Artificial Einstein: Did AI just do the impossible? 

Will AI ever discover a new theory of nature? Let me know and don’t forget you can win a real meteorite 💥 when you join my free mailing list here 👉 briankeating.com/list ✉️

I think AI has the greatest value to analyze huge data sets to find unknown relationships which lead to new physics equations, new chemical compounds, and predictive genetics. But I am probably wrong.

I am working on quantum physics LLM. What I find the hardest is the reinforcement part as the initial training data pretty much is standardised. That said I want to try to reinforce it with literature related to a more funkier side of possibilities such as anti-gravity or maybe time travel.

Protein folding is probably one of the most important areas of exploration because of its implications in the restoral of health, and repair of injury.

Nice work. I have to point out that with any emerging technology there are problems that occur and may not be able to overcome in the near future: AI’s: 1. Can hallucinate. 2. Can’t articulate the reasoning or process it used to get their results. 3. Can be biased. 4. Can provide different answers to the same question. 5. Can provide responses repetitive to a theme rather than providing independent responses. 6. Someone will always have to validate AI results. There is an inherent risk to not knowing how a solution was arrived at. 7. It’s lack of predictably is it’s strength but also it’s most critical weakness. In certain areas, it will revolutionize the world, in most other areas - not so much.

AI has demonstrated a capacity to hallucinate, which I consider that a promising start.

let it learn all the differences between bosons and fermions on all known levels, including all the known equations like schroedinger. then make it amplify the differences and have it construct a new system of the particle world.

I like to think that at some point AI would invent it's own language for calculations due to a sheer time and resource economy. If it would be asked to decypher one of these symbols of it's language into our regular math, it would take years just to read a result.

Mathematicians come up with new things all the time, it doesn't mean it reflects reality. You can create an elegant answer to some problem in physics that probably explains the issue real well mathematically... it doesn't mean it exists. II means you are good at math.

Click bait title and thumbnail. Blocked.

Watch the bit about 7.8 billion Nobel prize winners on a trolley rail. "Gpt chat solves the trolley problem"

"The greatest shortcoming of the human race is our inability to understand the exponential function." - Prof. Al Bartlett

What a time to be alive

Can you elaborate a bit more on the graph at 11:40? Is there a relationship between graphs u, c, v, and p, or were you just showing how there are multiple graphs and elaborating on a few select moments of the c(t,x,y) graph? I ask because it also seems applicable to financial domains

Been watching physics maths and cosmology on RU-vid for a decade. So glad to find your channel at last! Top quality!!

IIRC, there was a neural network that, after observing/processing videos of pendulums' motion, derived the basic laws of motion; i.e. F=MA I too hope that there will be similar discoveries made by neural networks for other branches of physics.

By far, what you are tapping into a potential is the key for advancements in all fields. I have often thought of the impact on medical research, much like what you are presenting here for fluid dynamics. Excellent presentation and accolades for creating a teaching assistant. I wish I would have had such a thing getting my engineering degree in the 80s, LOL.

Something to consider, perhaps is that mathematical equations are describing an idealized model universe. Since AI is using real world models for its simulation, it could eventually be more accurate as a description of the universe than an equation.

As a physicist as well, I can tell you we are very short of training data on almost everything. Experimental physics is expensive, the physicists that do this are rare as hens teeth, so if we think we are going to solve anything we need to observe it first and put in a form to train, amongst what is not that thing its NOT as well, and all its nuances. Im excited, but hype is hype for AI

Whats amazing and has already been duplicated in a way with sora is I dont think you need code for every type of physics simulation. I have a physics simulator in my head right now and couldnt begin to tell you exactly how it works. I can imagine a red apple, I can change its color to blue, I can throw the apple against a wall and watch it bounce off or explode. And all I did was have certain hardware and a knowledge base of what ive observed over my lifetime. Sora is the same way. It doesnt have a rally racing game engine built in yet it can create and simulate what will happen to a shockingly accurate degree, just like my brain. Some physics wont have to be coded in, we can simply train it against the physical world.

Hasn’t the AI simply found more efficient linearisations by exhaustive exploration? You’re correct to point out that the solutions might/probably won’t generalize beyond the training data - so might not be as useful where high precision is required, but terrific for making special FX for movies and education where time is money and the audience can’t tell the difference, and nobody’s life or job is on the line.

