Mystery of Entropy FINALLY Solved After 50 Years? (STEPHEN WOLFRAM)
ฝัง
- เผยแพร่เมื่อ 10 ก.ค. 2024
- Please check out Numerai - our sponsor @
numer.ai/mlst
Patreon: / mlst
Discord: / discord
The Second Law: Resolving the Mystery of the Second Law of Thermodynamics
Buy Stephen's book here - tinyurl.com/2jj2t9wa
The Language Game: How Improvisation Created Language and Changed the World by Morten H. Christiansen and Nick Chater
Buy here: tinyurl.com/35bvs8be
Stephen Wolfram starts by discussing the second law of thermodynamics - the idea that entropy, or disorder, tends to increase over time. He talks about how this law seems intuitively true, but has been difficult to prove. Wolfram outlines his decades-long quest to fully understand the second law, including failed early attempts to simulate particles mixing as a 12-year-old. He explains how irreversibility arises from the computational irreducibility of underlying physical processes coupled with our limited ability as observers to do the computations needed to "decrypt" the microscopic details.
The conversation then shifts to discussing language and how concepts allow us to communicate shared ideas between minds positioned in different parts of "rule space." Wolfram talks about the successes and limitations of using large language models to generate Wolfram Language code from natural language prompts. He sees it as a useful tool for getting started programming, but one still needs human refinement.
The final part of the conversation focuses on AI safety and governance. Wolfram notes uncontrolled actuation is where things can go wrong with AI systems. He discusses whether AI agents could have intrinsic experiences and goals, how we might build trust networks between AIs, and that managing a system of many AIs may be easier than a single AI. Wolfram emphasizes the need for more philosophical depth in thinking about AI aims, and draws connections between potential solutions and his work on computational irreducibility and physics.
Show notes: docs.google.com/document/d/1h...
Pod version: podcasters.spotify.com/pod/sh...
/ stephen_wolfram
TOC:
00:00:00 - Introduction
00:02:34 - Second law book
00:14:01 - Reversibility / entropy / observers / equivalence
00:34:22 - Concepts/language in the ruliad
00:49:04 - Comparison to free energy principle
00:53:58 - ChatGPT / Wolfram / Language
01:00:17 - AI risk
Panel: Dr. Tim Scarfe @ecsquendor / Dr. Keith Duggar @DoctorDuggar - วิทยาศาสตร์และเทคโนโลยี
![[FULL EP.01] LOL: Last One Laughing Thailand Thailand หัวเราะทีหลังปังกว่า | Prime Thailand](http://i.ytimg.com/vi/t4KunBRKn3s/mqdefault.jpg)
![[Full Episode] MasterChef Junior Thailand มาสเตอร์เชฟ จูเนียร์ ประเทศไทย Season 3 Episode 5](http://i.ytimg.com/vi/g5wpgeAIzMs/mqdefault.jpg)
Fantastic conversation. It amazes me how Stephen, while sometimes difficult to follow, from time to time he pops out with concepts and analogies simple enough to make sense of, and yet very powerful. Kudos to Tim & Keith for bringing this conversation!
I use Mathematica every waking hour of my life. I have been doing this for the last 30 years. Everything I have done in Physics is largely due to Mathematica. Hats off to Wolfram Inc.
Allow me to say: Uh-oh :)
Free the software my dudes. Donate to the free-libre software devs. Euthanize the rentiers. (Anything I can do with Mathekrapital I can do with Python + Nim).
Give an example of how u use it please
@@gemthomas calculating the time before he can go home from the job
wow, what's it like being so profoundly limited?
Holy cow what an amazing conversation. I feel privileged to be in the modern world and have access to this kind of brain bending, brain expanding stuff!
Listen to Nature and really expand it.
@@goldwhitedragonnot a very bright comment. As if Nature has a monopoly on what's real when all that listening or observing nature is going to do is raised a lot of abstract questions that require abstract tools to make any progress with. It's not like you look and immediately it's all visible. You have to penetrate it with a penetrating and contemplative mind. Also, a small mention on 'false causation', one of the biggest fallacies that emerged from 'listening to nature'and which took thousands of years to untangle ourselves from.
@@512Squared That's right. A contemplative and SINCERE mind, not one that says "not very bright."
@@goldwhitedragon deconstruct your own original reply and you will clearly see why it was not very bright while claiming to be bright. You took my reply that was calling that out and did more or less the same moral superiority bs. It's the insensitivity that is hardest to see. And yes, it has to be called out clearly, otherwise wolves are mistaken for sheep. And you projecting that back to me because I called you out on it? I say it again, not very bright. Funny, it was just one word 'really', but it revealed so much about how you look down on certain kinds of knowledge and inquiry, construct hierarchies that don't really exist, so yes, not very bright. But hey, you can step into clarity any time you want. Don't make it about me. If you to be want be enlightened, then drop the silly superiority and appreciate and celebrate knowledge and insight wherever it is found. Otherwise, grow a thicker skin to criticism rather than trying to make it about someone else. I doubt very much you'll take this in good faith, but good luck.
@@goldwhitedragon I mean seriously, a SINCERE mind would not make the original comment you made.
Please someone kindly point to the minute in this video where something new about entropy is actually said. While the man talking seems remarkable and very interesting to listen to I failed to identify how rambling about AI had anything to do with proving the second law of thermodynamics.
Wonderful content. I have listened Dr. Wolfram on a youtube video of MIT Physics LLM talks and I was shocked and aspired by his way of handling topics. Here is no different, brilliant questions, brilliant ideas not only technical but deeply philosophical ones. Thanks for bringing this to our scope.
"The most recognisable and brilliant scientist alive today"
What a crazy statement to start a podcast with. I really debate whether I can bother continuing if it starts in such a dumb way. I can imagine it will be full of mumbo jumbo that makes little sense with an audience that has zero background to evaluate or understand what he is saying, and takes everything as gold since they consider him the "most brilliant scientist alive". As a working theoretical physicists, I often end up frustrated after listening to stuff like that.
