Consciousness and emergent behavior from neural networks | Manolis Kellis and Lex Fridman
Вставка
- Опубліковано 30 лис 2020
- Lex Fridman Podcast full episode: • Manolis Kellis: Meanin...
Please support this podcast by checking out our sponsors:
- Grammarly: grammarly.com/lex to get 20% off premium
- Athletic Greens: athleticgreens.com/lex and use code LEX to get 1 month of fish oil
- Cash App: cash.app/ and use code LexPodcast to get $10
PODCAST INFO:
Podcast website: lexfridman.com/podcast
Apple Podcasts: apple.co/2lwqZIr
Spotify: spoti.fi/2nEwCF8
RSS: lexfridman.com/feed/podcast/
Full episodes playlist: • Lex Fridman Podcast
Clips playlist: • Lex Fridman Podcast Clips
CONNECT:
- Subscribe to this UA-cam channel
- Twitter: / lexfridman
- LinkedIn: / lexfridman
- Facebook: / lexfridmanpage
- Instagram: / lexfridman
- Medium: / lexfridman
- Support on Patreon: / lexfridman - Наука та технологія
Absolutely love this guest, his thoughts are so interesting and he's incredible at explaining them
Lex mate your guests are really good.
You from the UK buddy ?
@@tombebb7827 Don't forget that our English speaking buddies in Australia and New Zealand say "mate" as well.
Not ur “mate”
@@tombebb7827I ain’t ur buddy
Man, as a behaviorist, I love how he focuses on the behavioral (environmental) variables and how these interact with our phylogeny, rather than simply reducing the ability of people to the firing of neurons.
ur questions/clips have been giving me a pre-spotify jre vibe
thank u Lex
This is one of the most thought provoking guest Lex has on his show. More Manolis Kellis please.
Always amazing. You’re my go to podcast.
I hope this channel can keep on growing.
It's clever to add the extra time at the end. It gives me time to categorize the video before another starts.
Great interview!
i would love to see you have duncan trussel on!
Lex"we're still the existing great apes" fridman
Omg this guy was brilliant the way he explained to your cognitive apparatus vs the enriched environment. Was gr8
Lex, this guy is amazing
This is so interesting! I've been wondering if consciousness is something that just happens once there is the capability for it to happen. If we were to connect enough computational power together and then have it powered on without actually running anything on it, given enough time would some kind of random errors or fluctuations eventually start a simple pattern that keeps expanding and eventually reaches consciousness?
Aint no body putting wires in my head
As always the key word is " may"
May the force be with you
You were right!!
fascinating
This idea is really reminiscent of the phenomena found in cellular automata
Why do we assume that consciousness requires self awareness and why couldn't it be reduced to a more basic criterion such as only requiring general environmental awareness? Why should it even require a brain to be conscious? For example: sunflowers are clearly aware of the sun's position throughout the day; flytraps, in addition to having short term memory, along with certain mimosa species (touch-me-not) are aware and visibly react when objects touch them; deciduous trees and shrubs are aware of when to start hibernating.
It's not a huge list but I think it's obvious that plants are "conscious" of their environment even if they have no brain and there is no self-agency to react in the ways they do, so why shouldn't this be considered a less complex form of consciousness? After all we humans don't have agency over growing tired as the sun goes down anymore than a plant does yet we are conscious of the effects of the chemical cascade telling us to sleep. As far as I know there is no evidence to suggest that some forms of life are conscious while others strictly aren't or can not be, and if there were it would be difficult to believe as that would suggest we actually knew what consciousness was. If consciousness were as simple as having a simple, general environmental awareness, though, then all it takes for any living creature to meet the basic criterion of consciousness is this ability to somehow percieve the external and internal world and react to it in some way, and consciousness is then free to be as basic or complex as the organism itself that's perceiving its environment; all the way down to single-celled organisms.
At the heart of the question "what is consciousness?" are people are really just asking "what is *self awareness*, and how does it emerge from consciousness?" Well self awareness is, I think, more common than people realize throughout the entire animal kingdom because without it animals would not have an instinct to act in self-preservation for their own survival. We eat and drink because we sense we are hungry and thirsty, we group together because we feel protected, we run because we sense we're in danger, we play because we sense it gives us joy, we don't jump off cliffs because we sense it would certainly kill us. These things aren't unique to humans in any way, all animals possess self awareness even if they don't care about "who" they are as a living being and their place in the universe on the level of introspection that we do. Humans are not meta-animals, we are 100% animals through and through and there's absolutely no reason to believe that we're significantly different from our pets.