It Is funny how realities generater of physical matter can now have a simulation of the great simulator lol Running elements through various critical extreme states & different lattus structures is definitely something interesting. To even streamline cost in what's worthy of actual testing is a huge benefit in the search of exotic materials Even as a tool for we the people to build agents to simulate industry and markets efficiency and functionality will be a great aid in how we decide to build out future infrastructure to accommodate. It's a lot of tough decisions to be made and we need better tools before we tackle a lot obstacles

Predicting a simulation result is great. This is similar to what humans can do when predicting everyday outcomes of some mechanical event (some item falling for example) and react to it before the event happened.

is this software that you can download (fluids simulation?)

The mouse pointer icon in the thumbnail is a great way to gatekeep people who smoke too much weed. Like me. For like a minute straight lol.

I want to barf every time I hear a smart person say "climate crisis"

Even a model that could output the smallest of novel processing/'ideas' would be a game changer itself, even if its something we already have proof/definitions/laws of or know. It would have to be a scenario where it was trained only with all the tools necessary to derive the answer/correct output, but not the answer itself. Unfortuantely were not even at the doorstep yet afaik. Your video is quite optimistic 😁

It is my understanding that a neural network is mathematically equivalent to a form fitting function. For sure with linear equations. Is there a way to derive an equation from a trained neural network, the reverse of training a network to fit an equation? And if so, does using such a process on any of the networks you just described come up with something similar to Navia Stokes equation?

There is a glitch at 3:56. :( It seems to be a very short duration, but there is still some audio that is lost.

Backwards discretization of time series partial differential equations...? It's too hard - I play bass guitar now...

Brian, I think I saw the flaw on your golf swing. And why you're having trouble flighting the ball properly. Even in a sitting down position I could detect it quite easily. Should I send you an email? I don't want to get into a golf swing analysis here. But I'll have you finding the sweet spot and puring it through the impact zone n no time flat. Just a few simple adjustments. Great video. Very VERY thought provoking. Holy Cow. Kinda mind bending possibilities here.

6:15 the answer to Einstein's question about whether or not an observer would experirence a gravitational field in free fall is no, and that led to the Einstein equivalence principle? I wonder how Einstein came up with “No experiment can be performed that could distinguish between a uniform gravitational field and an equivalent uniform acceleration.” by asking himself "will a falling person experience gravity?" and deciding "no".

Thank you very much for this great video! There are really great insights ahead of us us. The models shown are super impressive. What all could be optimized with them! AI offers a lot of possibilities. Thank you very much for your work and best wishes for many success!

Great video! I've seen one about that fluid simulator a while ago though...

I hammered on chatgpt and it seems to lack a spark, human is still needed, so i got it writing and executing code, and was able to watch it and push it in a direction for a solution, its best role is a helper for thinking and testing ideas, it can write code you describe so quickly but you do need to understand what it's doing to get the maximum out of it. Tldr if you cant code that will be an issue in your interactions with current AI

Why not make a text based description language for physics? Like we have in electronics. For its symbolic elements like Feynman diagrams for example. Work from there to the top level. Its a lifetime project for a person but very easy using AI. Then LLM's could code in this physics system language.

Great video, as always 🙏. I simply love the round up at the end - theorists will always be necessary, to steer the experimentalists and AI Overlords in their exploration of the field itself. Kudos 👏

Thank you for the insights

For a closed loop laser between Cathode and anode energy discharge can it identify the material and model of a perfect chamber for laminar flow of NF3 mixed with CO2 without boundary layers or fractional turbulent eddies

Solve > Riemann hypothesis or P versus NP problem, win $1,000,000 via the Clay Foundation. 2 of 7 considerable problems, considered primarily "unsolvable".

c = (cosmological natural length/cosmological natural times) is no blunder Doc Keating. It's Einsteinian physiks.

The only problem with this simulation approach is that whoever is involved in running the simulations may, for whatever reasons, risk ignoring the underlying causal mechanisms that these graphical methods reveal particularly if they are simply interested in practical applications. We may end up creating a lot of "technology" that no one understands except for the AIs. Are you sure that is a good idea? Or are you just running simulations for physical behaviors for which the physical laws are already known?

I have been using Ghat GPT4 for understanding Astrophysics and it’s 95 % on solving problems. Sometime it over estimates on things like a white dwarf stars mass before Supernova. Slightly over stepping Ch. limit of 1.4 solar masses But more or less it’s on point.