Just because you’re a theoretical physicist doesn’t mean you’re not an arrogant asshole
Most working physicists don’t like Wolfram since he is in their field and makes way more money and has more notoriety than they could ever dream of. Or maybe there’s other reasons but he seems not well liked by the community
@@Bob-qq4is Do you have any background to evaluate whether anything Wolfram says makes any sense? Or just impressed by these podcasts?``
His whole fanbase are people that do not have any background to understand any of it. He spreads his ideas on podcasts where people can't tell whether it makes sense, instead of in peer-reviewed journals where it is evaluated by experts. Ever wondered why?
Jealous because he makes more money? Jim Simons is one of the richest people in the world, he is probably worth 20 times Wolfram. If not more. Yet his work is highly respected in the math and physics community.
Wolframs early work, before leaving academia, was pretty solid. His software, Mathematica, is really good. I use it every day. But his recent work is extremely disappointing. It's a massive combination of buzzwords, extreme claims and no genuine will to back up any of these claims.
He has been invited to give seminars or even online discussed his ideas with scientists. Whenever anybody asks into any detail about one extreme claim, he reverts to making 5 other unrelated extreme claims about his "theory". This is not in any way serious science.
If you are impressed by big words you don't understand on podcasts, and make up your opinion on the science community based on it, I have stuff to sell you man.
If anyone ever creates a caricature of Dr. Wolfram it would be Dr. Wolfram on a Podcast being asked a single question and then just not stopping answering the question for and the multiple side effects of the question for a couple of days and the host struggling to slip another question in to at least guide the conversation into a certain direction.
He's talking about fascinating stuff, though! I loved it.
@@epajarjestys9981 Parsing problem, much?
Interesting observation. When your brain is that vast and you don't share at scale* you end up with cortex entropy. The mind is at it's best when you are teaching others. It become natural wanting to just tell all. Grand unification. *At scale meaning wanting to be able to expand though at extreme rates. Out mouths and the listeners ears are too slow.
@@alabamacajun7791 ob the other hand: if can't understand something briefly, you haven't um bereits it thoroughly.
Try Chris Langan, America's smartest man and his CTMU. On Kurt jamungals show.
I understood it. I would love to learn more about how philosophically we could match the pace of the techical side. I do think he has cracked the entropy befuddlement.
Amazing conversation. Highest level content, presented in very understandable way without going too low. An intellectual pleasure for sure.
For 50 years my understanding of entropy has been breaking down and I can't put it back together.
The best thing in this interview is the pure joy in all your faces
It is truly a privilege having access to such luminaries! But of course, I enjoy these kinds of conversations wherever and whenever they happen ;-)
just a humble path integral in a big big ruliad 😁@@nomenec
That is why I watch Wolfram on video--his genuine enthusiasm and joy is something that can't be copied or faked
@@Dessoxyn he does seem a little behind the times on this entropy topic, diffusion models for example basically entirely hinge on least action principles which is tied at the hip with model entropy and bayesian inference etc. Seems like its been a crystal clear concept for computing since the 80s at least, ignoring wizards like Shannon way earlier who gave us the whole field of DSP.
@@AndreAnyone It's more likely that evolution and cognition exist because of it. It's like THE rabbit hole of physics and biology haha.
Mind-blowing conversation, enjoyed it a lot, thanks!
fantastic material as ever, thanks!
Top Quotes from Stephen Wolfram:
"I think I now really understand [the second law of thermodynamics]. I began to understand it back in the 1980s."
"The interplay between the underlying computational process...and our limited computational abilities kind of at the top."
"Computational irreducibility means you're just stuck going through every step in the computation."
"Science isn't going to be able to give you all the answers."
"Language is our encapsulation of things we care about versus things we don't."
"The workflow that I see really emerging is...I have a concept in my mind, and I want to get my computer to do it."
"Most of [the computational universe], we humans don't yet care about."
"If we leave AIs to their own devices, they're just gonna go out and explore other parts of the Ruliad."
"When you talk about goals...I can have an external theory of what their goals are, and so can I for your average AI."
"Right now, we're not having AIs connected to us...we are progressively connecting [them] to more and more actuation systems in the world, and we should care about that."
"I'm in a sense more concerned about the lack of depth of understanding on the philosophical side than on the technical side."
"It's inevitable that it's kind of like there's this thermodynamics of AIs."
"Is it easier to manage a billion AIs than it is to manage one AI? I think the answer may be yes."
"When you talk to people who were there when it started...they're always like, well, I think it's right, but we're not quite sure."
Paraphrasing: “Thousands of years of philosophy and, in the end, we have to write code”
@@RickDelmonico holy crap, that is what I call committing to a joke!
@@DavenH how old are you, like 12?
The first thing Steven did was commit to particles.
It can be shown, however, that all the particle-like phenomena can be explained by using properties of the wave functions/state vectors alone. Thus there is no evidence for particles. Wave-particle duality arises because the wave functions alone have both wave-like and particle-like properties.
Feynman's path integrals are not infinite, they're fractal.
hahaha 😂😂😂
Inspiring interview. It stimulates thoughts about how, as a writing tool, AI-assisted search can be good for looking up who in history had similar ideas. That way we can sort of cite ideas we had that mayb were not entirely original or perhaps needed more credibility. We may be standing on the shoulders of giants. But it doesn't do much good when we don't know who they are.
What "doesn't do much good" is when we don't understand what they say.
Very beautiful concepts and models presented here. Thank you for the inspiration.
I don't understand how the mystery of Entropy is FINALLY solved. This video doesn't contain a single well explained idea, rather just "all-connected-to-all-how-fascinating" type of thing
You are correct. They were just bullshitting you. ;-)
Yes yes yes. This is hitting all the right approach points... after years of reading physics and working with reinforced machine leaning, these ideas get to the heart of the universe and how we experience it. I will definately be getting Stevens book.
Nah, I don't get it. The more I listen to that guy, the more convinced I am, that this is
a) a very bad case of confusing the map with the terrain / the model with reality
b) taking an arbitrary model and trying to make everything fit at all cost
c) leading no where at all
d) me being much dumber than I thought
or a mix of them.
I'm mildly amazed intelligent people think big talking heads on a screen is better for communication of an idea than visual aids and even dynamic visual aids that could be programmed................
I just see big talking heads on a screen and too much verbiage (verbal + garbage).
