ideas that Newton claimed were his own! I'm not a psychologist, but when glory is to be had, it's often the case that someone will claim things they didn't do, but that’s a good reason for your viewers to do their own research and read about it for themselves. I mean, why would Newton burn a fellow colleague’s notes that could have had valuable information for future generations of scientists and the general public? These anecdotal facts have caused me to re-evaluate my estimation of both Newton… and Einstein.
Intelligence = Brain + environment + body = all possibilities? Interesting. So intelligence is a function of all three and yet these variables are dependent on each other or theirs a correlation among the independent variables? Intelligence(brain, environment, body)
What he said about intelligence being a connection between brain, body and environment is brilliant actually and I have an example: Can anyone claim that a footballer who can pass a ball to the other side of the field to a running teammate while trying to avoid a tackling opponent is not intelligent?
It would seem to me at a minimum fast processing speed, which is develop by repetetive training and natural talent. This seems like crystallised intelligence. If they could do that the first time they attempted it and then repeatedly do it, you could possibly say it's fluid intelligence.
I tend to disagree that there are problems we match to solve. Universe has no problem and it just is. We entertain ourselves and give meaning to our lives by imagining an order and moving our own world towards it. On the path to that imaginary order we “define” problems We try to solve.
This also means that billions of underprivileged humans on this planet might be stuck with great intelligence unable to get themselves near great problems.
It seems that some problem sets have a limit on the optimal solution. For example it seems that there is a limit on the perfect chess game, there is a limit on perfection. Intelligence might be one of those problem sets, perhaps it's more sensible to think of intelligence as a scale up/scale down depending on the problem. Spot instances of compute, that can be applied to a specific task. If 1 human brain is one spot instance, then you'd scale up and down as needed. It seems it gets interesting when 2+ human brains benefit from the calculation of one brain. That's how civilization benefits. It's a bit more ad hoc then than that, but you don't need a superintillegence to do your laundry, only a few cycles need to be used on that, perhaps a larger cycle to produce the washing machine, and a different set looking for optimizations on that process. Where it gets interesting is if you need to invent numerous sub solutions to solve the larger problem, each with its own suboptimizations, or unsolvable problems, so it would make sense for the intelligence to have some reality testing against the physical laws, and the ability to stop it's own runtime, if the compute costs get too great with no progress (frustration/distraction/giving up).
Intelligence is also presumably an expression of many interlinked minds within the environment and historically within the psyche that has evolved over many Millenia. Current predictions about AI seem preposterously optimistic and ignore what it took to create just the human brain.
there's no reason for it to be exponential, maybe as ai gets smarter it's harder for it to improve itself, maybe speed oscillates, maybe there's a bound. There is only one way to know
This seemed like a non sequitur; it didn't explain why he thinks a computer needs a body to be super intelligent. And who said a computer wont have a body anyway? Why does he assume computers will remain housed as they are today. Much more likely they will be fused with robotics. People used to say, machines will never play chess like us... write like us, etc. Growing up my mother used to say, "stop being so enamored with robotics, machines can't even walk, human locomotion is too complex! Want to avoid a robot? Walk up 3 stairs". And today, robots can already out perform us in a bipedal competition. Anyway, this interview seemed to be a case of the same thing. "I don't like the idea of being supplanted by cybernetics, so I will protest... without much of an argument".
His point is that a brain alone, isolated, cannot become smarter just by itself. Any GAI will need to be connected to the world and interact with it in order to develop useful intelligence. I think his point applies to some fields, but not others (like purely abstract fields)
Computer scientists are building in real world bodies or having the system interact in virtual reality. The environment is in fact being interacted with. I don’t understand or agree with the persons point of view. It seems obviously flawed to me.
Being intelligent means you can make predictions and have good memory, so not that hard to imagine a machine , which is not limited by biology, to be exponentially more intelligent. The machine will not be not limited in its ability to concentrate, simulate, memorize, and ideate like humans. It wont be limited to one body, experience, or even one points of view. Read Jeff Hawkins book On Intelligence for a clear discussion of what intelligence is.
I haven't read the book, but I don't think we can say with accuracy that a non-humanoid machine will be intelligent. This is because how we know intelligence is by our expression of it. If a machine does things it will do those things maybe appearing "intelligent", but it would be alien to us in such a way that we could not know for sure that it actually is intelligent beyond appearance. I personally do not think appearance is enough. Many things appear a certain way in which they are actually not.