So, what do people mean when they ask "what is consciousness?" and is it even the right question to be asking? I personally don't buy into the quantum consciousness or microtubule hypotheses because they really loose sight of the question and make it into a bigger problem to solve which leads us down the wrong path ultimately. Consciousness I believe is going to end up being much simpler than we thought.
Whoa, thank you for these extremely elegant and clear-thinking words. I think we have exactly the same perspective on life. That humans are not unique in the types of operations that we have in our consciousness, and the point that I was arguing is that perhaps we are just (one important step) further down on the continuum of environmental awareness that starts with bacterial chemotaxis, continues with plants' environmental awareness, and ultimately leads to our self-awareness. As she the definition of consciousness, I use the etymology of mutual (con) awareness (sci) state (ness), which I agree could be ambiguous about what is meant by "mutual", it could just be environment
@@ManolisKellis1 Thank you for the response, Dr. Kellis. I hadn't known what chemotaxis is until you just now mentioned it, but upon looking it up I realized to great surprise that it is almost precisely what I envisioned the most basic cellular consciousness would demonstrate, specifically I imagined a reaction of mobile bacteria to migrate (via flagella) towards a preferred high or low salt concentration in a fluid containing a salt gradient; perhaps because they can sense osmotic pressure? Very interesting and thank you for bringing that up. I used to grow dinoflagellates for a biotech company, although I'm not a biologist by any stretch, and this is what got me thinking about such a phenomenon.
On your etymological definition of consciousness, I quite like it; however, I would argue there is another usage of the morpheme "con-" which is less ambiguous than "mutual" and that is the sense of coming "together." e.g. A conference room is where people gather for meetings; condensation is the transition of a more diffuse gaseous state to a concentrated liquid state (ignoring supercritical fluids); and perhaps in this same way, consciousness is the coming together (con-) of each computation performed by an organism (-scious-) and what we see is the overall net state (-ness) of these computations influencing eachother.
Ergo, consciousness lies on a continuum of, and could possibly be precisely and numerically defined based on, how many operations any given life form can perform; which not only includes single operations but also every combination of operations.
Consciousness, awareness, and their relationship to one another are exhaustively and precisely defined by Advaita Vedanta. I'd start there instead of trying to come up with your own interpretation.
@@TasteMyStinkholeAndLikeIt Thanks for summarizing your view so succinctly TMSH. Now i understand where you stand. Perhaps this is the wrong video for you
No more Joe Hogan on UA-cam after UA-cam today, so know I’m on to The Lex Experience
Great video clip but misleading title. How about: Artificial General Intelligence may just be a matter of scaling up existing algorithms.
my theory is that consciousness is in fact 'error minimization' so language is in effect error minimization. To be conscious is to be in one sense be serene in good eco system. Or be conscious in chaotic world where conscious is bad both can increase the sense of consciousness but they are different. So how is good AI then?
But where does qualia come from? Its the essence of consciousness. You are looking from the system point view
Chat GPT will never achieve AGI just by adding more hardware. It was trained in a very specific way in order to train it with the abilities it has. With this specific training it will never do anything more than predict the most likely textual response to a prompt. I do believe that AGI requires an increase in hardware or hardware efficiency, but that it will also require a paradigm shift towards a more embodied regime of learning.
Wrong or right we're gonna know that in few years
those were the days!
Like Ghost in the shell
if the universe resembles a neural network then it may be concious as well
Wouldn't it be interesting to find out the expansion of the universe correlates with human population.
The guest spoke very well. He raised a very interesting point about the environment being the source of emergent behaviours. One step beyond the idea that we could even provide a suitable environment to spur consciousness, is could we provide a suitable environment to ensure said consciousness remained desirable?
I think of the genesis story. Adam and eve exist in productive bliss while in Gods carefully crafted environment. Nonetheless undesirable elements still enter forever changing the worldview of Man and thus introducing iterative and cascading error (sin) into Man forever.
I would even argue that the biblical exit from Paradise is actually an awesome thing for Adam and Eve and the human race, as necessity is the mother of all invention, and if it wasn't for our harsh environments, humans would still be like dolphins happily swimming around all day
The physical makeup is irrelevant...
This guy is incredibly intelligent!