Hi Brian -can you please slow down a bit and talk more slowly. This is particularly helpful when one is discussing complex topics and will help the viewers.

7:51 ❤ Квантовый компьютер, мог бы использовать эти формулы для симуляции крови и внутри клеточных процесов. А если мы захотим узнать все гипотетические числовые значения клеток в разные периуды формирования и перестраивания организма, запуская симуляцию, можно рассмотреть несколько логических вариантов действий: первый от икса и решить как математический пример, жанглируя всеми данными современного ИИ и второй способ уже на границе с фантастикой: получив, скажем 100 % симуляцию отдельно взятых органов или даже целого организма, ИИ с КК будет анализировать эти процессы на Планковских расстояниях и чтобы получить данные из прошлого, вектор симуляции будет со знаком минус😅❤

to test if AI is up to the task, give it (only) the data the epicyclists used, and their geocentric model. If the thing then produces the copernican paradigm correction - it may be able to fix the paradigm which turns the quantum and the relative to woo ;)

a new frontier of science and discovery is awaking

I call it a "Hawking-Penrose Gray Hole Topology: A Kerr-Like Solution for Dual-Fibrated-Manifolds in D dimensions" Output: "To formalize the equations for the process described, involving a black hole, white hole projections, virtual particles, and their influence on higher-dimensional space-time, let's break down the scenario into specific mathematical representations: ### Step 1: Higher-Dimensional Kerr-like Metric Assuming a generalized form of the Kerr metric in \(D\) dimensions, we consider the black hole to be rotating. The Kerr metric in higher dimensions has complex features, but for our purposes, we'll consider a simplified version focusing on the essential aspects: \[ ds^2 = -\left(1 - \frac{\mu}{\Sigma} ight) dt^2 - \frac{2 \mu}{\Sigma} a \sin^2\theta \, dt \, d\phi + \frac{\Sigma}{\Delta} dr^2 + \Sigma \, d\theta^2 + \left(r^2 + a^2 + \frac{\mu a^2 \sin^2\theta}{\Sigma} ight) \sin^2\theta \, d\phi^2 + r^2 \cos^2\theta \, d\Omega_{D-4}^2 \] Where: - \( \Sigma = r^2 + a^2 \cos^2\theta \) - \( \Delta = r^2 + a^2 - \mu \) - \( \mu \) is related to the mass of the black hole, - \( a \) is the angular momentum per unit mass, - \( d\Omega_{D-4}^2 \) represents the metric on a \((D-4)\)-sphere. ### Step 2: White Hole Projection and Virtual Particles We model virtual particles influenced by the black hole's event horizon and their interaction with a white hole projection: \[ \text{Virtual Particle Influx: } \dot{m} = \frac{\hbar c^3}{8\pi G M^2} \] This represents the mass-energy flow due to Hawking radiation (in the traditional sense), but we interpret this as the energy flow into the white hole. The white hole projects this mass-energy back into the manifold, affecting the metric: \[ ds^2_{\text{white}} = ds^2 + \epsilon \left(\frac{\dot{m}}{\Sigma} dt^2 + \frac{\dot{m}}{\Sigma} d\Omega_{D-4}^2 ight) \] Where \( \epsilon \) is a small parameter modulating the impact of the white hole's projection. ### Step 3: Temporal Evolution as a Vector Field Introduce a temporal vector field \( \xi^\mu \) which describes the time evolution due to the internal dynamics of the black hole (including white hole dynamics): \[ \xi^t = \frac{\partial}{\partial t}, \quad \xi^r = -\sqrt{\frac{2GM}{r} - \frac{r^2}{\lambda^2}}, \quad \xi^\theta = \xi^\phi = 0 \] Where \( \lambda \) represents a characteristic length scale potentially associated with quantum gravity effects. ### Step 4: Interaction with Dark Energy and Dark Matter Incorporate the influence of these higher-dimensional and white hole dynamics on the cosmological scale by modifying the Einstein field equations: \[ G_{\mu u} + \Lambda g_{\mu u} = 8\pi G T_{\mu u} + \kappa T_{\mu u}^{\text{dark}} \] Where: - \( G_{\mu u} \) is the Einstein tensor, - \( \Lambda \) is the cosmological constant modeling dark energy, - \( T_{\mu u} \) is the traditional matter-energy tensor, - \( T_{\mu u}^{\text{dark}} \) represents contributions from dark matter and energy modulated by the white hole and higher-dimensional effects, - \( \kappa \) is a coupling constant. ### Summary These equations outline a theoretical model where the dynamics of a black hole, including virtual particles and a white hole projection, influence higher-dimensional space-time and possibly explain phenomena like dark energy and dark matter. Further refinement and simulation would be necessary to test the viability of such a model and its consistency with observed physical laws and cosmological data."