@@acetate909 I'm not so skeptical about machines eventually having intelligence. After all we are machines, and are intelligent. I'm more skeptical about our ability to identify synthetic intelligence. Using limited definitions like predictive power, and memory seems likely to make us fail more at identifying intelligence. At some point we might have very "smart" machines that aren't actually intelligent at all. But we'd never know if it was true intelligence, or a facsimile of our definitions of intelligence. And I don't think it's safe to assume that if it quacks like a duck it's a duck when it comes to something as complex as computers.
I agree with what all he said. It's so true the meeting of right brain capacity with right problems is important to test your intelligence. IQ is just a number that one can increase with techniques and time but that doesn't make sure that you will be able to solve great problems. I like your every podcast @lexfridman. 🔥🔥
The title was misquoted tbh, its not most people but potential geniuses. Einstein in his Patent Clerk days still hadnt encountered 'his problem', however had extraordinary capability to solve it. But yea, its definitely not most people.
@@idontevenknow3707 Actually, he developed the first couple theories he is famous for while being a patent clerk. It happened not in academia because of his rebellious spirit and necessity in having money to live for, not because of choosing this work as a carrier pass.
This is a clip from a conversation with Francois Chollet from Sep 2019. New full episodes every Mon & Thu and 2-5 new clips every Tue & Fri. You can watch the full conversation here: ua-cam.com/video/Bo8MY4JpiXE/v-deo.html
(more links below)
Listen to it here: lexfridman.com/francois-chollet/ and subscribe on your podcast app.
Watch other clips on the clips playlist:
ua-cam.com/play/PLrAXtmErZgOeciFP3CBCIEElOJeitOr41.html
Watch full episodes on the full episode playlist:
ua-cam.com/play/PLrAXtmErZgOdP_8GztsuKi9nrraNbKKp4.html
Lex Fridman Wow am I first? Give me a shout out Lex! Love the podcast man. Best intellectual podcast out there right now.
I've long thought that there are likely many geniuses out there, but that many factors other than intelligence as well as external factors are required for those geniuses to make a noteworthy impact. The right events need to occur at the right times and in the right order, and they need ample resources, they need enough experience in a field to recognize a big problem. Imagine if Elon Musk had never moved to America, didn't have the opportunity to study physics, didn't have the opportunity to take an interest in electric cars while at university, was never involved with Paypal, etc.
Maybe even more crucial: imagine if he'd never had access to computers as a kid?
And with guys like Einstein and Musk, it's not just their intelligence, it's also massive amounts of determination, work ethic and belief in themselves that's needed for the scope of achievements they've been able to accomplish.
@@kjdskgj OP literally just said that.
@Deep Mind X if you're a child of average or even above average intelligence having a millionaire mother isn't going to push you into musk territory. wealth is definitely a factor, but its not "even more crucial" than whats mentioned above.
4:40 he said many people, not most people :P
Good catch. I changed it.
You also need the will to solve a problem.
Not will instincts
Both..
Instinct leads you towards the goal. Willpower makes you persist through the obstacles to reach it.
@@NorthGermanic maybe then, it's crucial to be careful of what is supplanting instinct and willpower within the hypothetically generally intelligent machine's toolset. instinct is (implicitly, though we may not be so sure) determined by a messy and drawn out web of interactive selection processes on radically different scales than that of the AI's selection processes. willpower is ill-defined for a machine if we keep its explicit necessity for 'impulse' as we currently define impulse, exertion, or obstacles for that matter. I get the feeling these aren't just minor semantic misgivings, but another serious glaring issue with the premise and promise of a general intelligence that doesn't include the endocrine, digestive, and interdependent biological (and societal, if we're to consider Chollet's external environment point) ecosystems that comprise an appropriately defined "intellect".
If I could drive a point with anyone working towards a functionally benevolent GAI, this^ would be it. Upvote if you agree because I'm not going to shop a treatise around or write any grants about it.
@S R I've wondered about that route for sure, but don't have the necessary understanding of the complexity of those systems in their totality, to make any kind of guess on how computable they are or will be any time soon. sounds simple when you put it that way (just "make the virtual environment correspond to physical reality" as we understand it) but if you account for the minute scales of relevant detail, from the specific information encoded in a body's DNA (let alone all of the genetic information and activity of the microflora/fauna inhabiting that body) all the way up to all of the relevant information necessary to process the environment functionally, you're really getting into the weeds there. I'd really be curious of any professional's thoughts on the viability range with our current state of the art and speculations as to future possibility
I need a problem
“Everyone is a genius. But if you judge a fish by its ability to climb a tree, it will live its whole life believing that it is stupid.” - Albert Einstein
This quote is so true. Everyone has a special understanding of something in their lives.