@@commentarytalk1446I’m sorry you find it too hard, that’s quite understandable, the sandpit is over there in the corner, do enjoy yourself…
@@paulklee5790 First of all "You're not sorry" so as such you're a liar. Secondly, you don't seem very adept at reading comprehension and would score zero if you read again. Thirdly you're a time-wasting troll using a mild form of insult to sound superior/humorous because of a sense of over-defensiveness. Try to take a good look in the mirror if you even human.
The part where he compares "internal experience" of a laptop with a human experience is just one example that shows how shallow and one dimensional some of his takes are.
I agree about him having just a single model (computation) and trying to force everything to fit it. For some problems it's great, for others - hopeless.
His bridge between temperature level of language models and thermodynamics state changes at the beginning, is a great demonstration of the level of thinking he does
I think applying the idea of phase transitions like this to other things is a great idea.
"please be as technical as possible, we want all the details"
boffin: "I really don't think you quite want that but I will go a certain distance"
ha ha ha
Incredible discussion about communication from different locations and how knowledge of events can be different for each observer. Reminds me of how polarized light is changed as it passes through a filter. Two people can see an event differently, but when they attempt to communicate what they saw, they end up communicating the event as the same. Their communication is transformed to align with the location it presently resides.
You gotta wonder if knowing too much leads you off into crazy land.
It does. I have this video to prove it.
Wait I need to see this video loool
There is only an infinitesimal separating the rational, irrational, and transcendental.
Great discussion!
Every now and again youtube suggests a video that im actually interested in. This is one of them
Wonderful interview with Stephan Wolfram. Interesting to hear him discuss the interplay between the second law of thermodynamics, i.e., entropy and how that relates to challenges in really understand how LLMs work.
We had quite different takeaways from this video. He claims to have had some epiphany about how 2nd law really works (an understanding that has evaded him for last 50 years until very recently). But imo he didn’t give us any of that explanation. He just derailed off into different random stories. If you got some understanding on how he interprets the 2nd Law from this, please share.
@@timjohnson3913 sabine h. covered her view on the topic a few weeks ago. the gist of it being that any initial state, no matter unlikely it is (all molecules in one half of the box) is just one of many, many, possible states and no less likely than any other. i tend to agree with her take on the subject.
@@timjohnson3913 What I understood is that Wolfram explains that the reason we perceive that entropy must always increase (that time’s arrow moves forward) is because our reality interpreter (our brain) processes information packets at speeds in the millisecond range whereas the processes that are actually taking place occur at speeds in the nanosecond range. He implies that this conclusion is a natural outflow of the analysis of how ideas are packaged and transmitted from one space/(reality interpreter) to another.
@@adminnvbs9166 and that makes any sense to you? Sure you gave a “what” explanation but I don’t see any connection to heat dissipating or broken eggs not coming together nor do I see any “why” that makes what you said a good explanation
@@timjohnson3913it has to do with “coarse graining” of the many possible configurations of micro-states. The coarse graining corresponds to the observable macro-states. The observable directionality of evolving macro-states is due to there being many more configurations of micro-states corresponding to later observable macro-states. For example, there’s more ways that the molecules of an egg can be observed to be scrambled than there are unscrambled. Roughly speaking…
Stephen this is a very insight producing presentation on Entropy and how it is affected by the limits of the resolution of our perceptions and measurements. Thanks for making such a great contribution to our understanding of this subject.
Surprised to have found a likeminded thinker. Thanks for sharing!
Legendary interview.
On the idea of computational irreducibility becoming obvious, this is probably a thing that's happened since we got computers. We've had practical experience of trying to write programs to do non-trivial things, and that's given us a feel for what kinds of things can be computed quickly and what can't.
I'm not sure that all that he's "discovering" is "specifically accurate" as much as it is "conditionally accurate". Suppose you had a barrel of the infinite. The way another intelligent species "begins to make sense" of what's in the barrel could conceivably result in a completely different ruliad than the one that he is "discovering" based on our current humanized conditions that represent our attempt to make sense of the same barrel of the infinite. In other words, not only is observer bias a possibility, but our inevitably "locked-in" anthropicly-limited perspective may produce a different set of explanatory rules. If we had a chance to compare our ruliad with another ruliad produced by another "alien" perspective, they would probably be different in some respects. The question is, would they have the same bones, or would they be fundamentally different? Seems like an important question, but it may be a bit too "stringy" to matter...meaning too "stringy" to be testable. Still, I wonder if perception affects a ruliad--specifically, the description of a ruliad? SW has already talked about how humans are invariably "locked in" to having a limited perceptional capacity, and we have to (whether we recognize it or not) limit our "intake of data" to what seems to "matter" to us. As has been stated before, no "model" of reality ever is reality...the amount of information is exceedingly infinite, and we can only ever focus on an excruciatingly small subset of that data. Anyway, interesting stuff.
The implications of computational irreducibility are not obvious to most people even now. Even people who think they understand it frequently slip into reductionism.
A good example of this is pro-life people who insist that life begins at conception and that a fetus is not different from a baby - that you skip immediately from not-person to person, and pointing to a reductive element, like a heartbeat, as evidence of a sort of instantaneous mapping of reality.
This view is not consistent at all with the reality Wolfram is describing or with the concept of computational irreducibility especially as it is so common in biological systems - that organisms have to pass through time in order to become what they become and that this process can’t be skipped over.
Gödel walks in: "Hold my beer ..."
also Gödel: "My friend, can I have some of that stuff you smoked, @gcewing?"
@@fenzelianEntropy is much more concrete in the case of genetics than heat. Think of the landscape model that virus physicist Peter Schuster describes in explaining the natural algorithm that allowed proteins to evolve into the efficient optimal folding machines that they are. A similar landscape model can be used to describe the universe of all possible gene configurations and organize them by the genetic differences (as distances) between them. For organisms to evolve, their gene structures must migrate across generations through this landscape. The law of entropy applies to this landscape because less gene arrangements produce successful complex life forms than produce successful simple life forms. And, likewise, gene arrangements that produce successful simple life forms are outnumbered by those that produce unsuccessful life. We know this based on entropy. Within this landscape may be more than 10⁶⁰⁰ possible arrangements. However many, it is a very large number so big that not even a smidgen of them could have occurred across 4 billion years even if every organism that ever existed in that time frame was genetically unique. Yet, in defiance of the static entropy characteristics of the genetics landscape mapping all life, gene configurations managed to randomly migrate to ones that produce ever increasingly complex life forms. All I can say is bull dung; it didn’t happen. Do the math. A natural selection genetic algorithm is not sufficient enough to solve such complex problems approximately optimizing dozens and dozens of interrelated phenotype systems in single organisms. Bull dung. No algorithm could solve such a search problem.