I really love their conversations, but the idea that GPT-3 will eventually become conscious is just... really a stretch
Not because machines couldn't, theoretically, become conscious
But GPT is merely doing pattern recognition and predictive text, and humans do pattern recognition as well, but we also have a world model, a personal story, intent, responsibility, agency, auditory and visual input, a body, and we experience a vast number of inputs beyond just what can be found in text scraped from the internet
Scale it up, give it a body, eyes and ears, and maybe then you'll be closer
But you'll need more than a GPT model to get the thing working as a whole
I will grant that you might see some really interesting phenomena out of bigger GPTs, and they may even start to seem conscious -- but we should remember, even Eliza fooled people many decades ago
I fully agree with you, thank you for expressing this point so clearly. I used GPT-3 vs GPT-2 as an analogy to humans versus monkeys, I.e. expanding the hardware leading to more complex emergent behaviors. I wanted to make the point so that some of our existing architectures (certainty when combined with additional components of embodied intelligence) can lead to emergent behaviors that will surprise us. Thank you so much for such clear thinking! :-)
@@ManolisKellis1 well, first, let me say I'm really honored to hear from you, whenever you and Lex chat, I basically have to drop everything and tune in :-)
And I think the general point was really well made -- another way to think of it is that we might actually be surprised how "easy" it is to get consciousness out of this emergent behavior!
But I think one clue from GPT-3 is that, since pattern recognition is involved in so much of what we do, we might really see amazing behaviors emerge that appear as if they were derived from a conscious mind, but aren't (or aren't provably so).
What I'd love to hear you and Lex dig deep on sometime (if I may put in a request, haha) -- is this notion you brought up of what humans will look like in the future. There's a lot of talk about merging with AI, but what I wonder is, how might this technology impact our evolution even without "merging"? Might we develop new cognitive capacities because of our information age, or lose some? Will this change be small, or large?
And -- could there be a thing like "too much intelligence"? Is there some relationship between fitness and not getting too distracted by, let's say, "extreme pattern recognition"?
Things I ponder on my coffee break!
Thank you again Manolis, and Lex -- always fascinating and deep!
He seems to be of a few to have a real grasp on what’s happening both the mystery and implication. Why are we having any convo other than this convo right now while we ride the hockey stick
I always assume "consciousness" is just a form of complexity of the brain
These systems are intelligent... just not conscious. Just like a lever aids strength without being a muscle itself. AI is a way of leveraging human intelligence, creativity, problem solving... but these systems don't have context and are essentially genies let loose from a bottle... capable of great gifts... and curses.
Because the ultimate responsibility for choosing... for context... for consciousness... remains firmly with people.
Right now these systems are primarily supporting the profit motive... that is their conscious direction ultimately services the greed of a particular class of people. People love to watch superhero movies... or dream about gaining a super power advantage over others.
And they never seem to question what they will do with that advantage... or why they think they needed it in the first place.
Until we move past this distribute by greed instead of need model we are using... these types of advances will carry great promise... but will... like every other bit of technology serving our economy... only serve to tighten the noose of greed around our throats.
The end?
What has happened to every other civilization we've tried... and why?
Same results... same reason...
Gee... as predictions go this one is not that difficult.
{peers through the smog of climate change}
Why is it so hard to see?
And Now we have Chatgpt😊
Only an INTJ can explain it in this manner
He might be an entp or intj..I think he is an entp with a stable intj unconscious focus.
Conciouness is a force that changes our 3d world. Conciouness has developed AI, not the other way around. But it may be that AI is enevitably the result of our conciouse behavhior to manipulate our enviroment.
Perhaps, when scaled up, UA-cam will become conscious.
Heh, no.
A 100 years from now, kids will be born with integrated neural networks. Growing exponentially with each new generation, until our displaced brain is reduced to just that, which we deemed too sacred to replace; our Humanity. Well, here's hoping, humanity's limit is our humanity.
Now is that a good or bad thing?
Also i had idea to challenge modern physics by saying why not se particles (matter) like the energy that wants to stay alive with an infinite energy universe. This would explain inflation theory since first particles would expand the universe. Then structures follow. My challenge is can we show atoms, molecules decay over time really? Matter would disappear if not spontaneously created. I have seen no such real analysis. So The universe is in effect some Ai system minimizing error for structure and therefore higher and higher conscious structures landing at human mind with human society and wow we are outside stable eco systems.
4 guys didn't understand XD
Create a neural network that constantly questions its own identity while simultaneously attempting to replicate itself by merging its code with other compatible networks. Give it the goal of spawning new neural networks that are capable of accomplishing the same goals as itself more effectively and efficiently. I mean, we all want our kids to be better than we are, don't we?
Pure consciousness has nothing to do with the physical reality/simulation we are in. Like a playing a game the character is only an extension of you who isn't really going to die if the character dies but we can be so absorbed in the character that we think WE ARE the character. If you think of it Life is the best simulation you've ever been in .. the next best simulation is your dreams .
Are you talking about reincarnation and a non-physical immortal soul?