All you really need is knowledge of self.
Once you go back and start thinking about the things that you were always naturally passionate about, you will find it uplifts you and you will thrive.
That just gave me some inspiration to tackle some challenging problems.
Someday Im gonna develop an unified theory in theoretical physics
Now thats a challenging task
Don't waste ur time on some stupid game and try to be a hero.
This is so profound. I totally agree with him.
I think he's trying to say that you need exposure to multiple disciplines so you can form new vectors to attack a problem, and have a problem that needs expertise in multiple disciplines to solve. These days nearly all scientists are regimented into a minute niche in their field and that's the limit of their thought and imagination.
Think of Andrew Wiles - in solving Fermat he had to essentially learn a whole new field of maths, incorporate several completely new proofs that he produced, and still bridge it all back into modular forms to begin to tackle the problem. And even then it took the true masters of those fields to step in and help him stay true to the underlying maths and not go off. Einstein had that as well. Where in academia today is that even fathomable? If it takes AI to be that latent Einstein or Wiles in modern times than that's what it takes, but it's not actually beyond human capabilities WITHOUT AI
I heard Eric Weinstein say something similar to your first point about the hyper-specialised nature of science and a lack of polymaths in modern society. It's not something I've considered before. Interesting
It means we need to get better at collaborating.
Also, it is perfectly plausible that a large improvement processing power will not produce equally large improvements in problem solving - it could ultimately be one of those things where twice the processing leads to a tiny increment in intelligence.
Most likely
This really resonates with the work of Professor Reuven Feuerstein whose theory involves thinking about thinking “meta-cognitively” - as a system with input, elaboration and output phases. The key / main point about the elaboration phase being defined as “awareness of the problem.” The idea being to actively look for problems needing to be solved rather than staying with the status quo / vested interests
Shortened version:
“not-for-profit” // “large body of published, peer-reviewed research demonstrating consistent findings of significant enhancement of cognitive performance for disadvantaged learners”
www.linkedin.com/pulse/recent-audit-cognitive-programs-university-auckland-james-colvin
*****
An important meta-analysis / summary of the evidence base:
eppi.ioe.ac.uk/cms/Portals/0/PDF%20reviews%20and%20summaries/t_s_rv2.pdf?ver=2006-03-02-125128-393
*****
Various documents gathered via google around the 2011 Nobel Peace prize nomination which was instigated by a Professor of “Pedagogical Science” from Italy. The one called “Biographical Notes” was the main letter:
www.dropbox.com/sh/1qro3otdqhc93y7/AABE29WEp3tvhwrXNWqg63Pwa?dl=0
********
A brief section of a much longer paper on the Johns Hopkins University web-site:
www.dropbox.com/s/h20trfc2zpxkwfb/Johns%20Hopkins%20University%20Taunton%20Info.docx?dl=0
The whole paper is here:
archive.education.jhu.edu/PD/newhorizons/strategies/topics/Instrumental%20Enrichment/hur.htm
*******
This is a great video - I’m willing to bet that if you watch to the end it’ll surprise you. Also - when he appears here the Professor is articulating clearly / towards the end of his long life some important and (I think) profound messages:
ua-cam.com/video/lqSQI6VJgLk/v-deo.html
i knew im the smartest dude in the universe, i just have 0 problems
I was going to say something similar and I just played final fantasy 8. Weird world
@@tykepope hey you can still be 2nd smartest
Squall Leonhart deal 🤣
May i apply for the 3rd spot?
@@shubhraneel1666 hop on the podium!
Open access to the library system has been an extension of my learning. That's missing in a post covid world and I agree with this speaker
Interesting thought, and I think he is probably correct about intelligence being a whole system and you can't tweak one thing and think you can predict the system will perform better. However, I do think that if we increase the storage, "RAM," and processing speed, that the range of problems our intelligence could make breakthroughs for would have to be wider.
The number of times Lex plays devil's advocate ♥
Isn't part of intelligence precisely to find the big problems that you want to grapple with?
Brain, body, environment, relationships, rich stimuli. Makes total sense. That being said, if aspects of cognition can be reverse engineered, then in due course of time we can get to machines that serve the same purpose as (at least) a portion of the brain. Then again, neural networks do that today.
Amazing! This is inspirational.
This is a great point.. but isn't it a separate issue to the idea of whether we can create an AGI, or perhaps more accurately , agents? How these agents then engage in the task of info processing is the remit of this video... am i wrong?
This is a brilliant and important idea folks! It has been my belief that the intelligence of the universe, of which we are part, is far more imaginative than a mirror image of itself (as we find in AI systems, robots, etc). The automata that already exist are far more fault resistant and intelligent than we give them credit for.