I can safely say that no one else can blow my mind like Wolfram does.
Watch some Michael Levin interviews and come back to me. No disrespect to Stephen Wolfram but Levin is at the bleeding edge of life science research.
great conversation ❤
Fantastic explanation.
An amazing achievement which is rarely mentioned is Cedric Villani's 'Proof of non-linear Landau damping and convergence to equilibrium for the Boltzmann equation'.
Does the time-independent Schrodinger make any dostinction between past, present and future?
Is the Poincare recurrence theorem and Boltzmann H - theorem a form of time reversibility?
Hi and Thank you for this engaging interview with Dr. Wolfram on the fascinating subject of entropy. I appreciate the depth of the conversation and the insights shared. However, I'd like to suggest one thing for future content:
Title Alignment: While the title is certainly eye-catching, it might be seen as an overstatement. Dr. Wolfram's insights are valuable, but the term "solved" may not fully represent the complexity of the subject. Perhaps a more nuanced title could better reflect the content.A more accurate title might be something like "New Insights into the Mystery of Entropy: A Conversation with Stephen Wolfram."
Thnx.
@@adamluhring2482 I guess you took the shut up and calculate directive too seriously. Most theoretical physicists do believe the evolution is unitary, even if it is still unclear how the observer enters into the picture. Basically, information is conserved, that is why black holes look so evil.
I would like to see how his model of a "changing hypergraph" deals with the fact that there is not a distinguished time direction in Minkowski space.
As a retired professor, I admired Stephen Wolfram, Donald Knuth, Stephen Timoshenko, etc... all these people are pioneers in their fields who operated at a higher level.
Truely phenomenal is right! Thanks for this talk gents !
Fantastic interview! A follow up one, very soon, as in a few days, would be a great idea! Probably, the most qualified person to hear on the so called dangers of LLMs.
I think its a strech say this is a conversation about science. At best its about philosophy of science. He is definitely a very smart and successful person. His theories are too hand wavey for a physicist. For e.g. his claim about space composed of atoms and our aggregate observation manifests as Continuum. A claim without a formal theory does not forward scientific progress. For e.g. I think a legit proposed theory like, loop quantum gravity suggested spacetime as discrete. But its predictions does not match observations. About rulial space, lets assume 10**600 is true. What then? Its as pointless as many worlds theory, something that can never be proven and doesnt make any observable predictions.
We will never be able to predict the universe hence the Uncertainty Principle. We are the part of the Entropy, Universe itself. We can always measure simulation that we are running but simulation can't predict itself with deterministic certainty. Standard model cels are bashing their heads trying to fine tune the particles yet they didn't really provide any useful conclusions which could be used today.
Just amazing. Every second of it. What a mind.
"In the end you've got to write code" spoke to my heart
When I hear Stephen I feel the concepts are re-statements of existing theories using his own language rather than being truly original. Stephen seems to be making parallels and restating concepts in his model. It's like a second order process description with a it's own naming convention. It's either brilliant or trivial and my limited mind can't quite grasp which.
I read one of his papers on the subject instead of listening to this talk as there's countless interesting diagams illustrating his investigations and I think it's a promising direction and tool-set to use. I would say the title is bogus (the author of the video confirms this in the comments sections elsewhere as it's a hook for yt sensationalism for views). However from the paper the most interest connection made was the concept of the complexity/entropy of a system and it's size AND how the observer is able to conceptualize it... ie limitations apply at both ends. I think that's illustrated well in the diagrams, its almost intuitively simply so probaby relatively robust and a nice addition or extension to former ideas about entropy - at least within my own very weak understanding of the subject! So yes it's not a solution as the title suggests but I think the tools of investigation expand understanding of 2nd Law of Thermodynamics in a very practical application.
As I pointed out before: Visuals aids are worth a lot more than talking heads to transmit ideas! The video's title is poor and the content visually is inadequate.
@@adamluhring2482 yes, I really quite like Stephen, and think that his intuition regarding the importance of understanding ourselves as specifically evolved creatures who observe the world in a very specific way, is a crucially important one. That said, though, I have to sadly agree with your synopsis of his work - despite the steady stream of talk about having finally discovered (solved, reconciled, understood, explained etc) some area or another of metaphysical inquiry... well, there's a _very_ steady stream of talk. I'm just not sure that there's much physical exercise going on to support all that hot air.
if you want original try David Deutsch
Wolfram once wanted his theory to be peer revied. I remember the review board gently called on him to find a better proof of his theories. Even in this video the only qualification is half a million reviews, but he didn't show any satisfaction.
I think you've done it, actually.
His unique take take on things might be an explanation.
Or it might be a re-framing.
In what sense is it useful? Does it improve our observations? Our understanding? Our predictions? Our control of natural phenomena?
The Positivists hoped that, beginning with mathematics, and progressing to physics, and by way of physics to chemistry, from chemistry to biology, from biology to psychology, from psychology to the social sciences, and then on to the arts, a logical language could eventually unite all of human experience and knowledge into a consistent system to understand, predict, and control everything. Sound crazy? Anyway, that was destroyed by Kurt Gödel in the 1930s.
Wolfram has a clear hypothesis: reality is fundamentally a binary computation. This is why he's on the sidelines of science. The correspondence between physics and mathematics is well known but not understood at all. Every scientist knows it's a mystery. There are many divergent opinions about it. They divide along the lines of whether mathematics is real or not. The promise of at once solving this deep mystery, and revealing the unifying principles of physical reality in the process, is the source of Wolfram's sensational appeal.
Stephen Wolfram is an experimental mathematician, an honorable profession. It's a type of inventor. Buckminster Fuller was one such. He also did good work. He also was a character. He also had fanatical followers.
What a great conversation to be a fly on the wall in.
Excellent content
So beautiful lecture
Easily the most fascinating guest you've ever had. I could listen to this man all day. 🙏❤️
I want to agree, but I had to stop after the first 10-12 minutes after hearing not a single new idea.