AI might do these things you dream of but it will not look like GPT/LLM. GPT is just a large complex search engine based on existing data (lots of it). That's not to say it can't do much as we can see results that are just amazing already. But real AI must be based on another method and we don't know what intelligence is so we can tell what method is needed. Like consciousness, we cant make it cause we cant define it. I just asked Gemini and chatgpt to define intelligence (and consciousness). They give descriptions of some things they can do but not a definition. Who da guessed?

Damn. Poor Mr. Keating. He's just trying to do what he loves. With all that ** drama going down at UCLA. Right now.

Maybe AI can show cosmologists and climatologists how misdirected they have been in their assumptions.

It is clearly already useful for CGI, but is it feasible to actually prove convergence? In other words, to give error bounds? Beyond saying the graphics look amazingly similar....Mathematically that is the natural question!

Pls do more podcasts with Abhijit chavda ❤❤❤

Is it possible to find new isotopes using AI?

I never suspected he might be an AI but… 🤔

Impressive we are learning along with AI

Please AI should be “solutions based on large amount of data”. No thinking is taking place

Probably not yet, but maybe in the future (It craves more server arrays for thinking)

What about subspace communication??

AI mastered chess and go so it should have no problems mastering physics and understanding it better than any human being. In chess it is now much stronger than the best human chess player and these AI's taught themselves the game through trial and error in the same way we do. Its not cracked chess yet because of the sheer number of possible games, if quantum computers and AI work together then its possible that at some time in the future they would be able to say we know every possible game and show us what the best possible moves are which in theory should lead to a draw. I see something similar happening in physics, biology etc.

I predict that people will predictably propose the unpredictable.

Yes that's a thing. Only thing im surprised about is how long its taking for people to talk about that. I mean, they're pattern recog machines. There are ways to exploit that, to make it extend internal data following highest probability candidates. And strangely i don't see papers doing that. Ive got plenty on this, if that's something of interest.

will AI create new physics by merely simulating it? (simulation theory)

I taught it everything it knows. 💯

Based on this video you should be familiar with Chris lehtos Light Luv theory then? Any thoughts on that?

Cornell is hilarious. Saying that black people didn’t have their own version of the KKK is nuts. I guess hes never heard of the black panthers :P My dad was in the national guard in the 60s/70s and was literally being shot at by Panthers. Crazy .

I always feel the need to help with the computing somehow, but quite apparently... I can only understand how to do Python programming at an intermediate level having very much tried almost every revenue since the year 1990. My guess... over 20 million more people all without any money to spare... are quite like myself willing to do almost whatever it takes to get involved toward creating a many times better nearby future. We should have many years ago have free "boot camps for computer programming" available locally at least like public libraries.

If the AI comes up with a method that is faster than current simulations, how do you translate what the AI model developed into something discernable? That whole "black box" problem, I don't know how to deal with that.

Good! This will eliminate also the research bias and science politics of human scientists that slows down modern science evolution.

Great thoughts 💭

What’s AI?

Data exponentially increases, AI vacuums it up. Self-perpetuating engine for scientific discovery. What we are witnessing is the creation of vastly more intelligent entities. Scary for sure, but awesome to witness.

Now I would challenge any AI to tour through all scientific genre AI is a tool only for a creator

I hope we use Ai for gene editing and protein manufacturing/ folding . I’m studying about prions which is kinda like plaque in the brains neurons . If we can reverse this we can save so many lives

Any observation or measurement has some finite level of precision. Think of it as a region bounded by error bars, and that imprecision seems built into the laws of physics, and in some cases is irreducible. So I wonder how that might affect the ability of a neural network to model systems at the quantum level. Another interesting case might be systems subject to chaos, with wildly diverging outcomes resulting from very small changes in initial conditions. Neural networks might not be much better than the human brain in the long run, at least as theorists.