?
Also potheads the ganja is just too strong.
In my case I’m pretty sure the brain is the bottleneck.
Francois believes that the way to build AGI should be top to bottom not vice versa. Understanding intelligence at a high level, can lead us in the direction on how to build it.
Bottom up approach to AGI is much much harder and nearly impossible as there are too many components in the brain and human body.
I think there are infinite scopes of problems and, like inflation, do not all scale at the same rate, so this one size fits all idea of intelligence is probably limited in scope
Surely like the laws of thermodynamics, the output cannot exceed the input. Unless there is some kind of mechanism for learning?
Thanks
Yeah, Einstein in patent clerk days was a true genius while Einstein in IAS days was a mere celebrity.
Dear Mr Fridman and Mr Chollet, Thank you Mr Fridman for your great show. You enlighten all of the world, and I hope Russia will come to its senses, or the government, and allow you to come back and do some shows in Moscow for example. Why am I writing. Well, I’m sure I will get a lot of flack from this, but isn’t it odd that Albert Einstein had so many great ideas working as a patent clerk in Berne, Switzerland? Does anyone see where I am going with this? Imagine being a patent clerk. What do you see all day long? Inventions, ideas, concepts. I’m not accusing anyone of everything, but it wouldn’t surprise me if Einstein lifted a few ideas and put them together in a different way to come up with his theory of relativity. We know that the first thing Isaac Newton did when he became president of the Science Academy was to have 17 portraits of himself made, obviously he knew he was going to go down in history. But probably the most intriguing thing he did was to burn the recently deceased notes of his ardent critic who was also a physicist and philosopher. His name escapes me something like Roberts I believe. But get this, he always claimed Newton had stolen some of his ideas that Newton claimed were his own! I'm not a psychologist, but when glory is to be had, it's often the case that someone will claim things they didn't do, but that’s a good reason for your viewers to do their own research and read about it for themselves. I mean, why would Newton burn a fellow colleague’s notes that could have had valuable information for future generations of scientists and the general public? These anecdotal facts have caused me to re-evaluate my estimation of both Newton… and Einstein.
Hats off to Newton.
He can't be compared.
Joe Rogan please listen to this
I’m pretty sure he’s telling me that I’m a genius and I just don’t know it. Hmmmmmyesssss, I think he might be on to something 🧐. Next video.
Hæhæhæ
🤣😂🤣
Intelligence = Brain + environment + body = all possibilities? Interesting. So intelligence is a function of all three and yet these variables are dependent on each other or theirs a correlation among the independent variables?
Intelligence(brain, environment, body)
What he said about intelligence being a connection between brain, body and environment is brilliant actually and I have an example: Can anyone claim that a footballer who can pass a ball to the other side of the field to a running teammate while trying to avoid a tackling opponent is not intelligent?
If it's lingard ,then no
@@davidregi7571 dead 🤣🤣🤣🤣🤣🤣🤣
It would seem to me at a minimum fast processing speed, which is develop by repetetive training and natural talent. This seems like crystallised intelligence. If they could do that the first time they attempted it and then repeatedly do it, you could possibly say it's fluid intelligence.
intelligence is not the learned skill. Intelligence is the inherent ability to solve novel problems and learn quickly.
I tend to disagree that there are problems we match to solve. Universe has no problem and it just is. We entertain ourselves and give meaning to our lives by imagining an order and moving our own world towards it. On the path to that imaginary order we “define” problems We try to solve.
This also means that billions of underprivileged humans on this planet might be stuck with great intelligence unable to get themselves near great problems.
I'd say that's being extremely optimistic. Most people live like NPCs
It seems that some problem sets have a limit on the optimal solution. For example it seems that there is a limit on the perfect chess game, there is a limit on perfection. Intelligence might be one of those problem sets, perhaps it's more sensible to think of intelligence as a scale up/scale down depending on the problem. Spot instances of compute, that can be applied to a specific task. If 1 human brain is one spot instance, then you'd scale up and down as needed. It seems it gets interesting when 2+ human brains benefit from the calculation of one brain. That's how civilization benefits. It's a bit more ad hoc then than that, but you don't need a superintillegence to do your laundry, only a few cycles need to be used on that, perhaps a larger cycle to produce the washing machine, and a different set looking for optimizations on that process.
Where it gets interesting is if you need to invent numerous sub solutions to solve the larger problem, each with its own suboptimizations, or unsolvable problems, so it would make sense for the intelligence to have some reality testing against the physical laws, and the ability to stop it's own runtime, if the compute costs get too great with no progress (frustration/distraction/giving up).