"In the end, you have to write code" this is a great notion. I came to this conclusion many years ago and Mr. Wolfram's work has only come to vindicate many notions I have come across in my search for meaning.
⁰⁰0⁰0⁰⁰⁰⁰00⁰00⁰⁰000000⁰⁰⁰⁰00⁰⁰⁰00⁰⁰0⁰⁰⁰0⁰⁰⁰⁰⁰⁰0⁰⁰😊
CTMU answers all questions. All.
Embedded in what language is are common concepts also embodied as word-groupings with connections.
A.I. is picking up mathematical patterns which, by virtue of linkage, comes close to modeling the deeper aspects below language, the hidden stuff which are assumptions we /brains need to comprehend and communicate.
Wow Wow! It was truly a treat to hear him talk about ideas that he is thinking about right now. It is like an artist describing each brush stroke, as he paints. I did not realize how much he is working on AI and the extent to which his language is playing a role in the development of Chat gpt. I think his new ideas about computational physics will move physics and engineering forward in leaps and bounds. I think his kind of thinking will be the key to modernizing how we live.
What agreat conversation. Would love to hear both of the other conversations mentioned in this interview (Bach and Friston). You guys are doing great stuff. Many thanks 🙏🙏
Yes to Joscha Bach. Such amazing minds we are privileged to experience! Thank you!
In terms of introducing Wolfram to Friston, I'd be curious to see Michael Levin as a 3rd party discussing some of these topics.
I emailed Levin yesterday to suggest that he speak with Wolfram.
I wonder if Friston and Levin would be able to get a word in!
It's funny how his reply in the video was basically to say "I find it's not worth talking to other people anymore."
Lol
As always, a very interesting conversation, but how do you test any of it?
Minite 18 is the best explanation of computational irreducibility i've ever heard
Thank you for another great episode.... I remember the first computer I programmed 1974. I worked for the vendor (NCR) and I felt so lucky that I had 8k of core on my computer whereas our customers hadc4k. The good all days.... thank you for reminding us older folk how lucky we are today.
We are all very lucky until the nukes start flying...
Awesome interview! It's funny how he talks about ostracising the AI at the end as an alternative means of punishment, reminds me of how banishment was a historical penalty for people of significant means, and in some ways it's considered worse than death - you don't get to become a martyr or a victim, you just become irrelevant. Also lets you hold being accepted back into the fold as a carrot.
That's a great point to keep in mind.
What a trip...
In a good way.
The more I hear about new Physics' ideas the more all that 'The Secret'/'Create your own Universe' lore makes sense.
Good job.
I met Stephen Wolfram once and I can't stand the high opinion of himself he has. The last thing this man needs is to be introduced as "perhaps the most recognizable and brilliant scientist alive today".
Like too many exceptionally brilliant people, he's got an exceptionally inflated false ego. But at least he offers something for real substantive thought and discussion. Anyone else who HASN'T done VERY IMPORTANT fundamental theoretical work, probably doesn't have anythging original or substantial to offer on the topic, yet likely has an unmeritedly inflated false ego...
Every time I listen to Stephen Wolfram I get excited. I get so many ideas I want to explore. Not just his ideas but he causes me to think more deeply about the things I’m working on. Wow great interview.
Wow, the statement from Dr. Wolfram that all of his thoughts and writing on such are designed to be overlaid upon a computational model. That kinda changes everything I have heard or thought about Dr. Wolfram's writings because for me personally I might have written something off as "oh well this is kind of like X" but does X also work as a computational model? That's so cool! No wonder he loves cellular automata and of course all of the legendary software contributions over his career, we can start to grasp the convergence. Thank you so much for the interview!
We are bounded in whatever ways you claim, but we also bounded by the limits of our language to explain, describe, and even discuss the matters you are addressing. With those limitations (constraints of discussion) you do a remarkable job with those difficulties against you.
Eh...It sounds to me like he took the second law and all it's subtleties (nevermind QM, nevermind chaos, nevermind probability) and transcribed it into a different language without resolving those subtleties.
One big thing, I think, is we HAVE to reconcile the kantian nature of our observations (or the observer problem in QM) and whether space-time are primal or not. What can he predict here? Otherwise it seems like "what's the point?🤷"
I'll have to read the book. But I do love the whole idea of all this complexity stuff 😍
Wow, can't wait to listen to this. I'll urgently watch today after I finish some work. Stephen Wolfram is always fascinating to listen to, even though I always feel I don't understand everything he says.
Just watched and absolutely loved it!
I'm not quite sure if Mr. Wolfram got into the details of the book he wanted to go into. It seems a bit like that strand of conversation got diverted. I should read the book. 📖
"I always feel I don't understand everything he says"
Embrace that feeling, it will lead to you learning things you otherwise wouldn't or inspire you to figure out what you don't understand.
@35:00 this is what Dr Hofstadter posited in "Analogy as the Fuel and Fire of Thinking." Assume language enforces a degree of regularity and conformity to qualia
Stephen, (extension of prior comment): How do we know whether the most important information is in the vast, eternal bit; vs. the tiniest, most discriminating bit? You mentioned that as we populate more and more of rulial space we cease to exist. I very much agree. As a species of observers, over our evolution, are we reaching ever more discriminating levels of perception? Increasing neural efficiency, or organization - order?
Can't wait for the Observer theory from Wolfram. One thing I wished you asked him about it is: How are you going to deal with the classic problem of self reference - the limits and structural bounds your theory will impose on our minds are themselves the result of these limits. You have to be outside of these limits yourself to know those limits are there.
there is no plausible solution for this problem, it's a philosophical problem from the domain of epistemology and there is no way of neither resolving it nor going around it. any theory can hence only hope and strive in compensating for it, incorporating this as a fact and simply moving on.
@@horasefu1438 Moving on exactly where though? Deriving a superluminal travel or finding out what's at the center of the black hole? I think this is still being stuck in the same old Platonic cave, it's just today we frame it in terms of bits and computation, and superpositions because we now have QM. Whatever the label for the problem or its source, it's still a problem that needs to be addressed. If not, it's either Pragmatism, in which case, it's Wittgenstein's “Whereof one cannot speak, thereof one must be silent.” or admittance that it's a cool story, perhaps a good plot for the next incarnation of the Matrix trilogy.