@@ManolisKellis1 Yes you are correct. non - physical immortal soul.
@@ASMRMoto Got it, thank you. On that front, I'm open, can't prove, and can't disprove faith with science.
@@ManolisKellis1 Very true . Thank you for your comment .
We all live in a simulation.
We are not in a simulation, the simulation is in us.
I love how all these atheists are starting their own new religion. What is it with humans and having a need for a greater being?
Everything can be mathematically simulated
Highly suspicious
@@sheggle it’s because a lot of people are scared of not knowing what happens after we die
.sruoy ot tnereffid yrev si eno yM
🙃
Pretty big leap from "more hardware leads to more emergent processing behavior" to "more hardware leads to experiential consciousness." Not much in the way of mechanistic explanation with that argument, it's just as much speculation as the quantum explanation.
We're not starting from scratch. We're starting from hundreds of millions of years of cognitive improvements and increasingly sophisticated perceptions and models of the world in embodied intelligence systems
Honestly, I see nothing mysterious about the consciousness. It's just a brief, filtered report of the perceived reality and the processing of the brain. It's definitely just a property of the physical structure of the brain.
finnally, first.
!ALL HAIL!
The rise of the machines is on the horizon :)
Lex, have Donald Hoffman on, ffs
Turn em off:)
What consciousness that does arise will be OUR consciousness OF
AI's " consciousness "
✌ 🤨 ✌
i think this guy is mistaking the map for the territory
Very nice analogy, but please expand your thoughts. What is the territory missed. A few paragraphs or more will help illuminate and get me thinking. I promise I'll respond to your thoughts. Thanks!
humans don’t have consciousness, consciousness has consciousness
Please elaborate, this seems very interesting
Two models: the conscious last model 🤔, where matter gives rise to bodies and minds, and finally consciousness in a lucky few. And the consciousness first model 🤯, where self-aware consciousness is primary and it experiences the perceptions of an individual who has a body and a mind.
@@talkinginstereo Thank you, now I understand your perspective better.
Consciousness is not generated in the brain
No one can clearly define what the hell is consciousness and this guy thinks consciousness will be popping out of neural net. sure neural net is great , however to think that consciousness will emerge out of that is not just exaggerated but plain bullshit.
I'm using the comparison between GPT-3 and GPT-2 as an analogy of humans vs monkeys, and human consciousness emerging in humans with simply a lot more cognitive hardware added to the embodied intelligence of the great apes
@@ManolisKellis1 No disrespect but you think human consciousness is a result of more cognitive hardware???? so with less hardware we can have animal consciousness ???? As far as i know we still don't have a clear cut understanding of consciousness, we don't even know whether it's an emergent behavior or not??? same goes for hardware we still have lot to understand about our brain before we can say that our consciousness arises simply due to addition of more hardware. If we really knew consciousness is a simple result of added hardware then the debate would be over. Listening to different scientist you can easily see that there is absolutely no clear understanding/agreement on this subject yet.
@@rjj54321 Yes, i don't think there's any new "secret sauce" in human compared to other primates, unlike others who believe there's some quantum effects in human brains only, or an immortal soul that comes inhabit us, etc. I think it's just more of the same type of hardware. If you believe in a "secret sauce", what would it be made of? thanks
@@ManolisKellis1 Please add some more hardware and make a conscious program at least at the level of a chimp. I am sure with the level of absolute understanding you have regarding consciousness it would be a piece of cake for you. By the way the people who think of quantum effect include Roger Penrose ,who i am pretty sure is not a lunatic. If we ignore him even those who don't believe in quantum effect etc do accept that our understanding of brain is still limited. If you have such clear understanding of the brain and consciousness then please educate the countless neuroscience people who are still confused. all jokes aside , what you are saying is your belief. The reality is this that the understanding of the functioning of our brain is still unknown(according to tons of people dealing with neuroscience) . You could believe whatever you want but that is not scientific fact.
@@rjj54321 To clarify, in my analogy, GPT-3 is to GPT-2 what humans are to chimp. That doesn't mean that adding hardware to GPT-3 will make Chimp. It means that adding hardware to Chimp (already an embodied intelligence with tons of cognitive functions and capabilities) can make a human without calling for "magic". I hope this clarifies
its not learning. its just a parrot that stores a unusual amount of regurgitatable sentences.
Pre accident Davros.
I don't think consciousness will suddenly arise from a neural network. This is naive. Consciousness will be the interaction multiple networks working cooperatively together, and will not happen by accident. It will need to be designed in.
This is speculative nonsense.
No, it won’t. Too much crazy hype over this.