Im a clerk. Nothing has happened still yet.
Intelligence is also presumably an expression of many interlinked minds within the environment and historically within the psyche that has evolved over many Millenia. Current predictions about AI seem preposterously optimistic and ignore what it took to create just the human brain.
there's no reason for it to be exponential, maybe as ai gets smarter it's harder for it to improve itself, maybe speed oscillates, maybe there's a bound. There is only one way to know
may be possible to find a theoretical bound. Complexity might explode beyond a certain level of intelligence.
This seemed like a non sequitur; it didn't explain why he thinks a computer needs a body to be super intelligent. And who said a computer wont have a body anyway? Why does he assume computers will remain housed as they are today. Much more likely they will be fused with robotics. People used to say, machines will never play chess like us... write like us, etc. Growing up my mother used to say, "stop being so enamored with robotics, machines can't even walk, human locomotion is too complex! Want to avoid a robot? Walk up 3 stairs". And today, robots can already out perform us in a bipedal competition. Anyway, this interview seemed to be a case of the same thing. "I don't like the idea of being supplanted by cybernetics, so I will protest... without much of an argument".
His point is that a brain alone, isolated, cannot become smarter just by itself. Any GAI will need to be connected to the world and interact with it in order to develop useful intelligence. I think his point applies to some fields, but not others (like purely abstract fields)
Computer scientists are building in real world bodies or having the system interact in virtual reality. The environment is in fact being interacted with. I don’t understand or agree with the persons point of view. It seems obviously flawed to me.
greetings from switzerland ;)
This all assumes that intelligence transfer is lossless from humans to machine
Do podcast with Ted.
That is correct, we at (Fortune 5) find all best talent and slave them. Thanks
Being intelligent means you can make predictions and have good memory, so not that hard to imagine a machine , which is not limited by biology, to be exponentially more intelligent. The machine will not be not limited in its ability to concentrate, simulate, memorize, and ideate like humans. It wont be limited to one body, experience, or even one points of view. Read Jeff Hawkins book On Intelligence for a clear discussion of what intelligence is.
I haven't read the book, but I don't think we can say with accuracy that a non-humanoid machine will be intelligent. This is because how we know intelligence is by our expression of it. If a machine does things it will do those things maybe appearing "intelligent", but it would be alien to us in such a way that we could not know for sure that it actually is intelligent beyond appearance. I personally do not think appearance is enough. Many things appear a certain way in which they are actually not.
@@acetate909 I'm not so skeptical about machines eventually having intelligence. After all we are machines, and are intelligent. I'm more skeptical about our ability to identify synthetic intelligence. Using limited definitions like predictive power, and memory seems likely to make us fail more at identifying intelligence. At some point we might have very "smart" machines that aren't actually intelligent at all. But we'd never know if it was true intelligence, or a facsimile of our definitions of intelligence. And I don't think it's safe to assume that if it quacks like a duck it's a duck when it comes to something as complex as computers.
I agree with what all he said. It's so true the meeting of right brain capacity with right problems is important to test your intelligence. IQ is just a number that one can increase with techniques and time but that doesn't make sure that you will be able to solve great problems. I like your every podcast @lexfridman. 🔥🔥
IQ was meant for little kids in France!
Actually IQ cannot be increased as of scientific evidence we have today, but hopefully we will get there someday.
I watch Rick and Morty, does that count?
AI? Hoops; many hoops!
I fought Shelton was a jenius
He is an akhtar
Copied little bit of evry Jennie and voila a teldon is produced
He married his second cousin and stole patents. Relax..
Well I don't have the IQ points :)) but I'll try.
Please change the intro
im totally content being one unrecognised Einstein, got 99 biches but a problem not one.
Intelligence doesn't come from the brain it comes from god
@@eternallight88 lol don't be silly. I'm not talking about that primitive idea of god theists believe in.
@@eternallight88 I said your premise is wrong.
@@eternallight88 What lol
It comes from carrots, idiot
It comes from god, but god put it in the brain. So idk what you’re really trying to say here
Such a terrible metaphor...
The title was misquoted tbh, its not most people but potential geniuses. Einstein in his Patent Clerk days still hadnt encountered 'his problem', however had extraordinary capability to solve it. But yea, its definitely not most people.
@@idontevenknow3707 Actually, he developed the first couple theories he is famous for while being a patent clerk. It happened not in academia because of his rebellious spirit and necessity in having money to live for, not because of choosing this work as a carrier pass.
@@sergey.pinigin good point