"I have a suspicion that there is a way out of this mess" is all he could say.
Next book.
Hilarious to hear him talk about 'If I made up a word and no one used it' and proceed to use the term rule-i-ad as if it were something. :)
It's a perfectly reasonable term. I've used it shortly after I heard it defined the first time, and I KNOW thousands of other people use it in the same conceptual sense, assuming "concept" and "sense" aren't qualia.
I love hearing Stephen Wolfram talk, even in the moments when gets a little incoherent :D
Also, as a linguist, I can appreciate the way you phrased your bit about the language game; you seem to have a good grasp on the matter
Are there real scientists who take all his stuff actually seriously?
no
AI will be crucial in helping us understand the universe. A different perspective, a different kind of intelligence will prove key. My pet theory is that Many impasses in different fields have their limitations rooted in the inherent thinking processes, language, logic of biological humans.
This is so cool. LLM predict/select the next word. The talk, e.g., about Tesla world-models for cars, is that the observers are the cameras with high resolution dynamic inputs, being retained in memory for perhaps 10 seconds of 60 frames per second.
The highway "world model" is trained (like the LLM) to predict all the bits in the next frame, up to maybe even the next full second of frames. So the model predicts a future. It can also predict 100 other possible futures (quickly), select the "best/most likely/desired/or goal-oriented" future, and send appropriate commands to the auto control system to realize that future. Meanwhile, it can also prepare appropriate (re)actions in the event that any of the other possible predicted futures actually occurs .. in the very next dynamic "frame" or ground truth it receives from the vision system.
If you think about it, this is what a driver does constantly, watching the cars, guessing what they might do next, deciding on what he or she is doing next; doing that, while worrying about the other drivers; and so forth on and on. To the extent that works in silicon, IMO, that really is totally amazing!
Sounds fascinating.
Just wondering whether Stephen is inspired by the philosophy of Giles Deleuze? A lot of the concepts he puts forward (e.g. the ruliad) sound similar to Deleuze’s rhizomes and plane of immanence
I just asked Google bard this question, it could be hallucinating though 🤷:
"The ruliad is a set of all possible computations, while a rhizome is a network of interconnected nodes. Both the ruliad and the rhizome are non-hierarchical and non-linear, meaning that they do not have a central organizing principle. They are also both constantly evolving and changing, as new nodes are added and old nodes are removed. However: there are also some key differences between the ruliad, rhizomes, and the plane of immanence. The ruliad is a more mathematical concept, while rhizomes and the plane of immanence are more philosophical concepts. The ruliad is also more focused on the future, while rhizomes and the plane of immanence are more focused on the present."
Delusional you said?...😂
0:56: 🔬 Dr. Stephen Wolfram's insatiable hunger for knowledge and research on ancient manuscripts and early scientists' conception of the universe.
16:50: 💡 The second law of thermodynamics is an interplay between the computational process and our ability to observe things.
34:28: 🧠 Minds in rural space can communicate through concepts, which act as packaged particles of information.
51:09: 🧠 The speaker finds it interesting that theoretical debates often need to be translated into code to be understood and grounded.
1:08:09: 🤔 The speaker discusses the future of AI and the challenges of AI governance.
Recap by Tammy AI
My personal points of resonance:
• At 50:07, the host mentions *"the sea of disorder".* Half an hour before TH-cam recommended this video to me, I had written a comment elsewhere on TH-cam mentioning "the sea of disorder". Good job, TH-cam algorithm!
• Like Dr. Wolfram, I enjoy GPT's talent at inventing beautiful and accurate brand new words very much too.
thank you dr. stephen wolfram, you are one of the greatest minds of our time, a true inspiration. I love Wolfram Alpha, it was extremely usefull during my studies.
TJ;DW: Clickbait title combined with a strained effort to make the 2nd Law of Thermodynamics somehow relevant to trendy AI buzzwords.
indeed
It's actually even worse. It's not just a clickbait bull session, this entire discussion of thermodynamic "irreversibility" is confabulated nonsense. To physicists, it is NOT "an enduring mystery" that subatomic physical laws are reversible, while "the behavior we observe in the universe is manifestly irreversible". All it takes to understand, for example, the behavior of a container of gas molecules is to ignore Wolram's mystifying word salad conflating entropy with encryption, and focus on the nuts and bolts.
Entropy is simply a measure of random disorder. There is nothing "irreversible" about the initial configuration of gas molecules in the box, and there is nothing that prevents them from returning to that initial state at any point in the future. The reason the gas molecules' movements APPEAR to be irreversible is because of combinatorial explosion: There are so incalculably many potential arrangements of the gas molecules in the box that it is extraordinarily UNLIKELY you will ever see them assume the same configuration twice (or any other specific configuration for that matter). That is the so-called "mystery of entropy", and it was well understand far earlier than 50 years ago.
AGI doesn't need survival instinct, survival can be instrumental to any goal.
Shit that's a profound insight, where we usually assume the survival instinct would take precedence over all, yet we have the concepts of self-sacrifice, heroism, or suicide that disregard that apparent sense of self preservation
@@Cabildabear Self-sacrifice developed in humans because of it's tribal history, it's still evolutionary in nature, not on personal, but on tribal level.
@@XOPOIIIO Its a beneficial trait to do the best thing for the continued propagation of your genes. If that means sacrificing yourself in a way that benefits your progeny, then it makes sense fro ma selfish gene perspective.
what the hell are you trying say?
@@6ixpool520 Exactly
Good stuff
The description of entropy increase around 17:00 in sounds like Bohm's idea of implicate order transposed into a computational frame of reference.
Good afternoon to you three. 😀
Wonderous shared discussion.
Fascinating indeed Stephen,
only know a wee bit of your work so far.
Has enabled my brain to upgrade alot of inner software and a sense of better understanding and might add direction.
Truly Grateful.
💜
Although understood pretty much nothing, still listened to it, being 1000'sX better than any music.
Mr. Wolfram is an absolute gem of our times.
th-cam.com/video/-wZ359spIVY/w-d-xo.html
25:00 "Abstract" is the word
36:00 An abstraction
36:30 A neuron
It was a great cocktail party conversation -- but without cocktails.