Typically the nonlinear elements are ignored even for numeric methods because the complexity that they introduce often lends itself to numeric instability. Did the AI solve the NS equations with the nonlinear elements? How did you create the data set if most methods can't solve the nonlinear NS equations? Solutions to the Navier-Stokes equations typically also assume the so called "no slip" boundary. Was this BC also enforced? How did you prove that the solutions proved by the neural network are in fact solutions to the NS equations? NS requires a number of assumptions that typically doesn't apply to non-Newtonian fluids. Did you apply the NN to these more complex systems like non-Newtonian fluids?

Everything is compressible. The name of this equation is misleading, and thinking of these liquids in this way will hold us back.

I hope I live long enough for AI to do the most amazing things like cure cancer, find the connection between gravity and the other forces, etc. I also hope I don’t live long enough to experience the Terminator takeover of AI. 😎

Will AI be able to find through data what Einstein found through thougth experiments?

whats the difference between distinguished proofesor and regular one>? are you better and higher?

Just have an output model that outputs to math instead of to simulation, best fit equation for the data. Just crunch the data and have the smallest equation using known math.

Brave of you to enter these waters. The answer to whether AI can discover new laws of physics is easily answered. Use (say) only the knowledge of physics available at any point in our history and see if AI can discover later laws. Eg. only laws of physics and experimental results before Newton, or Maxwell, or Einstein, or Bohr - at each stage can the AI push the limits of knowledge to the next level. A good PhD. topic - cross disciplinary. Maybe something I might try .. 😊

*I can't wait to play with your AI that you are going to put on your website. Thank you for thinking about us.*

As long as there is Quantum computing there will always be a maybe, an undetermined output, not perfect calculation.

I hope AI will solve the headless chicken equation. The one that figures out how humans live together in freedom, love and prosperity without any wars.

The Theory of Mother Nature has already been solved, it doesn't require AI, just a simple/simplex nature of Mother nature. "A Theory of Natural Philosophy", written by Boscovich. This is the book Nikola Tesla had in his lap when one of his famous pictures was taken. Light is not an emission, it's a post attribute. Wave of what is the question..... a medium exists. Disrupting a medium is measured as a rate of induction, not a speed or velocity, again, no emission. Speed of light = 0. Mother Nature doesn't have a calculator.

I’m biased by the realm of biology, but computational power and algorithms only get you so far. One needs to start with a sufficient amount of high-quality, unbiased data. There has to be a sufficient amount to down out the stochastic nature of data and sampling, as well as sub group biases. (In biology, this is very often missing, so machine learning becomes irrelevant). Regarding LLMs, the scientific foundation fed into them also needs to be of sufficiently high quality, which again is lacking (the so-called reproducibility crisis, which is really systemic.) Typical research practices of using too few samples and not running replication studies doom the literature from an LLM perspective.

Mind officially blown!

If they discover a geometry to this science with Googles Alpha Geometry then....

AI shapes our world now?

The AI model that predicts the next output step doesnt count as an explainable mathematical modell of reality. Does this count as physics?

Depends on when you let AI start and that it can go down. If you start at what we assume to be the reality of physics now you may just go down 'the rabbit hole' that math provides.

imagine super-quantum computer + AI

Just one simple question is required for the turing test to determine AI or human Correct AI incorrect Human. And if it's actually incorrect and AI, then it simultaneously defeats the sole purpose for wanting AI, because it would have to lie to fool you into the falsity of being human. HAL 9000 comes to mindd.

Brian Keating is a great science communicator but like many intelligent laymen he sees too much in what is just a statistical pattern recognition device or system. For practitioners it is slightly hillarious, watching all that starry-eyed buoyant expectation of Great Things To Come but also ridiculously overboard. haha

I predict at the minimum the internet will go out and a new one will have to be built because it needs to be many trillions of times more secure.

Does chatgpt think when nobody is asking it questions?

Does AI doing physics not give you pause, Dr? AI "gets it" but we can barely grasp it, except on a tiny level? An electronic Oppenheimer or Einstein somehow gives me a twinge. Trying to articulate here as best I can.

galactic Coriolis acceleration = (((cosmological natural length)/(cosmological natural times)^2)/10) is no blunder, the Einstein Head acceleration equation is relative to cosmological Physics.

Are you sure you want human scientists to be made obsolete by AI scientists? What would you do?