Exciting to hear what Stephen Wolfram has discovered. I believe he is at least on the track that entropy is like encrypted information rather than actual randomness, that it looks random because of the difficulty of measuring all the details, if that's still his view.
So disorder is encrypted order? Next time someone tells me that I'm out of order, I will just reply 'No, in fact I'm in order, just that I utilize privacy technologies'.
That's how randomness, noise, mysteries and magic appear to us. Anything of which we have insufficient information to form an illusion of having a grasp of.
Randomness is relative. If you know all the bits in a string it's not random to you. You just can't usefully compress it to communicate beyond the Shannon bounds. That's also the essence of the 2nd LoT. To usefully communicate a high entropy state of stuff in a box (up to anti-de Sitter space) you have to grossly over-simplify --- meaning you've chunked a lot and used macro variables, not micro, and many micro can realize the same macro.
It has nothing to do with computers and hypergraphs _per se._ It is computational by default, by definition, since it's just mathematical description. That doesn't mean anything profound. How else are you going to describe physics? With poetry? (Been done.) The math always works if you get the inputs right.
To me, fwiw, Wolfram comes across as a child nerd with bald hair. Higher IQ than me, have to say (though that's no guarantor of having useful insights). Not a grifter, but someone who is too easily surprised by trivialities.
I'm not a hater though. I like Stephen's enthusiasm. I wish he'd help "solve world poverty" though, which is not operationally a terribly hard thing, it is mighty politically hard. He needs to learn basic MMT , which he will not get chatting with the likes of NNT, instead see smithwillsuffice.github.io/ohanga-pai/questions/1_basic_ohangapai/ for some MMT basics.
("Solving poverty" --- i.e., solving the hard problem of instigating the political will to eliminate poverty via fair distribution of necessary output --- is a surer path to a Nobel Peace Prize than any hypergraph theorising is to a Physics Nobel or Math Abel. Justsyain. If that sort of prize is what tickles your ego.)
compressed might be better term than encrypted
Disorder and randomness are two terrible ways of understanding entropy. It boggles my mind anyone can think of science with this understanding of entropy. Entropy explains how one must expend energy to harness energy, thus eliminating the possibilities of perpetual motion. Wolfram starts to get into the Carnot cycle, talking about irreversible processes. If your take away from entropy is, "the world is devolving into chaos and randomness", you've entirely lost the plot. The harnessing of energy is always escaping our grasp, and the more we harness the quicker we speed up this process.
Think of harnessing all the energy released by fossil fuels. You have to collect the molecules and the heat from the atmosphere, and compact it back to the oil or gasoline. If we burned oil as energy to perform this process, we'd be contributing to the problem we're solving. How much oil has to be burned to reverse the burning process? You're better off cutting the losses and developing forms of energy generation that don't heat the atmosphere.
You gave it the wrong title - apart from some general remarks about Entropy - apparently all had forgotten that you wanted to talk how Stephen solved its mystery and instead showed how he got surprised by the capabilities of AI and LLMs like the rest of the world when ChatGPT came out - LLMs are completely random structures for any observer that does not know how to use them but they needed Gigawhathours to be created in their training process - they are ENERGY converted into INFORMATION - and this is what Stephen points out that ENTROY is not random for all observers but Energy converted to complex information - but still how he solved ENTROY is still a Mystery after watching this video
I think you need to look at that thought again. Stephen discusses entropy as a phenomenon of observer theory, and the emergence of the "law" of entropy as a result of the relationship of the observer to information rather than a phenomenon of the universe. I guess you could look back to seeing Maxwell's Demon as an observer with a different state than us.
That’s a revolutionary way to pronounce transition. 👌🏻 He’s an absolute treasure.
Have had trouble getting into Wolfram's lectures before, but this is very packed with accessible concepts to think about. Wonderful.
Fascinating conversation, I was particularly interested in the AI risk chapter, specifically the actuation concept. There used to be about 25 million "classical" programmers, now, pretty much anyone with the internet can program, to some degree using AI. Classical programmers learn ethics and responsibility alongside their skills development, but the newer AI assisted programmers have little to none of these accountable traits.
Stephen Wolfram correctly identified the scope of actuation, and how that could represent a significant challenge but then the conversation moved on to an AI developing internal agentiveness. But what about the AI assisted programmers using AI to develop systems capable of producing unstoppable systems? - either by accident, or deliberately. An AI doesn't need to develop its own agentiveness to be agentive. Agentiveness can be synthesized with automation code provided by naive, or disenfranchised, AI programmers.
I've come up with examples of potentially uncontrollable AI automations that could wreak significant damage to society and none of them require anything more than consumer hardware, open source models and basic automations.
If AI can help us do the right thing, it can certainly help us do the wrong thing and AGI is not a prerequisite for either.
The incidence of human malfeasance is statistically certain, therefore the incidence of human-prompted AI malfeasance is statistically certain. And, unfortunately, just becasue ethics and moral standards are (or at least were) part of programming study, they've obviouskly not made much of a difference over the past 70-80 years in terms of the proliferation of fundamentally immoral ventures such as health insurance, most of the MIC, energy companies, or the rest of the commons that's been subsumed by private enterprise.
Your point is spot on, I just don't think the oncoming negative consequences of actuaction are fundamentally different, they'll just be much worse than we've already got. Maybe that offers potential for positive solutions, if we work hard enough for it.
Making useful programs needs guidance. Generative systems do not have agency, only statistical reproduction. This means the average Internet user attempting to make something new with machine learning will not be able to utilize it in a very productive manner. Likewise an experienced programmer can make good use of such tools.
The primary goal in development of this tooling is primarily about lowering pay for skilled labor and secondarily about being able to pass legal responsibility for decision making to a mechanical process. Both of these things are not good.
@@orbatos I agree with your points on programming, I didn't mean to suggest that everyone can program effectively however the trajectory of the trends are fairly clear, we are lowering the bar for both skill and accountability into a broader set of users. Similarly automation can mitigate incompetence and failure rates (code, compile, test, repeat - until success). These are trajectories that I believe will lead to the development of significantly powerful tool that could be almost impossible to mitigate. If only one of these tools is developed we could have a major problem.
Meta is rumored to be releasing Llama 3 as an open source model to compete directly with GPT4 and I postulate this, with the safely rails inevitably removed will be sufficient to advance these trajectories well into the danger zone - on consumer hardware too. The consumer hardware part is important as it removes a barrier for malware producing systems to replicate exponentially.
I also agree that lowering the pay for skilled labor (pushing skills out of the market when we need them most), and especially the automation of decision making are not at all good. To me it seems like a lot of trajectories are heading in the opposite direction from where they need to be going and corporate goals and ego are at the wheel.
A nice conversation. I find it necessary to also mention the mathematical universe hypothesis, as it is extremely similar to the Ruliad idea. Perhaps one theory emphasizes computation while the other focuses on mathematical structure, with time being a factor in one and not the other.
Although time can be a factor in math structures too, so it's more of a different approach to the same thing maybe,
The two ideas are indeed similiar, but the mathematical universe, is a subset of the computational universe proposed by Wolfram...as the Ruliad contains all possible logical systems, including mathematics.
He has sections of his studies on meta-mathematics that you should look into that goes into the underlying ideas. "How Universal is the Concept of Numbers" is a good lecture to watch from him, i think you'll be surprised.
@NightmareCourtPictures
1st statement) Isn't it the exact opposite? I think Ruliad is a questionable and incomplete application of MUH. It is important to understand what math is. Beyond numbers and shapes, math studies abstract structures and their relations. There is no universally agreed-upon definition of mathematics, but it includes logic and many related areas. Math is turning complete (pretty sure, the internet seems to agree).
2nd statement). I took a look at the lecture, and he mentions some interesting ideas, but I don't see how it changes any of my points. I'm sure he is a smart guy. The main problem with all mathematical universe hypotheses (MUH) and "related ideas" is that they are unverifiable as far as I see. I'm glad Wolfram is trying to develop these ideas further, to perhaps arrive at something practical. That would be great. I see his ruliad as MUH with some more assumptions (making it a subset of MUH).
The concept of computational irreducibility is interesting, but I think it's likely wrong. It's a message of its own, though I'm happy to share (and complete the thoughts for myself). The reasoning might take quite a few words to explain, but is clear and I think it comes down to definitions and beyond those the implications. Further, his reasoning appears handwavy and as far as I understand and I think no rigorous math exists to prove his claims.
I may have mixed in some of my thoughts with those of the mathematical universe hypothesis, or maybe not. Either way, the above reasoning stands.
@@valterszakrevskis To your first statement : the mathematical universe, implies that the universe is made out of numbers...mainly to explain why universe gives us fine-tuned constants, and using this premise as a way to excuse the fine-tuning problem without invoking something like god.
The computational universe is similar except it is fundamentally computational...not defined out of numbers, and that lecture (how universal is the concept of numbers) is an important lecture in explaining this and why the Ruliad concept is not mathematical universe theory but *way more* than that.
To your second statement and third statement: Computational Irreducibility is an observed phenomenon of Computational Systems. He wrote a book called "A New Kind of Science" which you should watch the series he made on it (it's 14 parts 2 hours each sorry...but its a great watch) In that book he goes into pain-staking detail of the computational experiments he did. If you study any kind of computability, this book is like a right in one's face "Proof by Exhaustion" way of showing that these things are real and not hand waved. It's ironically hard to actually prove that any of it is wrong...because those experiments clearly show that it is just true...self-evident as he would say.
Studying the stuff for 2.5 years now and applying the work he laid out in applications for my own purposes...I would say with great confidence this is true, and it'll only be a matter of time before people start catching on.
Cheers,
My , at present, most concise definition of the (beautiful!) Second Law is, 'Good goes to Bad, unless. Good is created, or maintained, locally, but then more Bad is created elsewhere.'
I continue to glory in His Creation!
Eg the recently-encountered Carnally's Law. Wow!
So the old “how likely is this hand of cards” chestnut is a kind of entropy question. When the cards are “uninteresting” then we forget what they are and think about the number of hands that are, to our perception, interchangeable with that hand. So that hand doesn’t seem unlikely. But when it’s a royal flush, there aren’t many hands that are, to us, interchangeable with it.
@PhilHibbs
At The Circus in Bath there are 108 acorns on the parapets.
The Georgians loved gambling and card games and the card game Canasta uses four packs of cards so 108 cards.
So pattern matching and finding combinations in reality from a to b in time could be viewed as a form of SCSC (Satisfiability Conditioning and Secure Computation)
Always a pleasure to listen to Steven!
I loved mathematica's website. I browsed its wiki whenever I was curious about a concept back in the early 2000s. The wealth of interlinked knowledge was huge.
Mathematica was pretty good as well. But that was it. A wiki and a software. Then there was the web counterpart of the software back in the early 2010s.
But then I started seeing wolfram in the media. He kept talking about how he was part of great discoveries. Whenever there is some discovery, there he is, as if he is taking credit for it.
Is it me or is stephen wolfram just being delusional?
It is not you.
Same. I remember trying to come up with equations to make it draw things that I wanted. It was very basic all those years ago.
I don't think so. He happens to have the recognition to be able to talk about those sort of things, and his writings have put him in the spotlight.
But just because he's the only one you hear talking about those things, does it mean aren't any others?
Yes
This conversation starting point problem - refers to one formulation of LEonard Susskind formulations regarding entropy, that observed randomnes is described as a thermal equilibrium as a result of many tiny fractalised components which are too hard (useless) to account for. And next question derived from it ---> what if we can account them precisely and somehow practically apply that hidden part of information in real science. That`s how i percieve this extraordinary topic.
Fascinating
OMG I am 45 minutes in and I'm having so much god damn fun. I love MLST and I love Wolphram even more. This conversation is literally heaven on a plate. I'm salivating. I would have paid for this. Hundreds of pounds (don't get any ideas). Truly, thank you
Word to the Mothership
Some observer isn't born/created without some kind of it having some sort of risk assessment mechanism. I was playing with programming some particles, by reading the book The Nature of Code, and you quickly get to the part of particles avoiding other particles. If you want some agent to behave as some kind of entity, add some risk perceiving mechanism.
Wowzers, what an introduction
I came to hear Stephan but subbed because of Tim and Keith. Class acts!